add updated_at as an option

### Changes 🏗️

### Before

<img width="800" height="649" alt="Screenshot_2025-10-23_at_00 44 59"
src="https://github.com/user-attachments/assets/fd717d39-772a-4331-bc54-4db15a9a3107"
/>

### After

<img width="800" height="555" alt="Screenshot 2025-10-27 at 23 19 10"
src="https://github.com/user-attachments/assets/64878bd0-3a96-4b3a-8344-1a88c89de52e"
/>

### Checklist 📋

#### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] I have made a test plan
- [x] I have tested my changes according to the test plan:
  - [x] Try to signup with a non-approved email
  - [x] You see the modal with an updated copy

.github/copilot-instructions.md

@@ -12,6 +12,7 @@ This file provides comprehensive onboarding information for GitHub Copilot codin
 - **Infrastructure** - Docker configurations, CI/CD, and development tools
 
 **Primary Languages & Frameworks:**
+
 - **Backend**: Python 3.10-3.13, FastAPI, Prisma ORM, PostgreSQL, RabbitMQ
 - **Frontend**: TypeScript, Next.js 15, React, Tailwind CSS, Radix UI
 - **Development**: Docker, Poetry, pnpm, Playwright, Storybook
@@ -23,15 +24,17 @@ This file provides comprehensive onboarding information for GitHub Copilot codin
 **Always run these commands in the correct directory and in this order:**
 
 1. **Initial Setup** (required once):
+
    ```bash
    # Clone and enter repository
    git clone <repo> && cd AutoGPT
-   
+
    # Start all services (database, redis, rabbitmq, clamav)
    cd autogpt_platform && docker compose --profile local up deps --build --detach
    ```
 
 2. **Backend Setup** (always run before backend development):
+
    ```bash
    cd autogpt_platform/backend
    poetry install                    # Install dependencies
@@ -48,6 +51,7 @@ This file provides comprehensive onboarding information for GitHub Copilot codin
 ### Runtime Requirements
 
 **Critical:** Always ensure Docker services are running before starting development:
+
 ```bash
 cd autogpt_platform && docker compose --profile local up deps --build --detach
 ```
@@ -58,6 +62,7 @@ cd autogpt_platform && docker compose --profile local up deps --build --detach
 ### Development Commands
 
 **Backend Development:**
+
 ```bash
 cd autogpt_platform/backend
 poetry run serve                     # Start development server (port 8000)
@@ -68,6 +73,7 @@ poetry run lint                      # Lint code (ruff) - run after format
 ```
 
 **Frontend Development:**
+
 ```bash
 cd autogpt_platform/frontend
 pnpm dev                            # Start development server (port 3000) - use for active development
@@ -81,23 +87,27 @@ pnpm storybook                      # Start component development server
 ### Testing Strategy
 
 **Backend Tests:**
+
 - **Block Tests**: `poetry run pytest backend/blocks/test/test_block.py -xvs` (validates all blocks)
 - **Specific Block**: `poetry run pytest 'backend/blocks/test/test_block.py::test_available_blocks[BlockName]' -xvs`
 - **Snapshot Tests**: Use `--snapshot-update` when output changes, always review with `git diff`
 
 **Frontend Tests:**
+
 - **E2E Tests**: Always run `pnpm dev` before `pnpm test` (Playwright requires running instance)
 - **Component Tests**: Use Storybook for isolated component development
 
 ### Critical Validation Steps
 
 **Before committing changes:**
+
 1. Run `poetry run format` (backend) and `pnpm format` (frontend)
 2. Ensure all tests pass in modified areas
 3. Verify Docker services are still running
 4. Check that database migrations apply cleanly
 
 **Common Issues & Workarounds:**
+
 - **Prisma issues**: Run `poetry run prisma generate` after schema changes
 - **Permission errors**: Ensure Docker has proper permissions
 - **Port conflicts**: Check the `docker-compose.yml` file for the current list of exposed ports. You can list all mapped ports with:
@@ -108,6 +118,7 @@ pnpm storybook                      # Start component development server
 ### Core Architecture
 
 **AutoGPT Platform** (`autogpt_platform/`):
+
 - `backend/` - FastAPI server with async support
   - `backend/backend/` - Core API logic
   - `backend/blocks/` - Agent execution blocks
@@ -121,6 +132,7 @@ pnpm storybook                      # Start component development server
 - `docker-compose.yml` - Development stack orchestration
 
 **Key Configuration Files:**
+
 - `pyproject.toml` - Python dependencies and tooling
 - `package.json` - Node.js dependencies and scripts
 - `schema.prisma` - Database schema and migrations
@@ -136,6 +148,7 @@ pnpm storybook                      # Start component development server
 ### Development Workflow
 
 **GitHub Actions**: Multiple CI/CD workflows in `.github/workflows/`
+
 - `platform-backend-ci.yml` - Backend testing and validation
 - `platform-frontend-ci.yml` - Frontend testing and validation
 - `platform-fullstack-ci.yml` - End-to-end integration tests
@@ -146,11 +159,13 @@ pnpm storybook                      # Start component development server
 ### Key Source Files
 
 **Backend Entry Points:**
+
 - `backend/backend/server/server.py` - FastAPI application setup
 - `backend/backend/data/` - Database models and user management
 - `backend/blocks/` - Agent execution blocks and logic
 
 **Frontend Entry Points:**
+
 - `frontend/src/app/layout.tsx` - Root application layout
 - `frontend/src/app/page.tsx` - Home page
 - `frontend/src/lib/supabase/` - Authentication and database client
@@ -160,6 +175,7 @@ pnpm storybook                      # Start component development server
 ### Agent Block System
 
 Agents are built using a visual block-based system where each block performs a single action. Blocks are defined in `backend/blocks/` and must include:
+
 - Block definition with input/output schemas
 - Execution logic with proper error handling
 - Tests validating functionality
@@ -167,6 +183,7 @@ Agents are built using a visual block-based system where each block performs a s
 ### Database & ORM
 
 **Prisma ORM** with PostgreSQL backend including pgvector for embeddings:
+
 - Schema in `schema.prisma`
 - Migrations in `backend/migrations/`
 - Always run `prisma migrate dev` and `prisma generate` after schema changes
@@ -174,13 +191,15 @@ Agents are built using a visual block-based system where each block performs a s
 ## Environment Configuration
 
 ### Configuration Files Priority Order
+
 1. **Backend**: `/backend/.env.default` → `/backend/.env` (user overrides)
-2. **Frontend**: `/frontend/.env.default` → `/frontend/.env` (user overrides)  
+2. **Frontend**: `/frontend/.env.default` → `/frontend/.env` (user overrides)
 3. **Platform**: `/.env.default` (Supabase/shared) → `/.env` (user overrides)
 4. Docker Compose `environment:` sections override file-based config
 5. Shell environment variables have highest precedence
 
 ### Docker Environment Setup
+
 - All services use hardcoded defaults (no `${VARIABLE}` substitutions)
 - The `env_file` directive loads variables INTO containers at runtime
 - Backend/Frontend services use YAML anchors for consistent configuration
@@ -189,6 +208,7 @@ Agents are built using a visual block-based system where each block performs a s
 ## Advanced Development Patterns
 
 ### Adding New Blocks
+
 1. Create file in `/backend/backend/blocks/`
 2. Inherit from `Block` base class with input/output schemas
 3. Implement `run` method with proper error handling
@@ -198,6 +218,7 @@ Agents are built using a visual block-based system where each block performs a s
 7. Consider how inputs/outputs connect with other blocks in graph editor
 
 ### API Development
+
 1. Update routes in `/backend/backend/server/routers/`
 2. Add/update Pydantic models in same directory
 3. Write tests alongside route files
@@ -205,21 +226,76 @@ Agents are built using a visual block-based system where each block performs a s
 5. Run `poetry run test` to verify changes
 
 ### Frontend Development
-1. Components in `/frontend/src/components/`
-2. Use existing UI components from `/frontend/src/components/ui/`
-3. Add Storybook stories for component development
-4. Test user-facing features with Playwright E2E tests
-5. Update protected routes in middleware when needed
+
+**📖 Complete Frontend Guide**: See `autogpt_platform/frontend/CONTRIBUTING.md` and `autogpt_platform/frontend/.cursorrules` for comprehensive patterns and conventions.
+
+**Quick Reference:**
+
+**Component Structure:**
+
+- Separate render logic from data/behavior
+- Structure: `ComponentName/ComponentName.tsx` + `useComponentName.ts` + `helpers.ts`
+- Exception: Small components (3-4 lines of logic) can be inline
+- Render-only components can be direct files without folders
+
+**Data Fetching:**
+
+- Use generated API hooks from `@/app/api/__generated__/endpoints/`
+- Generated via Orval from backend OpenAPI spec
+- Pattern: `use{Method}{Version}{OperationName}`
+- Example: `useGetV2ListLibraryAgents`
+- Regenerate with: `pnpm generate:api`
+- **Never** use deprecated `BackendAPI` or `src/lib/autogpt-server-api/*`
+
+**Code Conventions:**
+
+- Use function declarations for components and handlers (not arrow functions)
+- Only arrow functions for small inline lambdas (map, filter, etc.)
+- Components: `PascalCase`, Hooks: `camelCase` with `use` prefix
+- No barrel files or `index.ts` re-exports
+- Minimal comments (code should be self-documenting)
+
+**Styling:**
+
+- Use Tailwind CSS utilities only
+- Use design system components from `src/components/` (atoms, molecules, organisms)
+- Never use `src/components/__legacy__/*`
+- Only use Phosphor Icons (`@phosphor-icons/react`)
+- Prefer design tokens over hardcoded values
+
+**Error Handling:**
+
+- Render errors: Use `<ErrorCard />` component
+- Mutation errors: Display with toast notifications
+- Manual exceptions: Use `Sentry.captureException()`
+- Global error boundaries already configured
+
+**Testing:**
+
+- Add/update Storybook stories for UI components (`pnpm storybook`)
+- Run Playwright E2E tests with `pnpm test`
+- Verify in Chromatic after PR
+
+**Architecture:**
+
+- Default to client components ("use client")
+- Server components only for SEO or extreme TTFB needs
+- Use React Query for server state (via generated hooks)
+- Co-locate UI state in components/hooks
 
 ### Security Guidelines
+
 **Cache Protection Middleware** (`/backend/backend/server/middleware/security.py`):
+
 - Default: Disables caching for ALL endpoints with `Cache-Control: no-store, no-cache, must-revalidate, private`
 - Uses allow list approach for cacheable paths (static assets, health checks, public pages)
 - Prevents sensitive data caching in browsers/proxies
 - Add new cacheable endpoints to `CACHEABLE_PATHS`
 
 ### CI/CD Alignment
+
 The repository has comprehensive CI workflows that test:
+
 - **Backend**: Python 3.11-3.13, services (Redis/RabbitMQ/ClamAV), Prisma migrations, Poetry lock validation
 - **Frontend**: Node.js 21, pnpm, Playwright with Docker Compose stack, API schema validation
 - **Integration**: Full-stack type checking and E2E testing
@@ -229,6 +305,7 @@ Match these patterns when developing locally - the copilot setup environment mir
 ## Collaboration with Other AI Assistants
 
 This repository is actively developed with assistance from Claude (via CLAUDE.md files). When working on this codebase:
+
 - Check for existing CLAUDE.md files that provide additional context
 - Follow established patterns and conventions already in the codebase
 - Maintain consistency with existing code style and architecture
@@ -237,8 +314,9 @@ This repository is actively developed with assistance from Claude (via CLAUDE.md
 ## Trust These Instructions
 
 These instructions are comprehensive and tested. Only perform additional searches if:
+
 1. Information here is incomplete for your specific task
 2. You encounter errors not covered by the workarounds
 3. You need to understand implementation details not covered above
 
-For detailed platform development patterns, refer to `autogpt_platform/CLAUDE.md` and `AGENTS.md` in the repository root.
+For detailed platform development patterns, refer to `autogpt_platform/CLAUDE.md` and `AGENTS.md` in the repository root.

autogpt_platform/CLAUDE.md

@@ -63,6 +63,9 @@ poetry run pytest path/to/test.py --snapshot-update
 # Install dependencies
 cd frontend && pnpm i
 
+# Generate API client from OpenAPI spec
+pnpm generate:api
+
 # Start development server
 pnpm dev
 
@@ -75,12 +78,23 @@ pnpm storybook
 # Build production
 pnpm build
 
+# Format and lint
+pnpm format
+
 # Type checking
 pnpm types
 ```
 
-We have a components library in autogpt_platform/frontend/src/components/atoms that should be used when adding new pages and components. 
+**📖 Complete Guide**: See `/frontend/CONTRIBUTING.md` and `/frontend/.cursorrules` for comprehensive frontend patterns.
 
+**Key Frontend Conventions:**
+
+- Separate render logic from data/behavior in components
+- Use generated API hooks from `@/app/api/__generated__/endpoints/`
+- Use function declarations (not arrow functions) for components/handlers
+- Use design system components from `src/components/` (atoms, molecules, organisms)
+- Only use Phosphor Icons
+- Never use `src/components/__legacy__/*` or deprecated `BackendAPI`
 
 ## Architecture Overview
 
@@ -95,11 +109,16 @@ We have a components library in autogpt_platform/frontend/src/components/atoms t
 
 ### Frontend Architecture
 
-- **Framework**: Next.js App Router with React Server Components
-- **State Management**: React hooks + Supabase client for real-time updates
+- **Framework**: Next.js 15 App Router (client-first approach)
+- **Data Fetching**: Type-safe generated API hooks via Orval + React Query
+- **State Management**: React Query for server state, co-located UI state in components/hooks
+- **Component Structure**: Separate render logic (`.tsx`) from business logic (`use*.ts` hooks)
 - **Workflow Builder**: Visual graph editor using @xyflow/react
-- **UI Components**: Radix UI primitives with Tailwind CSS styling
+- **UI Components**: shadcn/ui (Radix UI primitives) with Tailwind CSS styling
+- **Icons**: Phosphor Icons only
 - **Feature Flags**: LaunchDarkly integration
+- **Error Handling**: ErrorCard for render errors, toast for mutations, Sentry for exceptions
+- **Testing**: Playwright for E2E, Storybook for component development
 
 ### Key Concepts
 
@@ -153,6 +172,7 @@ Key models (defined in `/backend/schema.prisma`):
 **Adding a new block:**
 
 Follow the comprehensive [Block SDK Guide](../../../docs/content/platform/block-sdk-guide.md) which covers:
+
 - Provider configuration with `ProviderBuilder`
 - Block schema definition
 - Authentication (API keys, OAuth, webhooks)
@@ -160,6 +180,7 @@ Follow the comprehensive [Block SDK Guide](../../../docs/content/platform/block-
 - File organization
 
 Quick steps:
+
 1. Create new file in `/backend/backend/blocks/`
 2. Configure provider using `ProviderBuilder` in `_config.py`
 3. Inherit from `Block` base class
@@ -180,10 +201,20 @@ ex: do the inputs and outputs tie well together?
 
 **Frontend feature development:**
 
-1. Components go in `/frontend/src/components/`
-2. Use existing UI components from `/frontend/src/components/ui/`
-3. Add Storybook stories for new components
-4. Test with Playwright if user-facing
+See `/frontend/CONTRIBUTING.md` for complete patterns. Quick reference:
+
+1. **Pages**: Create in `src/app/(platform)/feature-name/page.tsx`
+   - Add `usePageName.ts` hook for logic
+   - Put sub-components in local `components/` folder
+2. **Components**: Structure as `ComponentName/ComponentName.tsx` + `useComponentName.ts` + `helpers.ts`
+   - Use design system components from `src/components/` (atoms, molecules, organisms)
+   - Never use `src/components/__legacy__/*`
+3. **Data fetching**: Use generated API hooks from `@/app/api/__generated__/endpoints/`
+   - Regenerate with `pnpm generate:api`
+   - Pattern: `use{Method}{Version}{OperationName}`
+4. **Styling**: Tailwind CSS only, use design tokens, Phosphor Icons only
+5. **Testing**: Add Storybook stories for new components, Playwright for E2E
+6. **Code conventions**: Function declarations (not arrow functions) for components/handlers
 
 ### Security Implementation
 

autogpt_platform/Makefile

@@ -0,0 +1,57 @@
+.PHONY: start-core stop-core logs-core format lint migrate run-backend run-frontend
+
+# Run just Supabase + Redis + RabbitMQ
+start-core:
+	docker compose up -d deps
+
+# Stop core services
+stop-core:
+	docker compose stop deps
+
+reset-db:
+	rm -rf db/docker/volumes/db/data
+	cd backend && poetry run prisma migrate deploy
+	cd backend && poetry run prisma generate
+	
+# View logs for core services
+logs-core:
+	docker compose logs -f deps
+
+# Run formatting and linting for backend and frontend
+format:
+	cd backend && poetry run format
+	cd frontend && pnpm format
+	cd frontend && pnpm lint
+
+init-env:
+	cp -n .env.default .env || true
+	cd backend && cp -n .env.default .env || true
+	cd frontend && cp -n .env.default .env || true
+
+
+# Run migrations for backend
+migrate:
+	cd backend && poetry run prisma migrate deploy
+	cd backend && poetry run prisma generate
+
+run-backend:
+	cd backend && poetry run app
+
+run-frontend:
+	cd frontend && pnpm dev
+
+test-data:
+	cd backend && poetry run python test/test_data_creator.py
+	
+help:
+	@echo "Usage: make <target>"
+	@echo "Targets:"
+	@echo "  start-core - Start just the core services (Supabase, Redis, RabbitMQ) in background"
+	@echo "  stop-core - Stop the core services"
+	@echo "  reset-db - Reset the database by deleting the volume"
+	@echo "  logs-core - Tail the logs for core services"
+	@echo "  format - Format & lint backend (Python) and frontend (TypeScript) code"
+	@echo "  migrate - Run backend database migrations"
+	@echo "  run-backend - Run the backend FastAPI server"
+	@echo "  run-frontend - Run the frontend Next.js development server"
+	@echo "  test-data - Run the test data creator"

autogpt_platform/README.md

@@ -38,6 +38,37 @@ To run the AutoGPT Platform, follow these steps:
 
 4. After all the services are in ready state, open your browser and navigate to `http://localhost:3000` to access the AutoGPT Platform frontend.
 
+### Running Just Core services
+
+You can now run the following to enable just the core services.
+
+```
+# For help
+make help
+
+# Run just Supabase + Redis + RabbitMQ
+make start-core
+
+# Stop core services
+make stop-core
+
+# View logs from core services 
+make logs-core
+
+# Run formatting and linting for backend and frontend
+make format
+
+# Run migrations for backend database
+make migrate
+
+# Run backend server
+make run-backend
+
+# Run frontend development server
+make run-frontend
+
+```
+
 ### Docker Compose Commands
 
 Here are some useful Docker Compose commands for managing your AutoGPT Platform:

autogpt_platform/autogpt_libs/autogpt_libs/logging/config.py

@@ -94,42 +94,36 @@ def configure_logging(force_cloud_logging: bool = False) -> None:
     config = LoggingConfig()
     log_handlers: list[logging.Handler] = []
 
+    structured_logging = config.enable_cloud_logging or force_cloud_logging
+
     # Console output handlers
-    stdout = logging.StreamHandler(stream=sys.stdout)
-    stdout.setLevel(config.level)
-    stdout.addFilter(BelowLevelFilter(logging.WARNING))
-    if config.level == logging.DEBUG:
-        stdout.setFormatter(AGPTFormatter(DEBUG_LOG_FORMAT))
-    else:
-        stdout.setFormatter(AGPTFormatter(SIMPLE_LOG_FORMAT))
+    if not structured_logging:
+        stdout = logging.StreamHandler(stream=sys.stdout)
+        stdout.setLevel(config.level)
+        stdout.addFilter(BelowLevelFilter(logging.WARNING))
+        if config.level == logging.DEBUG:
+            stdout.setFormatter(AGPTFormatter(DEBUG_LOG_FORMAT))
+        else:
+            stdout.setFormatter(AGPTFormatter(SIMPLE_LOG_FORMAT))
 
-    stderr = logging.StreamHandler()
-    stderr.setLevel(logging.WARNING)
-    if config.level == logging.DEBUG:
-        stderr.setFormatter(AGPTFormatter(DEBUG_LOG_FORMAT))
-    else:
-        stderr.setFormatter(AGPTFormatter(SIMPLE_LOG_FORMAT))
+        stderr = logging.StreamHandler()
+        stderr.setLevel(logging.WARNING)
+        if config.level == logging.DEBUG:
+            stderr.setFormatter(AGPTFormatter(DEBUG_LOG_FORMAT))
+        else:
+            stderr.setFormatter(AGPTFormatter(SIMPLE_LOG_FORMAT))
 
-    log_handlers += [stdout, stderr]
+        log_handlers += [stdout, stderr]
 
     # Cloud logging setup
-    if config.enable_cloud_logging or force_cloud_logging:
-        import google.cloud.logging
-        from google.cloud.logging.handlers import CloudLoggingHandler
-        from google.cloud.logging_v2.handlers.transports import (
-            BackgroundThreadTransport,
-        )
+    else:
+        # Use Google Cloud Structured Log Handler. Log entries are printed to stdout
+        # in a JSON format which is automatically picked up by Google Cloud Logging.
+        from google.cloud.logging.handlers import StructuredLogHandler
 
-        client = google.cloud.logging.Client()
-        # Use BackgroundThreadTransport to prevent blocking the main thread
-        # and deadlocks when gRPC calls to Google Cloud Logging hang
-        cloud_handler = CloudLoggingHandler(
-            client,
-            name="autogpt_logs",
-            transport=BackgroundThreadTransport,
-        )
-        cloud_handler.setLevel(config.level)
-        log_handlers.append(cloud_handler)
+        structured_log_handler = StructuredLogHandler(stream=sys.stdout)
+        structured_log_handler.setLevel(config.level)
+        log_handlers.append(structured_log_handler)
 
     # File logging setup
     if config.enable_file_logging:
@@ -185,7 +179,13 @@ def configure_logging(force_cloud_logging: bool = False) -> None:
 
     # Configure the root logger
     logging.basicConfig(
-        format=DEBUG_LOG_FORMAT if config.level == logging.DEBUG else SIMPLE_LOG_FORMAT,
+        format=(
+            "%(levelname)s  %(message)s"
+            if structured_logging
+            else (
+                DEBUG_LOG_FORMAT if config.level == logging.DEBUG else SIMPLE_LOG_FORMAT
+            )
+        ),
         level=config.level,
         handlers=log_handlers,
     )

autogpt_platform/autogpt_libs/autogpt_libs/utils/cache.py

@@ -1,339 +0,0 @@
-import asyncio
-import inspect
-import logging
-import threading
-import time
-from functools import wraps
-from typing import (
-    Any,
-    Callable,
-    ParamSpec,
-    Protocol,
-    TypeVar,
-    cast,
-    runtime_checkable,
-)
-
-P = ParamSpec("P")
-R = TypeVar("R")
-R_co = TypeVar("R_co", covariant=True)
-
-logger = logging.getLogger(__name__)
-
-
-def _make_hashable_key(
-    args: tuple[Any, ...], kwargs: dict[str, Any]
-) -> tuple[Any, ...]:
-    """
-    Convert args and kwargs into a hashable cache key.
-
-    Handles unhashable types like dict, list, set by converting them to
-    their sorted string representations.
-    """
-
-    def make_hashable(obj: Any) -> Any:
-        """Recursively convert an object to a hashable representation."""
-        if isinstance(obj, dict):
-            # Sort dict items to ensure consistent ordering
-            return (
-                "__dict__",
-                tuple(sorted((k, make_hashable(v)) for k, v in obj.items())),
-            )
-        elif isinstance(obj, (list, tuple)):
-            return ("__list__", tuple(make_hashable(item) for item in obj))
-        elif isinstance(obj, set):
-            return ("__set__", tuple(sorted(make_hashable(item) for item in obj)))
-        elif hasattr(obj, "__dict__"):
-            # Handle objects with __dict__ attribute
-            return ("__obj__", obj.__class__.__name__, make_hashable(obj.__dict__))
-        else:
-            # For basic hashable types (str, int, bool, None, etc.)
-            try:
-                hash(obj)
-                return obj
-            except TypeError:
-                # Fallback: convert to string representation
-                return ("__str__", str(obj))
-
-    hashable_args = tuple(make_hashable(arg) for arg in args)
-    hashable_kwargs = tuple(sorted((k, make_hashable(v)) for k, v in kwargs.items()))
-    return (hashable_args, hashable_kwargs)
-
-
-@runtime_checkable
-class CachedFunction(Protocol[P, R_co]):
-    """Protocol for cached functions with cache management methods."""
-
-    def cache_clear(self) -> None:
-        """Clear all cached entries."""
-        return None
-
-    def cache_info(self) -> dict[str, int | None]:
-        """Get cache statistics."""
-        return {}
-
-    def cache_delete(self, *args: P.args, **kwargs: P.kwargs) -> bool:
-        """Delete a specific cache entry by its arguments. Returns True if entry existed."""
-        return False
-
-    def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R_co:
-        """Call the cached function."""
-        return None  # type: ignore
-
-
-def cached(
-    *,
-    maxsize: int = 128,
-    ttl_seconds: int | None = None,
-) -> Callable[[Callable], CachedFunction]:
-    """
-    Thundering herd safe cache decorator for both sync and async functions.
-
-    Uses double-checked locking to prevent multiple threads/coroutines from
-    executing the expensive operation simultaneously during cache misses.
-
-    Args:
-        func: The function to cache (when used without parentheses)
-        maxsize: Maximum number of cached entries
-        ttl_seconds: Time to live in seconds. If None, entries never expire
-
-    Returns:
-        Decorated function or decorator
-
-    Example:
-        @cache()  # Default: maxsize=128, no TTL
-        def expensive_sync_operation(param: str) -> dict:
-            return {"result": param}
-
-        @cache()  # Works with async too
-        async def expensive_async_operation(param: str) -> dict:
-            return {"result": param}
-
-        @cache(maxsize=1000, ttl_seconds=300)  # Custom maxsize and TTL
-        def another_operation(param: str) -> dict:
-            return {"result": param}
-    """
-
-    def decorator(target_func):
-        # Cache storage and per-event-loop locks
-        cache_storage = {}
-        _event_loop_locks = {}  # Maps event loop to its asyncio.Lock
-
-        if inspect.iscoroutinefunction(target_func):
-
-            def _get_cache_lock():
-                """Get or create an asyncio.Lock for the current event loop."""
-                try:
-                    loop = asyncio.get_running_loop()
-                except RuntimeError:
-                    # No event loop, use None as default key
-                    loop = None
-
-                if loop not in _event_loop_locks:
-                    return _event_loop_locks.setdefault(loop, asyncio.Lock())
-                return _event_loop_locks[loop]
-
-            @wraps(target_func)
-            async def async_wrapper(*args: P.args, **kwargs: P.kwargs):
-                key = _make_hashable_key(args, kwargs)
-                current_time = time.time()
-
-                # Fast path: check cache without lock
-                if key in cache_storage:
-                    if ttl_seconds is None:
-                        logger.debug(f"Cache hit for {target_func.__name__}")
-                        return cache_storage[key]
-                    else:
-                        cached_data = cache_storage[key]
-                        if isinstance(cached_data, tuple):
-                            result, timestamp = cached_data
-                            if current_time - timestamp < ttl_seconds:
-                                logger.debug(f"Cache hit for {target_func.__name__}")
-                                return result
-
-                # Slow path: acquire lock for cache miss/expiry
-                async with _get_cache_lock():
-                    # Double-check: another coroutine might have populated cache
-                    if key in cache_storage:
-                        if ttl_seconds is None:
-                            return cache_storage[key]
-                        else:
-                            cached_data = cache_storage[key]
-                            if isinstance(cached_data, tuple):
-                                result, timestamp = cached_data
-                                if current_time - timestamp < ttl_seconds:
-                                    return result
-
-                    # Cache miss - execute function
-                    logger.debug(f"Cache miss for {target_func.__name__}")
-                    result = await target_func(*args, **kwargs)
-
-                    # Store result
-                    if ttl_seconds is None:
-                        cache_storage[key] = result
-                    else:
-                        cache_storage[key] = (result, current_time)
-
-                    # Cleanup if needed
-                    if len(cache_storage) > maxsize:
-                        cutoff = maxsize // 2
-                        oldest_keys = (
-                            list(cache_storage.keys())[:-cutoff] if cutoff > 0 else []
-                        )
-                        for old_key in oldest_keys:
-                            cache_storage.pop(old_key, None)
-
-                    return result
-
-            wrapper = async_wrapper
-
-        else:
-            # Sync function with threading.Lock
-            cache_lock = threading.Lock()
-
-            @wraps(target_func)
-            def sync_wrapper(*args: P.args, **kwargs: P.kwargs):
-                key = _make_hashable_key(args, kwargs)
-                current_time = time.time()
-
-                # Fast path: check cache without lock
-                if key in cache_storage:
-                    if ttl_seconds is None:
-                        logger.debug(f"Cache hit for {target_func.__name__}")
-                        return cache_storage[key]
-                    else:
-                        cached_data = cache_storage[key]
-                        if isinstance(cached_data, tuple):
-                            result, timestamp = cached_data
-                            if current_time - timestamp < ttl_seconds:
-                                logger.debug(f"Cache hit for {target_func.__name__}")
-                                return result
-
-                # Slow path: acquire lock for cache miss/expiry
-                with cache_lock:
-                    # Double-check: another thread might have populated cache
-                    if key in cache_storage:
-                        if ttl_seconds is None:
-                            return cache_storage[key]
-                        else:
-                            cached_data = cache_storage[key]
-                            if isinstance(cached_data, tuple):
-                                result, timestamp = cached_data
-                                if current_time - timestamp < ttl_seconds:
-                                    return result
-
-                    # Cache miss - execute function
-                    logger.debug(f"Cache miss for {target_func.__name__}")
-                    result = target_func(*args, **kwargs)
-
-                    # Store result
-                    if ttl_seconds is None:
-                        cache_storage[key] = result
-                    else:
-                        cache_storage[key] = (result, current_time)
-
-                    # Cleanup if needed
-                    if len(cache_storage) > maxsize:
-                        cutoff = maxsize // 2
-                        oldest_keys = (
-                            list(cache_storage.keys())[:-cutoff] if cutoff > 0 else []
-                        )
-                        for old_key in oldest_keys:
-                            cache_storage.pop(old_key, None)
-
-                    return result
-
-            wrapper = sync_wrapper
-
-        # Add cache management methods
-        def cache_clear() -> None:
-            cache_storage.clear()
-
-        def cache_info() -> dict[str, int | None]:
-            return {
-                "size": len(cache_storage),
-                "maxsize": maxsize,
-                "ttl_seconds": ttl_seconds,
-            }
-
-        def cache_delete(*args, **kwargs) -> bool:
-            """Delete a specific cache entry. Returns True if entry existed."""
-            key = _make_hashable_key(args, kwargs)
-            if key in cache_storage:
-                del cache_storage[key]
-                return True
-            return False
-
-        setattr(wrapper, "cache_clear", cache_clear)
-        setattr(wrapper, "cache_info", cache_info)
-        setattr(wrapper, "cache_delete", cache_delete)
-
-        return cast(CachedFunction, wrapper)
-
-    return decorator
-
-
-def thread_cached(func):
-    """
-    Thread-local cache decorator for both sync and async functions.
-
-    Each thread gets its own cache, which is useful for request-scoped caching
-    in web applications where you want to cache within a single request but
-    not across requests.
-
-    Args:
-        func: The function to cache
-
-    Returns:
-        Decorated function with thread-local caching
-
-    Example:
-        @thread_cached
-        def expensive_operation(param: str) -> dict:
-            return {"result": param}
-
-        @thread_cached  # Works with async too
-        async def expensive_async_operation(param: str) -> dict:
-            return {"result": param}
-    """
-    thread_local = threading.local()
-
-    def _clear():
-        if hasattr(thread_local, "cache"):
-            del thread_local.cache
-
-    if inspect.iscoroutinefunction(func):
-
-        @wraps(func)
-        async def async_wrapper(*args, **kwargs):
-            cache = getattr(thread_local, "cache", None)
-            if cache is None:
-                cache = thread_local.cache = {}
-            key = _make_hashable_key(args, kwargs)
-            if key not in cache:
-                cache[key] = await func(*args, **kwargs)
-            return cache[key]
-
-        setattr(async_wrapper, "clear_cache", _clear)
-        return async_wrapper
-
-    else:
-
-        @wraps(func)
-        def sync_wrapper(*args, **kwargs):
-            cache = getattr(thread_local, "cache", None)
-            if cache is None:
-                cache = thread_local.cache = {}
-            key = _make_hashable_key(args, kwargs)
-            if key not in cache:
-                cache[key] = func(*args, **kwargs)
-            return cache[key]
-
-        setattr(sync_wrapper, "clear_cache", _clear)
-        return sync_wrapper
-
-
-def clear_thread_cache(func: Callable) -> None:
-    """Clear thread-local cache for a function."""
-    if clear := getattr(func, "clear_cache", None):
-        clear()

autogpt_platform/backend/Dockerfile

@@ -47,6 +47,7 @@ RUN poetry install --no-ansi --no-root
 
 # Generate Prisma client
 COPY autogpt_platform/backend/schema.prisma ./
+COPY autogpt_platform/backend/backend/data/partial_types.py ./backend/data/partial_types.py
 RUN poetry run prisma generate
 
 FROM debian:13-slim AS server_dependencies
@@ -92,6 +93,7 @@ FROM server_dependencies AS migrate
 
 # Migration stage only needs schema and migrations - much lighter than full backend
 COPY autogpt_platform/backend/schema.prisma /app/autogpt_platform/backend/
+COPY autogpt_platform/backend/backend/data/partial_types.py /app/autogpt_platform/backend/backend/data/partial_types.py
 COPY autogpt_platform/backend/migrations /app/autogpt_platform/backend/migrations
 
 FROM server_dependencies AS server

autogpt_platform/backend/backend/blocks/__init__.py

@@ -5,7 +5,7 @@ import re
 from pathlib import Path
 from typing import TYPE_CHECKING, TypeVar
 
-from autogpt_libs.utils.cache import cached
+from backend.util.cache import cached
 
 logger = logging.getLogger(__name__)
 
@@ -16,7 +16,7 @@ if TYPE_CHECKING:
 T = TypeVar("T")
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def load_all_blocks() -> dict[str, type["Block"]]:
     from backend.data.block import Block
     from backend.util.settings import Config

autogpt_platform/backend/backend/blocks/data_manipulation.py

@@ -66,6 +66,7 @@ class AddToDictionaryBlock(Block):
         dictionary: dict[Any, Any] = SchemaField(
             default_factory=dict,
             description="The dictionary to add the entry to. If not provided, a new dictionary will be created.",
+            advanced=False,
         )
         key: str = SchemaField(
             default="",

autogpt_platform/backend/backend/blocks/discord/bot_blocks.py

@@ -4,13 +4,13 @@ import mimetypes
 from pathlib import Path
 from typing import Any
 
-import aiohttp
 import discord
 from pydantic import SecretStr
 
 from backend.data.block import Block, BlockCategory, BlockOutput, BlockSchema
 from backend.data.model import APIKeyCredentials, SchemaField
 from backend.util.file import store_media_file
+from backend.util.request import Requests
 from backend.util.type import MediaFileType
 
 from ._auth import (
@@ -114,10 +114,9 @@ class ReadDiscordMessagesBlock(Block):
             if message.attachments:
                 attachment = message.attachments[0]  # Process the first attachment
                 if attachment.filename.endswith((".txt", ".py")):
-                    async with aiohttp.ClientSession() as session:
-                        async with session.get(attachment.url) as response:
-                            file_content = response.text()
-                            self.output_data += f"\n\nFile from user: {attachment.filename}\nContent: {file_content}"
+                    response = await Requests().get(attachment.url)
+                    file_content = response.text()
+                    self.output_data += f"\n\nFile from user: {attachment.filename}\nContent: {file_content}"
 
             await client.close()
 
@@ -171,11 +170,11 @@ class SendDiscordMessageBlock(Block):
             description="The content of the message to send"
         )
         channel_name: str = SchemaField(
-            description="The name of the channel the message will be sent to"
+            description="Channel ID or channel name to send the message to"
         )
         server_name: str = SchemaField(
-            description="The name of the server where the channel is located",
-            advanced=True,  # Optional field for server name
+            description="Server name (only needed if using channel name)",
+            advanced=True,
             default="",
         )
 
@@ -231,25 +230,49 @@ class SendDiscordMessageBlock(Block):
         @client.event
         async def on_ready():
             print(f"Logged in as {client.user}")
-            for guild in client.guilds:
-                if server_name and guild.name != server_name:
-                    continue
-                for channel in guild.text_channels:
-                    if channel.name == channel_name:
-                        # Split message into chunks if it exceeds 2000 characters
-                        chunks = self.chunk_message(message_content)
-                        last_message = None
-                        for chunk in chunks:
-                            last_message = await channel.send(chunk)
-                        result["status"] = "Message sent"
-                        result["message_id"] = (
-                            str(last_message.id) if last_message else ""
-                        )
-                        result["channel_id"] = str(channel.id)
-                        await client.close()
-                        return
+            channel = None
 
-            result["status"] = "Channel not found"
+            # Try to parse as channel ID first
+            try:
+                channel_id = int(channel_name)
+                channel = client.get_channel(channel_id)
+            except ValueError:
+                # Not a valid ID, will try name lookup
+                pass
+
+            # If not found by ID (or not an ID), try name lookup
+            if not channel:
+                for guild in client.guilds:
+                    if server_name and guild.name != server_name:
+                        continue
+                    for ch in guild.text_channels:
+                        if ch.name == channel_name:
+                            channel = ch
+                            break
+                    if channel:
+                        break
+
+            if not channel:
+                result["status"] = f"Channel not found: {channel_name}"
+                await client.close()
+                return
+
+            # Type check - ensure it's a text channel that can send messages
+            if not hasattr(channel, "send"):
+                result["status"] = (
+                    f"Channel {channel_name} cannot receive messages (not a text channel)"
+                )
+                await client.close()
+                return
+
+            # Split message into chunks if it exceeds 2000 characters
+            chunks = self.chunk_message(message_content)
+            last_message = None
+            for chunk in chunks:
+                last_message = await channel.send(chunk)  # type: ignore
+            result["status"] = "Message sent"
+            result["message_id"] = str(last_message.id) if last_message else ""
+            result["channel_id"] = str(channel.id)
             await client.close()
 
         await client.start(token)
@@ -675,16 +698,15 @@ class SendDiscordFileBlock(Block):
 
                 elif file.startswith(("http://", "https://")):
                     # URL - download the file
-                    async with aiohttp.ClientSession() as session:
-                        async with session.get(file) as response:
-                            file_bytes = await response.read()
+                    response = await Requests().get(file)
+                    file_bytes = response.content
 
-                            # Try to get filename from URL if not provided
-                            if not filename:
-                                from urllib.parse import urlparse
+                    # Try to get filename from URL if not provided
+                    if not filename:
+                        from urllib.parse import urlparse
 
-                                path = urlparse(file).path
-                                detected_filename = Path(path).name or "download"
+                        path = urlparse(file).path
+                        detected_filename = Path(path).name or "download"
                 else:
                     # Local file path - read from stored media file
                     # This would be a path from a previous block's output

autogpt_platform/backend/backend/blocks/iteration.py

@@ -2,7 +2,7 @@ from typing import Any
 
 from backend.data.block import Block, BlockCategory, BlockOutput, BlockSchema
 from backend.data.model import SchemaField
-from backend.util.json import json
+from backend.util.json import loads
 
 
 class StepThroughItemsBlock(Block):
@@ -68,7 +68,7 @@ class StepThroughItemsBlock(Block):
                     raise ValueError(
                         f"Input too large: {len(data)} bytes > {MAX_ITEM_SIZE} bytes"
                     )
-                items = json.loads(data)
+                items = loads(data)
             else:
                 items = data
 

autogpt_platform/backend/backend/blocks/jina/fact_checker.py

@@ -1,5 +1,8 @@
+from typing import List
 from urllib.parse import quote
 
+from typing_extensions import TypedDict
+
 from backend.blocks.jina._auth import (
     JinaCredentials,
     JinaCredentialsField,
@@ -10,6 +13,12 @@ from backend.data.model import SchemaField
 from backend.util.request import Requests
 
 
+class Reference(TypedDict):
+    url: str
+    keyQuote: str
+    isSupportive: bool
+
+
 class FactCheckerBlock(Block):
     class Input(BlockSchema):
         statement: str = SchemaField(
@@ -23,6 +32,10 @@ class FactCheckerBlock(Block):
         )
         result: bool = SchemaField(description="The result of the factuality check")
         reason: str = SchemaField(description="The reason for the factuality result")
+        references: List[Reference] = SchemaField(
+            description="List of references supporting or contradicting the statement",
+            default=[],
+        )
         error: str = SchemaField(description="Error message if the check fails")
 
     def __init__(self):
@@ -53,5 +66,11 @@ class FactCheckerBlock(Block):
             yield "factuality", data["factuality"]
             yield "result", data["result"]
             yield "reason", data["reason"]
+
+            # Yield references if present in the response
+            if "references" in data:
+                yield "references", data["references"]
+            else:
+                yield "references", []
         else:
             raise RuntimeError(f"Expected 'data' key not found in response: {data}")

autogpt_platform/backend/backend/blocks/linear/_config.py

@@ -62,10 +62,10 @@ TEST_CREDENTIALS_OAUTH = OAuth2Credentials(
     title="Mock Linear API key",
     username="mock-linear-username",
     access_token=SecretStr("mock-linear-access-token"),
-    access_token_expires_at=None,
+    access_token_expires_at=1672531200,  # Mock expiration time for short-lived token
     refresh_token=SecretStr("mock-linear-refresh-token"),
     refresh_token_expires_at=None,
-    scopes=["mock-linear-scopes"],
+    scopes=["read", "write"],
 )
 
 TEST_CREDENTIALS_API_KEY = APIKeyCredentials(

autogpt_platform/backend/backend/blocks/linear/_oauth.py

@@ -2,7 +2,9 @@
 Linear OAuth handler implementation.
 """
 
+import base64
 import json
+import time
 from typing import Optional
 from urllib.parse import urlencode
 
@@ -38,8 +40,9 @@ class LinearOAuthHandler(BaseOAuthHandler):
         self.client_secret = client_secret
         self.redirect_uri = redirect_uri
         self.auth_base_url = "https://linear.app/oauth/authorize"
-        self.token_url = "https://api.linear.app/oauth/token"  # Correct token URL
+        self.token_url = "https://api.linear.app/oauth/token"
         self.revoke_url = "https://api.linear.app/oauth/revoke"
+        self.migrate_url = "https://api.linear.app/oauth/migrate_old_token"
 
     def get_login_url(
         self, scopes: list[str], state: str, code_challenge: Optional[str]
@@ -82,19 +85,84 @@ class LinearOAuthHandler(BaseOAuthHandler):
 
         return True  # Linear doesn't return JSON on successful revoke
 
+    async def migrate_old_token(
+        self, credentials: OAuth2Credentials
+    ) -> OAuth2Credentials:
+        """
+        Migrate an old long-lived token to a new short-lived token with refresh token.
+
+        This uses Linear's /oauth/migrate_old_token endpoint to exchange current
+        long-lived tokens for short-lived tokens with refresh tokens without
+        requiring users to re-authorize.
+        """
+        if not credentials.access_token:
+            raise ValueError("No access token to migrate")
+
+        request_body = {
+            "client_id": self.client_id,
+            "client_secret": self.client_secret,
+        }
+
+        headers = {
+            "Authorization": f"Bearer {credentials.access_token.get_secret_value()}",
+            "Content-Type": "application/x-www-form-urlencoded",
+        }
+
+        response = await Requests().post(
+            self.migrate_url, data=request_body, headers=headers
+        )
+
+        if not response.ok:
+            try:
+                error_data = response.json()
+                error_message = error_data.get("error", "Unknown error")
+                error_description = error_data.get("error_description", "")
+                if error_description:
+                    error_message = f"{error_message}: {error_description}"
+            except json.JSONDecodeError:
+                error_message = response.text
+            raise LinearAPIException(
+                f"Failed to migrate Linear token ({response.status}): {error_message}",
+                response.status,
+            )
+
+        token_data = response.json()
+
+        # Extract token expiration
+        now = int(time.time())
+        expires_in = token_data.get("expires_in")
+        access_token_expires_at = None
+        if expires_in:
+            access_token_expires_at = now + expires_in
+
+        new_credentials = OAuth2Credentials(
+            provider=self.PROVIDER_NAME,
+            title=credentials.title,
+            username=credentials.username,
+            access_token=token_data["access_token"],
+            scopes=credentials.scopes,  # Preserve original scopes
+            refresh_token=token_data.get("refresh_token"),
+            access_token_expires_at=access_token_expires_at,
+            refresh_token_expires_at=None,
+        )
+
+        new_credentials.id = credentials.id
+        return new_credentials
+
     async def _refresh_tokens(
         self, credentials: OAuth2Credentials
     ) -> OAuth2Credentials:
         if not credentials.refresh_token:
             raise ValueError(
-                "No refresh token available."
-            )  # Linear uses non-expiring tokens
+                "No refresh token available. Token may need to be migrated to the new refresh token system."
+            )
 
         return await self._request_tokens(
             {
                 "refresh_token": credentials.refresh_token.get_secret_value(),
                 "grant_type": "refresh_token",
-            }
+            },
+            current_credentials=credentials,
         )
 
     async def _request_tokens(
@@ -102,16 +170,33 @@ class LinearOAuthHandler(BaseOAuthHandler):
         params: dict[str, str],
         current_credentials: Optional[OAuth2Credentials] = None,
     ) -> OAuth2Credentials:
+        # Determine if this is a refresh token request
+        is_refresh = params.get("grant_type") == "refresh_token"
+
+        # Build request body with appropriate grant_type
         request_body = {
             "client_id": self.client_id,
             "client_secret": self.client_secret,
-            "grant_type": "authorization_code",  # Ensure grant_type is correct
             **params,
         }
 
-        headers = {
-            "Content-Type": "application/x-www-form-urlencoded"
-        }  # Correct header for token request
+        # Set default grant_type if not provided
+        if "grant_type" not in request_body:
+            request_body["grant_type"] = "authorization_code"
+
+        headers = {"Content-Type": "application/x-www-form-urlencoded"}
+
+        # For refresh token requests, support HTTP Basic Authentication as recommended
+        if is_refresh:
+            # Option 1: Use HTTP Basic Auth (preferred by Linear)
+            client_credentials = f"{self.client_id}:{self.client_secret}"
+            encoded_credentials = base64.b64encode(client_credentials.encode()).decode()
+            headers["Authorization"] = f"Basic {encoded_credentials}"
+
+            # Remove client credentials from body when using Basic Auth
+            request_body.pop("client_id", None)
+            request_body.pop("client_secret", None)
+
         response = await Requests().post(
             self.token_url, data=request_body, headers=headers
         )
@@ -120,6 +205,9 @@ class LinearOAuthHandler(BaseOAuthHandler):
             try:
                 error_data = response.json()
                 error_message = error_data.get("error", "Unknown error")
+                error_description = error_data.get("error_description", "")
+                if error_description:
+                    error_message = f"{error_message}: {error_description}"
             except json.JSONDecodeError:
                 error_message = response.text
             raise LinearAPIException(
@@ -129,27 +217,84 @@ class LinearOAuthHandler(BaseOAuthHandler):
 
         token_data = response.json()
 
-        # Note: Linear access tokens do not expire, so we set expires_at to None
+        # Extract token expiration if provided (for new refresh token implementation)
+        now = int(time.time())
+        expires_in = token_data.get("expires_in")
+        access_token_expires_at = None
+        if expires_in:
+            access_token_expires_at = now + expires_in
+
+        # Get username - preserve from current credentials if refreshing
+        username = None
+        if current_credentials and is_refresh:
+            username = current_credentials.username
+        elif "user" in token_data:
+            username = token_data["user"].get("name", "Unknown User")
+        else:
+            # Fetch username using the access token
+            username = await self._request_username(token_data["access_token"])
+
         new_credentials = OAuth2Credentials(
             provider=self.PROVIDER_NAME,
             title=current_credentials.title if current_credentials else None,
-            username=token_data.get("user", {}).get(
-                "name", "Unknown User"
-            ),  # extract name or set appropriate
+            username=username or "Unknown User",
             access_token=token_data["access_token"],
-            scopes=token_data["scope"].split(
-                ","
-            ),  # Linear returns comma-separated scopes
-            refresh_token=token_data.get(
-                "refresh_token"
-            ),  # Linear uses non-expiring tokens so this might be null
-            access_token_expires_at=None,
-            refresh_token_expires_at=None,
+            scopes=(
+                token_data["scope"].split(",")
+                if "scope" in token_data
+                else (current_credentials.scopes if current_credentials else [])
+            ),
+            refresh_token=token_data.get("refresh_token"),
+            access_token_expires_at=access_token_expires_at,
+            refresh_token_expires_at=None,  # Linear doesn't provide refresh token expiration
         )
+
         if current_credentials:
             new_credentials.id = current_credentials.id
+
         return new_credentials
 
+    async def get_access_token(self, credentials: OAuth2Credentials) -> str:
+        """
+        Returns a valid access token, handling migration and refresh as needed.
+
+        This overrides the base implementation to handle Linear's token migration
+        from old long-lived tokens to new short-lived tokens with refresh tokens.
+        """
+        # If token has no expiration and no refresh token, it might be an old token
+        # that needs migration
+        if (
+            credentials.access_token_expires_at is None
+            and credentials.refresh_token is None
+        ):
+            try:
+                # Attempt to migrate the old token
+                migrated_credentials = await self.migrate_old_token(credentials)
+                # Update the credentials store would need to be handled by the caller
+                # For now, use the migrated credentials for this request
+                credentials = migrated_credentials
+            except LinearAPIException:
+                # Migration failed, try to use the old token as-is
+                # This maintains backward compatibility
+                pass
+
+        # Use the standard refresh logic from the base class
+        if self.needs_refresh(credentials):
+            credentials = await self.refresh_tokens(credentials)
+
+        return credentials.access_token.get_secret_value()
+
+    def needs_migration(self, credentials: OAuth2Credentials) -> bool:
+        """
+        Check if credentials represent an old long-lived token that needs migration.
+
+        Old tokens have no expiration time and no refresh token.
+        """
+        return (
+            credentials.access_token_expires_at is None
+            and credentials.refresh_token is None
+        )
+
     async def _request_username(self, access_token: str) -> Optional[str]:
         # Use the LinearClient to fetch user details using GraphQL
         from ._api import LinearClient

autogpt_platform/backend/backend/blocks/linear/models.py

@@ -37,5 +37,5 @@ class Project(BaseModel):
     name: str
     description: str
     priority: int
-    progress: int
-    content: str
+    progress: float
+    content: str | None

autogpt_platform/backend/backend/blocks/llm.py

@@ -102,9 +102,8 @@ class LlmModel(str, Enum, metaclass=LlmModelMeta):
     CLAUDE_4_OPUS = "claude-opus-4-20250514"
     CLAUDE_4_SONNET = "claude-sonnet-4-20250514"
     CLAUDE_4_5_SONNET = "claude-sonnet-4-5-20250929"
+    CLAUDE_4_5_HAIKU = "claude-haiku-4-5-20251001"
     CLAUDE_3_7_SONNET = "claude-3-7-sonnet-20250219"
-    CLAUDE_3_5_SONNET = "claude-3-5-sonnet-latest"
-    CLAUDE_3_5_HAIKU = "claude-3-5-haiku-latest"
     CLAUDE_3_HAIKU = "claude-3-haiku-20240307"
     # AI/ML API models
     AIML_API_QWEN2_5_72B = "Qwen/Qwen2.5-72B-Instruct-Turbo"
@@ -217,15 +216,12 @@ MODEL_METADATA = {
     LlmModel.CLAUDE_4_5_SONNET: ModelMetadata(
         "anthropic", 200000, 64000
     ),  # claude-sonnet-4-5-20250929
+    LlmModel.CLAUDE_4_5_HAIKU: ModelMetadata(
+        "anthropic", 200000, 64000
+    ),  # claude-haiku-4-5-20251001
     LlmModel.CLAUDE_3_7_SONNET: ModelMetadata(
         "anthropic", 200000, 64000
     ),  # claude-3-7-sonnet-20250219
-    LlmModel.CLAUDE_3_5_SONNET: ModelMetadata(
-        "anthropic", 200000, 8192
-    ),  # claude-3-5-sonnet-20241022
-    LlmModel.CLAUDE_3_5_HAIKU: ModelMetadata(
-        "anthropic", 200000, 8192
-    ),  # claude-3-5-haiku-20241022
     LlmModel.CLAUDE_3_HAIKU: ModelMetadata(
         "anthropic", 200000, 4096
     ),  # claude-3-haiku-20240307
@@ -1558,7 +1554,9 @@ class AIConversationBlock(AIBlockBase):
                 ("prompt", list),
             ],
             test_mock={
-                "llm_call": lambda *args, **kwargs: "The 2020 World Series was played at Globe Life Field in Arlington, Texas."
+                "llm_call": lambda *args, **kwargs: dict(
+                    response="The 2020 World Series was played at Globe Life Field in Arlington, Texas."
+                )
             },
         )
 
@@ -1587,7 +1585,7 @@ class AIConversationBlock(AIBlockBase):
             ),
             credentials=credentials,
         )
-        yield "response", response
+        yield "response", response["response"]
         yield "prompt", self.prompt
 
 

autogpt_platform/backend/backend/blocks/perplexity.py

@@ -0,0 +1,226 @@
+# flake8: noqa: E501
+import logging
+from enum import Enum
+from typing import Any, Literal
+
+import openai
+from pydantic import SecretStr
+
+from backend.data.block import Block, BlockCategory, BlockOutput, BlockSchema
+from backend.data.model import (
+    APIKeyCredentials,
+    CredentialsField,
+    CredentialsMetaInput,
+    NodeExecutionStats,
+    SchemaField,
+)
+from backend.integrations.providers import ProviderName
+from backend.util.logging import TruncatedLogger
+
+logger = TruncatedLogger(logging.getLogger(__name__), "[Perplexity-Block]")
+
+
+class PerplexityModel(str, Enum):
+    """Perplexity sonar models available via OpenRouter"""
+
+    SONAR = "perplexity/sonar"
+    SONAR_PRO = "perplexity/sonar-pro"
+    SONAR_DEEP_RESEARCH = "perplexity/sonar-deep-research"
+
+
+PerplexityCredentials = CredentialsMetaInput[
+    Literal[ProviderName.OPEN_ROUTER], Literal["api_key"]
+]
+
+TEST_CREDENTIALS = APIKeyCredentials(
+    id="test-perplexity-creds",
+    provider="open_router",
+    api_key=SecretStr("mock-openrouter-api-key"),
+    title="Mock OpenRouter API key",
+    expires_at=None,
+)
+TEST_CREDENTIALS_INPUT = {
+    "provider": TEST_CREDENTIALS.provider,
+    "id": TEST_CREDENTIALS.id,
+    "type": TEST_CREDENTIALS.type,
+    "title": TEST_CREDENTIALS.title,
+}
+
+
+def PerplexityCredentialsField() -> PerplexityCredentials:
+    return CredentialsField(
+        description="OpenRouter API key for accessing Perplexity models.",
+    )
+
+
+class PerplexityBlock(Block):
+    class Input(BlockSchema):
+        prompt: str = SchemaField(
+            description="The query to send to the Perplexity model.",
+            placeholder="Enter your query here...",
+        )
+        model: PerplexityModel = SchemaField(
+            title="Perplexity Model",
+            default=PerplexityModel.SONAR,
+            description="The Perplexity sonar model to use.",
+            advanced=False,
+        )
+        credentials: PerplexityCredentials = PerplexityCredentialsField()
+        system_prompt: str = SchemaField(
+            title="System Prompt",
+            default="",
+            description="Optional system prompt to provide context to the model.",
+            advanced=True,
+        )
+        max_tokens: int | None = SchemaField(
+            advanced=True,
+            default=None,
+            description="The maximum number of tokens to generate.",
+        )
+
+    class Output(BlockSchema):
+        response: str = SchemaField(
+            description="The response from the Perplexity model."
+        )
+        annotations: list[dict[str, Any]] = SchemaField(
+            description="List of URL citations and annotations from the response."
+        )
+        error: str = SchemaField(description="Error message if the API call failed.")
+
+    def __init__(self):
+        super().__init__(
+            id="c8a5f2e9-8b3d-4a7e-9f6c-1d5e3c9b7a4f",
+            description="Query Perplexity's sonar models with real-time web search capabilities and receive annotated responses with source citations.",
+            categories={BlockCategory.AI, BlockCategory.SEARCH},
+            input_schema=PerplexityBlock.Input,
+            output_schema=PerplexityBlock.Output,
+            test_input={
+                "prompt": "What is the weather today?",
+                "model": PerplexityModel.SONAR,
+                "credentials": TEST_CREDENTIALS_INPUT,
+            },
+            test_credentials=TEST_CREDENTIALS,
+            test_output=[
+                ("response", "The weather varies by location..."),
+                ("annotations", list),
+            ],
+            test_mock={
+                "call_perplexity": lambda *args, **kwargs: {
+                    "response": "The weather varies by location...",
+                    "annotations": [
+                        {
+                            "type": "url_citation",
+                            "url_citation": {
+                                "title": "weather.com",
+                                "url": "https://weather.com",
+                            },
+                        }
+                    ],
+                }
+            },
+        )
+        self.execution_stats = NodeExecutionStats()
+
+    async def call_perplexity(
+        self,
+        credentials: APIKeyCredentials,
+        model: PerplexityModel,
+        prompt: str,
+        system_prompt: str = "",
+        max_tokens: int | None = None,
+    ) -> dict[str, Any]:
+        """Call Perplexity via OpenRouter and extract annotations."""
+        client = openai.AsyncOpenAI(
+            base_url="https://openrouter.ai/api/v1",
+            api_key=credentials.api_key.get_secret_value(),
+        )
+
+        messages = []
+        if system_prompt:
+            messages.append({"role": "system", "content": system_prompt})
+        messages.append({"role": "user", "content": prompt})
+
+        try:
+            response = await client.chat.completions.create(
+                extra_headers={
+                    "HTTP-Referer": "https://agpt.co",
+                    "X-Title": "AutoGPT",
+                },
+                model=model.value,
+                messages=messages,
+                max_tokens=max_tokens,
+            )
+
+            if not response.choices:
+                raise ValueError("No response from Perplexity via OpenRouter.")
+
+            # Extract the response content
+            response_content = response.choices[0].message.content or ""
+
+            # Extract annotations if present in the message
+            annotations = []
+            if hasattr(response.choices[0].message, "annotations"):
+                # If annotations are directly available
+                annotations = response.choices[0].message.annotations
+            else:
+                # Check if there's a raw response with annotations
+                raw = getattr(response.choices[0].message, "_raw_response", None)
+                if isinstance(raw, dict) and "annotations" in raw:
+                    annotations = raw["annotations"]
+
+            if not annotations and hasattr(response, "model_extra"):
+                # Check model_extra for annotations
+                model_extra = response.model_extra
+                if isinstance(model_extra, dict):
+                    # Check in choices
+                    if "choices" in model_extra and len(model_extra["choices"]) > 0:
+                        choice = model_extra["choices"][0]
+                        if "message" in choice and "annotations" in choice["message"]:
+                            annotations = choice["message"]["annotations"]
+
+            # Also check the raw response object for annotations
+            if not annotations:
+                raw = getattr(response, "_raw_response", None)
+                if isinstance(raw, dict):
+                    # Check various possible locations for annotations
+                    if "annotations" in raw:
+                        annotations = raw["annotations"]
+                    elif "choices" in raw and len(raw["choices"]) > 0:
+                        choice = raw["choices"][0]
+                        if "message" in choice and "annotations" in choice["message"]:
+                            annotations = choice["message"]["annotations"]
+
+            # Update execution stats
+            if response.usage:
+                self.execution_stats.input_token_count = response.usage.prompt_tokens
+                self.execution_stats.output_token_count = (
+                    response.usage.completion_tokens
+                )
+
+            return {"response": response_content, "annotations": annotations or []}
+
+        except Exception as e:
+            logger.error(f"Error calling Perplexity: {e}")
+            raise
+
+    async def run(
+        self, input_data: Input, *, credentials: APIKeyCredentials, **kwargs
+    ) -> BlockOutput:
+        logger.debug(f"Running Perplexity block with model: {input_data.model}")
+
+        try:
+            result = await self.call_perplexity(
+                credentials=credentials,
+                model=input_data.model,
+                prompt=input_data.prompt,
+                system_prompt=input_data.system_prompt,
+                max_tokens=input_data.max_tokens,
+            )
+
+            yield "response", result["response"]
+            yield "annotations", result["annotations"]
+
+        except Exception as e:
+            error_msg = f"Error calling Perplexity: {str(e)}"
+            logger.error(error_msg)
+            yield "error", error_msg

autogpt_platform/backend/backend/blocks/rss.py

@@ -1,7 +1,5 @@
 import asyncio
 import logging
-import urllib.parse
-import urllib.request
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
@@ -10,6 +8,7 @@ import pydantic
 
 from backend.data.block import Block, BlockCategory, BlockOutput, BlockSchema
 from backend.data.model import SchemaField
+from backend.util.request import Requests
 
 
 class RSSEntry(pydantic.BaseModel):
@@ -103,35 +102,29 @@ class ReadRSSFeedBlock(Block):
         )
 
     @staticmethod
-    def parse_feed(url: str) -> dict[str, Any]:
+    async def parse_feed(url: str) -> dict[str, Any]:
         # Security fix: Add protection against memory exhaustion attacks
         MAX_FEED_SIZE = 10 * 1024 * 1024  # 10MB limit for RSS feeds
 
-        # Validate URL
-        parsed_url = urllib.parse.urlparse(url)
-        if parsed_url.scheme not in ("http", "https"):
-            raise ValueError(f"Invalid URL scheme: {parsed_url.scheme}")
-
-        # Download with size limit
+        # Download feed content with size limit
         try:
-            with urllib.request.urlopen(url, timeout=30) as response:
-                # Check content length if available
-                content_length = response.headers.get("Content-Length")
-                if content_length and int(content_length) > MAX_FEED_SIZE:
-                    raise ValueError(
-                        f"Feed too large: {content_length} bytes exceeds {MAX_FEED_SIZE} limit"
-                    )
+            response = await Requests(raise_for_status=True).get(url)
 
-                # Read with size limit
-                content = response.read(MAX_FEED_SIZE + 1)
-                if len(content) > MAX_FEED_SIZE:
-                    raise ValueError(
-                        f"Feed too large: exceeds {MAX_FEED_SIZE} byte limit"
-                    )
+            # Check content length if available
+            content_length = response.headers.get("Content-Length")
+            if content_length and int(content_length) > MAX_FEED_SIZE:
+                raise ValueError(
+                    f"Feed too large: {content_length} bytes exceeds {MAX_FEED_SIZE} limit"
+                )
 
-                # Parse with feedparser using the validated content
-                # feedparser has built-in protection against XML attacks
-                return feedparser.parse(content)  # type: ignore
+            # Get content with size limit
+            content = response.content
+            if len(content) > MAX_FEED_SIZE:
+                raise ValueError(f"Feed too large: exceeds {MAX_FEED_SIZE} byte limit")
+
+            # Parse with feedparser using the validated content
+            # feedparser has built-in protection against XML attacks
+            return feedparser.parse(content)  # type: ignore
         except Exception as e:
             # Log error and return empty feed
             logging.warning(f"Failed to parse RSS feed from {url}: {e}")
@@ -145,7 +138,7 @@ class ReadRSSFeedBlock(Block):
         while keep_going:
             keep_going = input_data.run_continuously
 
-            feed = self.parse_feed(input_data.rss_url)
+            feed = await self.parse_feed(input_data.rss_url)
             all_entries = []
 
             for entry in feed["entries"]:

autogpt_platform/backend/backend/blocks/stagehand/blocks.py

@@ -1,6 +1,7 @@
 import logging
 import signal
 import threading
+import warnings
 from contextlib import contextmanager
 from enum import Enum
 
@@ -26,6 +27,13 @@ from backend.sdk import (
     SchemaField,
 )
 
+# Suppress false positive cleanup warning of litellm (a dependency of stagehand)
+warnings.filterwarnings(
+    "ignore",
+    message="coroutine 'close_litellm_async_clients' was never awaited",
+    category=RuntimeWarning,
+)
+
 # Store the original method
 original_register_signal_handlers = stagehand.main.Stagehand._register_signal_handlers
 

autogpt_platform/backend/backend/blocks/test/test_llm.py

@@ -362,7 +362,7 @@ class TestLLMStatsTracking:
         assert block.execution_stats.llm_call_count == 1
 
         # Check output
-        assert outputs["response"] == {"response": "AI response to conversation"}
+        assert outputs["response"] == "AI response to conversation"
 
     @pytest.mark.asyncio
     async def test_ai_list_generator_with_retries(self):

autogpt_platform/backend/backend/blocks/youtube.py

@@ -1,6 +1,7 @@
 from urllib.parse import parse_qs, urlparse
 
 from youtube_transcript_api._api import YouTubeTranscriptApi
+from youtube_transcript_api._errors import NoTranscriptFound
 from youtube_transcript_api._transcripts import FetchedTranscript
 from youtube_transcript_api.formatters import TextFormatter
 
@@ -64,7 +65,29 @@ class TranscribeYoutubeVideoBlock(Block):
 
     @staticmethod
     def get_transcript(video_id: str) -> FetchedTranscript:
-        return YouTubeTranscriptApi().fetch(video_id=video_id)
+        """
+        Get transcript for a video, preferring English but falling back to any available language.
+
+        :param video_id: The YouTube video ID
+        :return: The fetched transcript
+        :raises: Any exception except NoTranscriptFound for requested languages
+        """
+        api = YouTubeTranscriptApi()
+        try:
+            # Try to get English transcript first (default behavior)
+            return api.fetch(video_id=video_id)
+        except NoTranscriptFound:
+            # If English is not available, get the first available transcript
+            transcript_list = api.list(video_id)
+            # Try manually created transcripts first, then generated ones
+            available_transcripts = list(
+                transcript_list._manually_created_transcripts.values()
+            ) + list(transcript_list._generated_transcripts.values())
+            if available_transcripts:
+                # Fetch the first available transcript
+                return available_transcripts[0].fetch()
+            # If no transcripts at all, re-raise the original error
+            raise
 
     @staticmethod
     def format_transcript(transcript: FetchedTranscript) -> str:

autogpt_platform/backend/backend/cli.py

@@ -45,9 +45,6 @@ class MainApp(AppProcess):
 
         app.main(silent=True)
 
-    def cleanup(self):
-        pass
-
 
 @click.group()
 def main():

autogpt_platform/backend/backend/data/block.py

@@ -20,7 +20,6 @@ from typing import (
 
 import jsonref
 import jsonschema
-from autogpt_libs.utils.cache import cached
 from prisma.models import AgentBlock
 from prisma.types import AgentBlockCreateInput
 from pydantic import BaseModel
@@ -28,6 +27,7 @@ from pydantic import BaseModel
 from backend.data.model import NodeExecutionStats
 from backend.integrations.providers import ProviderName
 from backend.util import json
+from backend.util.cache import cached
 from backend.util.settings import Config
 
 from .model import (
@@ -722,7 +722,7 @@ def get_block(block_id: str) -> Block[BlockSchema, BlockSchema] | None:
     return cls() if cls else None
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def get_webhook_block_ids() -> Sequence[str]:
     return [
         id
@@ -731,7 +731,7 @@ def get_webhook_block_ids() -> Sequence[str]:
     ]
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def get_io_block_ids() -> Sequence[str]:
     return [
         id

autogpt_platform/backend/backend/data/block_cost_config.py

@@ -1,7 +1,11 @@
 from typing import Type
 
 from backend.blocks.ai_music_generator import AIMusicGeneratorBlock
-from backend.blocks.ai_shortform_video_block import AIShortformVideoCreatorBlock
+from backend.blocks.ai_shortform_video_block import (
+    AIAdMakerVideoCreatorBlock,
+    AIScreenshotToVideoAdBlock,
+    AIShortformVideoCreatorBlock,
+)
 from backend.blocks.apollo.organization import SearchOrganizationsBlock
 from backend.blocks.apollo.people import SearchPeopleBlock
 from backend.blocks.apollo.person import GetPersonDetailBlock
@@ -69,10 +73,9 @@ MODEL_COST: dict[LlmModel, int] = {
     LlmModel.CLAUDE_4_1_OPUS: 21,
     LlmModel.CLAUDE_4_OPUS: 21,
     LlmModel.CLAUDE_4_SONNET: 5,
+    LlmModel.CLAUDE_4_5_HAIKU: 4,
     LlmModel.CLAUDE_4_5_SONNET: 9,
     LlmModel.CLAUDE_3_7_SONNET: 5,
-    LlmModel.CLAUDE_3_5_SONNET: 4,
-    LlmModel.CLAUDE_3_5_HAIKU: 1,  # $0.80 / $4.00
     LlmModel.CLAUDE_3_HAIKU: 1,
     LlmModel.AIML_API_QWEN2_5_72B: 1,
     LlmModel.AIML_API_LLAMA3_1_70B: 1,
@@ -322,7 +325,31 @@ BLOCK_COSTS: dict[Type[Block], list[BlockCost]] = {
     ],
     AIShortformVideoCreatorBlock: [
         BlockCost(
-            cost_amount=50,
+            cost_amount=307,
+            cost_filter={
+                "credentials": {
+                    "id": revid_credentials.id,
+                    "provider": revid_credentials.provider,
+                    "type": revid_credentials.type,
+                }
+            },
+        )
+    ],
+    AIAdMakerVideoCreatorBlock: [
+        BlockCost(
+            cost_amount=714,
+            cost_filter={
+                "credentials": {
+                    "id": revid_credentials.id,
+                    "provider": revid_credentials.provider,
+                    "type": revid_credentials.type,
+                }
+            },
+        )
+    ],
+    AIScreenshotToVideoAdBlock: [
+        BlockCost(
+            cost_amount=612,
             cost_filter={
                 "credentials": {
                     "id": revid_credentials.id,

autogpt_platform/backend/backend/data/credit.py

@@ -5,7 +5,6 @@ from datetime import datetime, timezone
 from typing import TYPE_CHECKING, Any, cast
 
 import stripe
-from prisma import Json
 from prisma.enums import (
     CreditRefundRequestStatus,
     CreditTransactionType,
@@ -13,16 +12,12 @@ from prisma.enums import (
     OnboardingStep,
 )
 from prisma.errors import UniqueViolationError
-from prisma.models import CreditRefundRequest, CreditTransaction, User
-from prisma.types import (
-    CreditRefundRequestCreateInput,
-    CreditTransactionCreateInput,
-    CreditTransactionWhereInput,
-)
+from prisma.models import CreditRefundRequest, CreditTransaction, User, UserBalance
+from prisma.types import CreditRefundRequestCreateInput, CreditTransactionWhereInput
 from pydantic import BaseModel
 
-from backend.data import db
 from backend.data.block_cost_config import BLOCK_COSTS
+from backend.data.db import query_raw_with_schema
 from backend.data.includes import MAX_CREDIT_REFUND_REQUESTS_FETCH
 from backend.data.model import (
     AutoTopUpConfig,
@@ -36,7 +31,8 @@ from backend.data.user import get_user_by_id, get_user_email_by_id
 from backend.notifications.notifications import queue_notification_async
 from backend.server.v2.admin.model import UserHistoryResponse
 from backend.util.exceptions import InsufficientBalanceError
-from backend.util.json import SafeJson
+from backend.util.feature_flag import Flag, is_feature_enabled
+from backend.util.json import SafeJson, dumps
 from backend.util.models import Pagination
 from backend.util.retry import func_retry
 from backend.util.settings import Settings
@@ -49,6 +45,10 @@ stripe.api_key = settings.secrets.stripe_api_key
 logger = logging.getLogger(__name__)
 base_url = settings.config.frontend_base_url or settings.config.platform_base_url
 
+# Constants for test compatibility
+POSTGRES_INT_MAX = 2147483647
+POSTGRES_INT_MIN = -2147483648
+
 
 class UsageTransactionMetadata(BaseModel):
     graph_exec_id: str | None = None
@@ -139,14 +139,20 @@ class UserCreditBase(ABC):
         pass
 
     @abstractmethod
-    async def onboarding_reward(self, user_id: str, credits: int, step: OnboardingStep):
+    async def onboarding_reward(
+        self, user_id: str, credits: int, step: OnboardingStep
+    ) -> bool:
         """
         Reward the user with credits for completing an onboarding step.
         Won't reward if the user has already received credits for the step.
 
         Args:
             user_id (str): The user ID.
+            credits (int): The amount to reward.
             step (OnboardingStep): The onboarding step.
+
+        Returns:
+            bool: True if rewarded, False if already rewarded.
         """
         pass
 
@@ -236,6 +242,12 @@ class UserCreditBase(ABC):
         """
         Returns the current balance of the user & the latest balance snapshot time.
         """
+        # Check UserBalance first for efficiency and consistency
+        user_balance = await UserBalance.prisma().find_unique(where={"userId": user_id})
+        if user_balance:
+            return user_balance.balance, user_balance.updatedAt
+
+        # Fallback to transaction history computation if UserBalance doesn't exist
         top_time = self.time_now()
         snapshot = await CreditTransaction.prisma().find_first(
             where={
@@ -250,72 +262,86 @@ class UserCreditBase(ABC):
         snapshot_balance = snapshot.runningBalance or 0 if snapshot else 0
         snapshot_time = snapshot.createdAt if snapshot else datetime_min
 
-        # Get transactions after the snapshot, this should not exist, but just in case.
-        transactions = await CreditTransaction.prisma().group_by(
-            by=["userId"],
-            sum={"amount": True},
-            max={"createdAt": True},
-            where={
-                "userId": user_id,
-                "createdAt": {
-                    "gt": snapshot_time,
-                    "lte": top_time,
-                },
-                "isActive": True,
-            },
-        )
-        transaction_balance = (
-            int(transactions[0].get("_sum", {}).get("amount", 0) + snapshot_balance)
-            if transactions
-            else snapshot_balance
-        )
-        transaction_time = (
-            datetime.fromisoformat(
-                str(transactions[0].get("_max", {}).get("createdAt", datetime_min))
-            )
-            if transactions
-            else snapshot_time
-        )
-        return transaction_balance, transaction_time
+        return snapshot_balance, snapshot_time
 
     @func_retry
     async def _enable_transaction(
         self,
         transaction_key: str,
         user_id: str,
-        metadata: Json,
+        metadata: SafeJson,
         new_transaction_key: str | None = None,
     ):
-        transaction = await CreditTransaction.prisma().find_first_or_raise(
-            where={"transactionKey": transaction_key, "userId": user_id}
+        # First check if transaction exists and is inactive (safety check)
+        transaction = await CreditTransaction.prisma().find_first(
+            where={
+                "transactionKey": transaction_key,
+                "userId": user_id,
+                "isActive": False,
+            }
         )
-        if transaction.isActive:
-            return
+        if not transaction:
+            # Transaction doesn't exist or is already active, return early
+            return None
 
-        async with db.locked_transaction(f"usr_trx_{user_id}"):
-
-            transaction = await CreditTransaction.prisma().find_first_or_raise(
-                where={"transactionKey": transaction_key, "userId": user_id}
+        # Atomic operation to enable transaction and update user balance using UserBalance
+        result = await query_raw_with_schema(
+            """
+            WITH user_balance_lock AS (
+                SELECT 
+                    $2::text as userId, 
+                    COALESCE(
+                        (SELECT balance FROM {schema_prefix}"UserBalance" WHERE "userId" = $2 FOR UPDATE),
+                        -- Fallback: compute balance from transaction history if UserBalance doesn't exist
+                        (SELECT COALESCE(ct."runningBalance", 0) 
+                         FROM {schema_prefix}"CreditTransaction" ct 
+                         WHERE ct."userId" = $2 
+                           AND ct."isActive" = true 
+                           AND ct."runningBalance" IS NOT NULL 
+                         ORDER BY ct."createdAt" DESC 
+                         LIMIT 1),
+                        0
+                    ) as balance
+            ),
+            transaction_check AS (
+                SELECT * FROM {schema_prefix}"CreditTransaction" 
+                WHERE "transactionKey" = $1 AND "userId" = $2 AND "isActive" = false
+            ),
+            balance_update AS (
+                INSERT INTO {schema_prefix}"UserBalance" ("userId", "balance", "updatedAt")
+                SELECT 
+                    $2::text,
+                    user_balance_lock.balance + transaction_check.amount,
+                    CURRENT_TIMESTAMP
+                FROM user_balance_lock, transaction_check
+                ON CONFLICT ("userId") DO UPDATE SET
+                    "balance" = EXCLUDED."balance",
+                    "updatedAt" = EXCLUDED."updatedAt"
+                RETURNING "balance", "updatedAt"
+            ),
+            transaction_update AS (
+                UPDATE {schema_prefix}"CreditTransaction"
+                SET "transactionKey" = COALESCE($4, $1),
+                    "isActive" = true,
+                    "runningBalance" = balance_update.balance,
+                    "createdAt" = balance_update."updatedAt",
+                    "metadata" = $3::jsonb
+                FROM balance_update, transaction_check
+                WHERE {schema_prefix}"CreditTransaction"."transactionKey" = transaction_check."transactionKey"
+                  AND {schema_prefix}"CreditTransaction"."userId" = transaction_check."userId"
+                RETURNING {schema_prefix}"CreditTransaction"."runningBalance"
             )
-            if transaction.isActive:
-                return
+            SELECT "runningBalance" as balance FROM transaction_update;
+            """,
+            transaction_key,  # $1
+            user_id,  # $2
+            dumps(metadata.data),  # $3 - use pre-serialized JSON string for JSONB
+            new_transaction_key,  # $4
+        )
 
-            user_balance, _ = await self._get_credits(user_id)
-            await CreditTransaction.prisma().update(
-                where={
-                    "creditTransactionIdentifier": {
-                        "transactionKey": transaction_key,
-                        "userId": user_id,
-                    }
-                },
-                data={
-                    "transactionKey": new_transaction_key or transaction_key,
-                    "isActive": True,
-                    "runningBalance": user_balance + transaction.amount,
-                    "createdAt": self.time_now(),
-                    "metadata": metadata,
-                },
-            )
+        if result:
+            # UserBalance is already updated by the CTE
+            return result[0]["balance"]
 
     async def _add_transaction(
         self,
@@ -326,12 +352,54 @@ class UserCreditBase(ABC):
         transaction_key: str | None = None,
         ceiling_balance: int | None = None,
         fail_insufficient_credits: bool = True,
-        metadata: Json = SafeJson({}),
+        metadata: SafeJson = SafeJson({}),
     ) -> tuple[int, str]:
         """
         Add a new transaction for the user.
         This is the only method that should be used to add a new transaction.
 
+        ATOMIC OPERATION DESIGN DECISION:
+        ================================
+        This method uses PostgreSQL row-level locking (FOR UPDATE) for atomic credit operations.
+        After extensive analysis of concurrency patterns and correctness requirements, we determined
+        that the FOR UPDATE approach is necessary despite the latency overhead.
+
+        WHY FOR UPDATE LOCKING IS REQUIRED:
+        ----------------------------------
+        1. **Data Consistency**: Credit operations must be ACID-compliant. The balance check,
+           calculation, and update must be atomic to prevent race conditions where:
+           - Multiple spend operations could exceed available balance
+           - Lost update problems could occur with concurrent top-ups
+           - Refunds could create negative balances incorrectly
+
+        2. **Serializability**: FOR UPDATE ensures operations are serialized at the database level,
+           guaranteeing that each transaction sees a consistent view of the balance before applying changes.
+
+        3. **Correctness Over Performance**: Financial operations require absolute correctness.
+           The ~10-50ms latency increase from row locking is acceptable for the guarantee that
+           no user will ever have an incorrect balance due to race conditions.
+
+        4. **PostgreSQL Optimization**: Modern PostgreSQL versions optimize row locks efficiently.
+           The performance cost is minimal compared to the complexity and risk of lock-free approaches.
+
+        ALTERNATIVES CONSIDERED AND REJECTED:
+        ------------------------------------
+        - **Optimistic Concurrency**: Using version numbers or timestamps would require complex
+          retry logic and could still fail under high contention scenarios.
+        - **Application-Level Locking**: Redis locks or similar would add network overhead and
+          single points of failure while being less reliable than database locks.
+        - **Event Sourcing**: Would require complete architectural changes and eventual consistency
+          models that don't fit our real-time balance requirements.
+
+        PERFORMANCE CHARACTERISTICS:
+        ---------------------------
+        - Single user operations: 10-50ms latency (acceptable for financial operations)
+        - Concurrent operations on same user: Serialized (prevents data corruption)
+        - Concurrent operations on different users: Fully parallel (no blocking)
+
+        This design prioritizes correctness and data integrity over raw performance,
+        which is the appropriate choice for a credit/payment system.
+
         Args:
             user_id (str): The user ID.
             amount (int): The amount of credits to add.
@@ -345,40 +413,142 @@ class UserCreditBase(ABC):
         Returns:
             tuple[int, str]: The new balance & the transaction key.
         """
-        async with db.locked_transaction(f"usr_trx_{user_id}"):
-            # Get latest balance snapshot
-            user_balance, _ = await self._get_credits(user_id)
-
-            if ceiling_balance and amount > 0 and user_balance >= ceiling_balance:
+        # Quick validation for ceiling balance to avoid unnecessary database operations
+        if ceiling_balance and amount > 0:
+            current_balance, _ = await self._get_credits(user_id)
+            if current_balance >= ceiling_balance:
                 raise ValueError(
-                    f"You already have enough balance of ${user_balance/100}, top-up is not required when you already have at least ${ceiling_balance/100}"
+                    f"You already have enough balance of ${current_balance/100}, top-up is not required when you already have at least ${ceiling_balance/100}"
                 )
 
-            if amount < 0 and user_balance + amount < 0:
-                if fail_insufficient_credits:
-                    raise InsufficientBalanceError(
-                        message=f"Insufficient balance of ${user_balance/100}, where this will cost ${abs(amount)/100}",
-                        user_id=user_id,
-                        balance=user_balance,
-                        amount=amount,
+        # Single unified atomic operation for all transaction types using UserBalance
+        try:
+            result = await query_raw_with_schema(
+                """
+                WITH user_balance_lock AS (
+                    SELECT 
+                        $1::text as userId, 
+                        -- CRITICAL: FOR UPDATE lock prevents concurrent modifications to the same user's balance
+                        -- This ensures atomic read-modify-write operations and prevents race conditions
+                        COALESCE(
+                            (SELECT balance FROM {schema_prefix}"UserBalance" WHERE "userId" = $1 FOR UPDATE),
+                            -- Fallback: compute balance from transaction history if UserBalance doesn't exist
+                            (SELECT COALESCE(ct."runningBalance", 0) 
+                             FROM {schema_prefix}"CreditTransaction" ct 
+                             WHERE ct."userId" = $1 
+                               AND ct."isActive" = true 
+                               AND ct."runningBalance" IS NOT NULL 
+                             ORDER BY ct."createdAt" DESC 
+                             LIMIT 1),
+                            0
+                        ) as balance
+                ),
+                balance_update AS (
+                    INSERT INTO {schema_prefix}"UserBalance" ("userId", "balance", "updatedAt")
+                    SELECT 
+                        $1::text,
+                        CASE 
+                            -- For inactive transactions: Don't update balance
+                            WHEN $5::boolean = false THEN user_balance_lock.balance
+                            -- For ceiling balance (amount > 0): Apply ceiling
+                            WHEN $2 > 0 AND $7::int IS NOT NULL AND user_balance_lock.balance > $7::int - $2 THEN $7::int
+                            -- For regular operations: Apply with overflow/underflow protection  
+                            WHEN user_balance_lock.balance + $2 > $6::int THEN $6::int
+                            WHEN user_balance_lock.balance + $2 < $10::int THEN $10::int
+                            ELSE user_balance_lock.balance + $2
+                        END,
+                        CURRENT_TIMESTAMP
+                    FROM user_balance_lock
+                    WHERE (
+                        $5::boolean = false OR  -- Allow inactive transactions
+                        $2 >= 0 OR              -- Allow positive amounts (top-ups, grants)
+                        $8::boolean = false OR  -- Allow when insufficient balance check is disabled
+                        user_balance_lock.balance + $2 >= 0  -- Allow spending only when sufficient balance
                     )
+                    ON CONFLICT ("userId") DO UPDATE SET
+                        "balance" = EXCLUDED."balance",
+                        "updatedAt" = EXCLUDED."updatedAt"
+                    RETURNING "balance", "updatedAt"
+                ),
+                transaction_insert AS (
+                    INSERT INTO {schema_prefix}"CreditTransaction" (
+                        "userId", "amount", "type", "runningBalance", 
+                        "metadata", "isActive", "createdAt", "transactionKey"
+                    )
+                    SELECT 
+                        $1::text,
+                        $2::int,
+                        $3::text::{schema_prefix}"CreditTransactionType",
+                        CASE 
+                            -- For inactive transactions: Set runningBalance to original balance (don't apply the change yet)
+                            WHEN $5::boolean = false THEN user_balance_lock.balance
+                            ELSE COALESCE(balance_update.balance, user_balance_lock.balance)
+                        END,
+                        $4::jsonb,
+                        $5::boolean,
+                        COALESCE(balance_update."updatedAt", CURRENT_TIMESTAMP),
+                        COALESCE($9, gen_random_uuid()::text)
+                    FROM user_balance_lock
+                    LEFT JOIN balance_update ON true
+                    WHERE (
+                        $5::boolean = false OR  -- Allow inactive transactions
+                        $2 >= 0 OR              -- Allow positive amounts (top-ups, grants)
+                        $8::boolean = false OR  -- Allow when insufficient balance check is disabled
+                        user_balance_lock.balance + $2 >= 0  -- Allow spending only when sufficient balance
+                    )
+                    RETURNING "runningBalance", "transactionKey"
+                )
+                SELECT "runningBalance" as balance, "transactionKey" FROM transaction_insert;
+                """,
+                user_id,  # $1
+                amount,  # $2
+                transaction_type.value,  # $3
+                dumps(metadata.data),  # $4 - use pre-serialized JSON string for JSONB
+                is_active,  # $5
+                POSTGRES_INT_MAX,  # $6 - overflow protection
+                ceiling_balance,  # $7 - ceiling balance (nullable)
+                fail_insufficient_credits,  # $8 - check balance for spending
+                transaction_key,  # $9 - transaction key (nullable)
+                POSTGRES_INT_MIN,  # $10 - underflow protection
+            )
+        except Exception as e:
+            # Convert raw SQL unique constraint violations to UniqueViolationError
+            # for consistent exception handling throughout the codebase
+            error_str = str(e).lower()
+            if (
+                "already exists" in error_str
+                or "duplicate key" in error_str
+                or "unique constraint" in error_str
+            ):
+                # Extract table and constraint info for better error messages
+                # Re-raise as a UniqueViolationError but with proper format
+                # Create a minimal data structure that the error constructor expects
+                raise UniqueViolationError({"error": str(e), "user_facing_error": {}})
+            # For any other error, re-raise as-is
+            raise
 
-                amount = min(-user_balance, 0)
+        if result:
+            new_balance, tx_key = result[0]["balance"], result[0]["transactionKey"]
+            # UserBalance is already updated by the CTE
+            return new_balance, tx_key
 
-            # Create the transaction
-            transaction_data: CreditTransactionCreateInput = {
-                "userId": user_id,
-                "amount": amount,
-                "runningBalance": user_balance + amount,
-                "type": transaction_type,
-                "metadata": metadata,
-                "isActive": is_active,
-                "createdAt": self.time_now(),
-            }
-            if transaction_key:
-                transaction_data["transactionKey"] = transaction_key
-            tx = await CreditTransaction.prisma().create(data=transaction_data)
-            return user_balance + amount, tx.transactionKey
+        # If no result, either user doesn't exist or insufficient balance
+        user = await User.prisma().find_unique(where={"id": user_id})
+        if not user:
+            raise ValueError(f"User {user_id} not found")
+
+        # Must be insufficient balance for spending operation
+        if amount < 0 and fail_insufficient_credits:
+            current_balance, _ = await self._get_credits(user_id)
+            raise InsufficientBalanceError(
+                message=f"Insufficient balance of ${current_balance/100}, where this will cost ${abs(amount)/100}",
+                user_id=user_id,
+                balance=current_balance,
+                amount=amount,
+            )
+
+        # Unexpected case
+        raise ValueError(f"Transaction failed for user {user_id}, amount {amount}")
 
 
 class UserCredit(UserCreditBase):
@@ -453,9 +623,10 @@ class UserCredit(UserCreditBase):
                     {"reason": f"Reward for completing {step.value} onboarding step."}
                 ),
             )
+            return True
         except UniqueViolationError:
-            # Already rewarded for this step
-            pass
+            # User already received this reward
+            return False
 
     async def top_up_refund(
         self, user_id: str, transaction_key: str, metadata: dict[str, str]
@@ -644,7 +815,7 @@ class UserCredit(UserCreditBase):
     ):
         # init metadata, without sharing it with the world
         metadata = metadata or {}
-        if not metadata["reason"]:
+        if not metadata.get("reason"):
             match top_up_type:
                 case TopUpType.MANUAL:
                     metadata["reason"] = {"reason": f"Top up credits for {user_id}"}
@@ -974,8 +1145,8 @@ class DisabledUserCredit(UserCreditBase):
     async def top_up_credits(self, *args, **kwargs):
         pass
 
-    async def onboarding_reward(self, *args, **kwargs):
-        pass
+    async def onboarding_reward(self, *args, **kwargs) -> bool:
+        return True
 
     async def top_up_intent(self, *args, **kwargs) -> str:
         return ""
@@ -993,14 +1164,31 @@ class DisabledUserCredit(UserCreditBase):
         pass
 
 
-def get_user_credit_model() -> UserCreditBase:
+async def get_user_credit_model(user_id: str) -> UserCreditBase:
+    """
+    Get the credit model for a user, considering LaunchDarkly flags.
+
+    Args:
+        user_id (str): The user ID to check flags for.
+
+    Returns:
+        UserCreditBase: The appropriate credit model for the user
+    """
     if not settings.config.enable_credit:
         return DisabledUserCredit()
 
-    if settings.config.enable_beta_monthly_credit:
-        return BetaUserCredit(settings.config.num_user_credits_refill)
+    # Check LaunchDarkly flag for payment pilot users
+    # Default to False (beta monthly credit behavior) to maintain current behavior
+    is_payment_enabled = await is_feature_enabled(
+        Flag.ENABLE_PLATFORM_PAYMENT, user_id, default=False
+    )
 
-    return UserCredit()
+    if is_payment_enabled:
+        # Payment enabled users get UserCredit (no monthly refills, enable payments)
+        return UserCredit()
+    else:
+        # Default behavior: users get beta monthly credits
+        return BetaUserCredit(settings.config.num_user_credits_refill)
 
 
 def get_block_costs() -> dict[str, list["BlockCost"]]:
@@ -1090,7 +1278,8 @@ async def admin_get_user_history(
             )
             reason = metadata.get("reason", "No reason provided")
 
-        balance, last_update = await get_user_credit_model()._get_credits(tx.userId)
+        user_credit_model = await get_user_credit_model(tx.userId)
+        balance, _ = await user_credit_model._get_credits(tx.userId)
 
         history.append(
             UserTransaction(

autogpt_platform/backend/backend/data/credit_ceiling_test.py

@@ -0,0 +1,172 @@
+"""
+Test ceiling balance functionality to ensure auto top-up limits work correctly.
+
+This test was added to cover a previously untested code path that could lead to
+incorrect balance capping behavior.
+"""
+
+from uuid import uuid4
+
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.errors import UniqueViolationError
+from prisma.models import CreditTransaction, User, UserBalance
+
+from backend.data.credit import UserCredit
+from backend.util.json import SafeJson
+from backend.util.test import SpinTestServer
+
+
+async def create_test_user(user_id: str) -> None:
+    """Create a test user for ceiling tests."""
+    try:
+        await User.prisma().create(
+            data={
+                "id": user_id,
+                "email": f"test-{user_id}@example.com",
+                "name": f"Test User {user_id[:8]}",
+            }
+        )
+    except UniqueViolationError:
+        # User already exists, continue
+        pass
+
+    await UserBalance.prisma().upsert(
+        where={"userId": user_id},
+        data={"create": {"userId": user_id, "balance": 0}, "update": {"balance": 0}},
+    )
+
+
+async def cleanup_test_user(user_id: str) -> None:
+    """Clean up test user and their transactions."""
+    try:
+        await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+        await User.prisma().delete_many(where={"id": user_id})
+    except Exception as e:
+        # Log cleanup failures but don't fail the test
+        print(f"Warning: Failed to cleanup test user {user_id}: {e}")
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_ceiling_balance_rejects_when_above_threshold(server: SpinTestServer):
+    """Test that ceiling balance correctly rejects top-ups when balance is above threshold."""
+    credit_system = UserCredit()
+    user_id = f"ceiling-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user balance of 1000 ($10) using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=1000,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "initial_balance"}),
+        )
+        current_balance = await credit_system.get_credits(user_id)
+        assert current_balance == 1000
+
+        # Try to add 200 more with ceiling of 800 (should reject since 1000 > 800)
+        with pytest.raises(ValueError, match="You already have enough balance"):
+            await credit_system._add_transaction(
+                user_id=user_id,
+                amount=200,
+                transaction_type=CreditTransactionType.TOP_UP,
+                ceiling_balance=800,  # Ceiling lower than current balance
+            )
+
+        # Balance should remain unchanged
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 1000, f"Balance should remain 1000, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_ceiling_balance_clamps_when_would_exceed(server: SpinTestServer):
+    """Test that ceiling balance correctly clamps amounts that would exceed the ceiling."""
+    credit_system = UserCredit()
+    user_id = f"ceiling-clamp-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user balance of 500 ($5) using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=500,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "initial_balance"}),
+        )
+
+        # Add 800 more with ceiling of 1000 (should clamp to 1000, not reach 1300)
+        final_balance, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=800,
+            transaction_type=CreditTransactionType.TOP_UP,
+            ceiling_balance=1000,  # Ceiling should clamp 500 + 800 = 1300 to 1000
+        )
+
+        # Balance should be clamped to ceiling
+        assert (
+            final_balance == 1000
+        ), f"Balance should be clamped to 1000, got {final_balance}"
+
+        # Verify with get_credits too
+        stored_balance = await credit_system.get_credits(user_id)
+        assert (
+            stored_balance == 1000
+        ), f"Stored balance should be 1000, got {stored_balance}"
+
+        # Verify transaction shows the clamped amount
+        transactions = await CreditTransaction.prisma().find_many(
+            where={"userId": user_id, "type": CreditTransactionType.TOP_UP},
+            order={"createdAt": "desc"},
+        )
+
+        # Should have 2 transactions: 500 + (500 to reach ceiling of 1000)
+        assert len(transactions) == 2
+
+        # The second transaction should show it only added 500, not 800
+        second_tx = transactions[0]  # Most recent
+        assert second_tx.runningBalance == 1000
+        # The actual amount recorded could be 800 (what was requested) but balance was clamped
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_ceiling_balance_allows_when_under_threshold(server: SpinTestServer):
+    """Test that ceiling balance allows top-ups when balance is under threshold."""
+    credit_system = UserCredit()
+    user_id = f"ceiling-under-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user balance of 300 ($3) using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=300,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "initial_balance"}),
+        )
+
+        # Add 200 more with ceiling of 1000 (should succeed: 300 + 200 = 500 < 1000)
+        final_balance, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=200,
+            transaction_type=CreditTransactionType.TOP_UP,
+            ceiling_balance=1000,
+        )
+
+        # Balance should be exactly 500
+        assert final_balance == 500, f"Balance should be 500, got {final_balance}"
+
+        # Verify with get_credits too
+        stored_balance = await credit_system.get_credits(user_id)
+        assert (
+            stored_balance == 500
+        ), f"Stored balance should be 500, got {stored_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)

autogpt_platform/backend/backend/data/credit_concurrency_test.py

@@ -0,0 +1,737 @@
+"""
+Concurrency and atomicity tests for the credit system.
+
+These tests ensure the credit system handles high-concurrency scenarios correctly
+without race conditions, deadlocks, or inconsistent state.
+"""
+
+import asyncio
+import random
+from uuid import uuid4
+
+import prisma.enums
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.errors import UniqueViolationError
+from prisma.models import CreditTransaction, User, UserBalance
+
+from backend.data.credit import POSTGRES_INT_MAX, UsageTransactionMetadata, UserCredit
+from backend.util.exceptions import InsufficientBalanceError
+from backend.util.json import SafeJson
+from backend.util.test import SpinTestServer
+
+# Test with both UserCredit and BetaUserCredit if needed
+credit_system = UserCredit()
+
+
+async def create_test_user(user_id: str) -> None:
+    """Create a test user with initial balance."""
+    try:
+        await User.prisma().create(
+            data={
+                "id": user_id,
+                "email": f"test-{user_id}@example.com",
+                "name": f"Test User {user_id[:8]}",
+            }
+        )
+    except UniqueViolationError:
+        # User already exists, continue
+        pass
+
+    # Ensure UserBalance record exists
+    await UserBalance.prisma().upsert(
+        where={"userId": user_id},
+        data={"create": {"userId": user_id, "balance": 0}, "update": {"balance": 0}},
+    )
+
+
+async def cleanup_test_user(user_id: str) -> None:
+    """Clean up test user and their transactions."""
+    try:
+        await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+        await UserBalance.prisma().delete_many(where={"userId": user_id})
+        await User.prisma().delete_many(where={"id": user_id})
+    except Exception as e:
+        # Log cleanup failures but don't fail the test
+        print(f"Warning: Failed to cleanup test user {user_id}: {e}")
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_spends_same_user(server: SpinTestServer):
+    """Test multiple concurrent spends from the same user don't cause race conditions."""
+    user_id = f"concurrent-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user initial balance using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=1000,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "initial_balance"}),
+        )
+
+        # Try to spend 10 x $1 concurrently
+        async def spend_one_dollar(idx: int):
+            try:
+                return await credit_system.spend_credits(
+                    user_id,
+                    100,  # $1
+                    UsageTransactionMetadata(
+                        graph_exec_id=f"concurrent-{idx}",
+                        reason=f"Concurrent spend {idx}",
+                    ),
+                )
+            except InsufficientBalanceError:
+                return None
+
+        # Run 10 concurrent spends
+        results = await asyncio.gather(
+            *[spend_one_dollar(i) for i in range(10)], return_exceptions=True
+        )
+
+        # Count successful spends
+        successful = [
+            r for r in results if r is not None and not isinstance(r, Exception)
+        ]
+        failed = [r for r in results if isinstance(r, InsufficientBalanceError)]
+
+        # All 10 should succeed since we have exactly $10
+        assert len(successful) == 10, f"Expected 10 successful, got {len(successful)}"
+        assert len(failed) == 0, f"Expected 0 failures, got {len(failed)}"
+
+        # Final balance should be exactly 0
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 0, f"Expected balance 0, got {final_balance}"
+
+        # Verify transaction history is consistent
+        transactions = await CreditTransaction.prisma().find_many(
+            where={"userId": user_id, "type": prisma.enums.CreditTransactionType.USAGE}
+        )
+        assert (
+            len(transactions) == 10
+        ), f"Expected 10 transactions, got {len(transactions)}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_spends_insufficient_balance(server: SpinTestServer):
+    """Test that concurrent spends correctly enforce balance limits."""
+    user_id = f"insufficient-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user limited balance using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=500,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "limited_balance"}),
+        )
+
+        # Try to spend 10 x $1 concurrently (but only have $5)
+        async def spend_one_dollar(idx: int):
+            try:
+                return await credit_system.spend_credits(
+                    user_id,
+                    100,  # $1
+                    UsageTransactionMetadata(
+                        graph_exec_id=f"insufficient-{idx}",
+                        reason=f"Insufficient spend {idx}",
+                    ),
+                )
+            except InsufficientBalanceError:
+                return "FAILED"
+
+        # Run 10 concurrent spends
+        results = await asyncio.gather(
+            *[spend_one_dollar(i) for i in range(10)], return_exceptions=True
+        )
+
+        # Count successful vs failed
+        successful = [
+            r
+            for r in results
+            if r not in ["FAILED", None] and not isinstance(r, Exception)
+        ]
+        failed = [r for r in results if r == "FAILED"]
+
+        # Exactly 5 should succeed, 5 should fail
+        assert len(successful) == 5, f"Expected 5 successful, got {len(successful)}"
+        assert len(failed) == 5, f"Expected 5 failures, got {len(failed)}"
+
+        # Final balance should be exactly 0
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 0, f"Expected balance 0, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_mixed_operations(server: SpinTestServer):
+    """Test concurrent mix of spends, top-ups, and balance checks."""
+    user_id = f"mixed-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Initial balance using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=1000,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "initial_balance"}),
+        )
+
+        # Mix of operations
+        async def mixed_operations():
+            operations = []
+
+            # 5 spends of $1 each
+            for i in range(5):
+                operations.append(
+                    credit_system.spend_credits(
+                        user_id,
+                        100,
+                        UsageTransactionMetadata(reason=f"Mixed spend {i}"),
+                    )
+                )
+
+            # 3 top-ups of $2 each using internal method
+            for i in range(3):
+                operations.append(
+                    credit_system._add_transaction(
+                        user_id=user_id,
+                        amount=200,
+                        transaction_type=CreditTransactionType.TOP_UP,
+                        metadata=SafeJson({"test": f"concurrent_topup_{i}"}),
+                    )
+                )
+
+            # 10 balance checks
+            for i in range(10):
+                operations.append(credit_system.get_credits(user_id))
+
+            return await asyncio.gather(*operations, return_exceptions=True)
+
+        results = await mixed_operations()
+
+        # Check no exceptions occurred
+        exceptions = [
+            r
+            for r in results
+            if isinstance(r, Exception) and not isinstance(r, InsufficientBalanceError)
+        ]
+        assert len(exceptions) == 0, f"Unexpected exceptions: {exceptions}"
+
+        # Final balance should be: 1000 - 500 + 600 = 1100
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 1100, f"Expected balance 1100, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_race_condition_exact_balance(server: SpinTestServer):
+    """Test spending exact balance amount concurrently doesn't go negative."""
+    user_id = f"exact-balance-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give exact amount using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=100,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "exact_amount"}),
+        )
+
+        # Try to spend $1 twice concurrently
+        async def spend_exact():
+            try:
+                return await credit_system.spend_credits(
+                    user_id, 100, UsageTransactionMetadata(reason="Exact spend")
+                )
+            except InsufficientBalanceError:
+                return "FAILED"
+
+        # Both try to spend the full balance
+        result1, result2 = await asyncio.gather(spend_exact(), spend_exact())
+
+        # Exactly one should succeed
+        results = [result1, result2]
+        successful = [
+            r for r in results if r != "FAILED" and not isinstance(r, Exception)
+        ]
+        failed = [r for r in results if r == "FAILED"]
+
+        assert len(successful) == 1, f"Expected 1 success, got {len(successful)}"
+        assert len(failed) == 1, f"Expected 1 failure, got {len(failed)}"
+
+        # Balance should be exactly 0, never negative
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 0, f"Expected balance 0, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_onboarding_reward_idempotency(server: SpinTestServer):
+    """Test that onboarding rewards are idempotent (can't be claimed twice)."""
+    user_id = f"onboarding-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Use WELCOME step which is defined in the OnboardingStep enum
+        # Try to claim same reward multiple times concurrently
+        async def claim_reward():
+            try:
+                result = await credit_system.onboarding_reward(
+                    user_id, 500, prisma.enums.OnboardingStep.WELCOME
+                )
+                return "SUCCESS" if result else "DUPLICATE"
+            except Exception as e:
+                print(f"Claim reward failed: {e}")
+                return "FAILED"
+
+        # Try 5 concurrent claims of the same reward
+        results = await asyncio.gather(*[claim_reward() for _ in range(5)])
+
+        # Count results
+        success_count = results.count("SUCCESS")
+        failed_count = results.count("FAILED")
+
+        # At least one should succeed, others should be duplicates
+        assert success_count >= 1, f"At least one claim should succeed, got {results}"
+        assert failed_count == 0, f"No claims should fail, got {results}"
+
+        # Check balance - should only have 500, not 2500
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 500, f"Expected balance 500, got {final_balance}"
+
+        # Check only one transaction exists
+        transactions = await CreditTransaction.prisma().find_many(
+            where={
+                "userId": user_id,
+                "type": prisma.enums.CreditTransactionType.GRANT,
+                "transactionKey": f"REWARD-{user_id}-WELCOME",
+            }
+        )
+        assert (
+            len(transactions) == 1
+        ), f"Expected 1 reward transaction, got {len(transactions)}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_integer_overflow_protection(server: SpinTestServer):
+    """Test that integer overflow is prevented by clamping to POSTGRES_INT_MAX."""
+    user_id = f"overflow-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Try to add amount that would overflow
+        max_int = POSTGRES_INT_MAX
+
+        # First, set balance near max
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": max_int - 100},
+                "update": {"balance": max_int - 100},
+            },
+        )
+
+        # Try to add more than possible - should clamp to POSTGRES_INT_MAX
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=200,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "overflow_protection"}),
+        )
+
+        # Balance should be clamped to max_int, not overflowed
+        final_balance = await credit_system.get_credits(user_id)
+        assert (
+            final_balance == max_int
+        ), f"Balance should be clamped to {max_int}, got {final_balance}"
+
+        # Verify transaction was created with clamped amount
+        transactions = await CreditTransaction.prisma().find_many(
+            where={
+                "userId": user_id,
+                "type": prisma.enums.CreditTransactionType.TOP_UP,
+            },
+            order={"createdAt": "desc"},
+        )
+        assert len(transactions) > 0, "Transaction should be created"
+        assert (
+            transactions[0].runningBalance == max_int
+        ), "Transaction should show clamped balance"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_high_concurrency_stress(server: SpinTestServer):
+    """Stress test with many concurrent operations."""
+    user_id = f"stress-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Initial balance using internal method (bypasses Stripe)
+        initial_balance = 10000  # $100
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=initial_balance,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "stress_test_balance"}),
+        )
+
+        # Run many concurrent operations
+        async def random_operation(idx: int):
+            operation = random.choice(["spend", "check"])
+
+            if operation == "spend":
+                amount = random.randint(1, 50)  # $0.01 to $0.50
+                try:
+                    return (
+                        "spend",
+                        amount,
+                        await credit_system.spend_credits(
+                            user_id,
+                            amount,
+                            UsageTransactionMetadata(reason=f"Stress {idx}"),
+                        ),
+                    )
+                except InsufficientBalanceError:
+                    return ("spend_failed", amount, None)
+            else:
+                balance = await credit_system.get_credits(user_id)
+                return ("check", 0, balance)
+
+        # Run 100 concurrent operations
+        results = await asyncio.gather(
+            *[random_operation(i) for i in range(100)], return_exceptions=True
+        )
+
+        # Calculate expected final balance
+        total_spent = sum(
+            r[1]
+            for r in results
+            if not isinstance(r, Exception) and isinstance(r, tuple) and r[0] == "spend"
+        )
+        expected_balance = initial_balance - total_spent
+
+        # Verify final balance
+        final_balance = await credit_system.get_credits(user_id)
+        assert (
+            final_balance == expected_balance
+        ), f"Expected {expected_balance}, got {final_balance}"
+        assert final_balance >= 0, "Balance went negative!"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_multiple_spends_sufficient_balance(server: SpinTestServer):
+    """Test multiple concurrent spends when there's sufficient balance for all."""
+    user_id = f"multi-spend-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Give user 150 balance ($1.50) using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=150,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "sufficient_balance"}),
+        )
+
+        # Track individual timing to see serialization
+        timings = {}
+
+        async def spend_with_detailed_timing(amount: int, label: str):
+            start = asyncio.get_event_loop().time()
+            try:
+                await credit_system.spend_credits(
+                    user_id,
+                    amount,
+                    UsageTransactionMetadata(
+                        graph_exec_id=f"concurrent-{label}",
+                        reason=f"Concurrent spend {label}",
+                    ),
+                )
+                end = asyncio.get_event_loop().time()
+                timings[label] = {"start": start, "end": end, "duration": end - start}
+                return f"{label}-SUCCESS"
+            except Exception as e:
+                end = asyncio.get_event_loop().time()
+                timings[label] = {
+                    "start": start,
+                    "end": end,
+                    "duration": end - start,
+                    "error": str(e),
+                }
+                return f"{label}-FAILED: {e}"
+
+        # Run concurrent spends: 10, 20, 30 (total 60, well under 150)
+        overall_start = asyncio.get_event_loop().time()
+        results = await asyncio.gather(
+            spend_with_detailed_timing(10, "spend-10"),
+            spend_with_detailed_timing(20, "spend-20"),
+            spend_with_detailed_timing(30, "spend-30"),
+            return_exceptions=True,
+        )
+        overall_end = asyncio.get_event_loop().time()
+
+        print(f"Results: {results}")
+        print(f"Overall duration: {overall_end - overall_start:.4f}s")
+
+        # Analyze timing to detect serialization vs true concurrency
+        print("\nTiming analysis:")
+        for label, timing in timings.items():
+            print(
+                f"  {label}: started at {timing['start']:.4f}, ended at {timing['end']:.4f}, duration {timing['duration']:.4f}s"
+            )
+
+        # Check if operations overlapped (true concurrency) or were serialized
+        sorted_timings = sorted(timings.items(), key=lambda x: x[1]["start"])
+        print("\nExecution order by start time:")
+        for i, (label, timing) in enumerate(sorted_timings):
+            print(f"  {i+1}. {label}: {timing['start']:.4f} -> {timing['end']:.4f}")
+
+        # Check for overlap (true concurrency) vs serialization
+        overlaps = []
+        for i in range(len(sorted_timings) - 1):
+            current = sorted_timings[i]
+            next_op = sorted_timings[i + 1]
+            if current[1]["end"] > next_op[1]["start"]:
+                overlaps.append(f"{current[0]} overlaps with {next_op[0]}")
+
+        if overlaps:
+            print(f"✅ TRUE CONCURRENCY detected: {overlaps}")
+        else:
+            print("🔒 SERIALIZATION detected: No overlapping execution times")
+
+        # Check final balance
+        final_balance = await credit_system.get_credits(user_id)
+        print(f"Final balance: {final_balance}")
+
+        # Count successes/failures
+        successful = [r for r in results if "SUCCESS" in str(r)]
+        failed = [r for r in results if "FAILED" in str(r)]
+
+        print(f"Successful: {len(successful)}, Failed: {len(failed)}")
+
+        # All should succeed since 150 - (10 + 20 + 30) = 90 > 0
+        assert (
+            len(successful) == 3
+        ), f"Expected all 3 to succeed, got {len(successful)} successes: {results}"
+        assert final_balance == 90, f"Expected balance 90, got {final_balance}"
+
+        # Check transaction timestamps to confirm database-level serialization
+        transactions = await CreditTransaction.prisma().find_many(
+            where={"userId": user_id, "type": prisma.enums.CreditTransactionType.USAGE},
+            order={"createdAt": "asc"},
+        )
+        print("\nDatabase transaction order (by createdAt):")
+        for i, tx in enumerate(transactions):
+            print(
+                f"  {i+1}. Amount {tx.amount}, Running balance: {tx.runningBalance}, Created: {tx.createdAt}"
+            )
+
+        # Verify running balances are chronologically consistent (ordered by createdAt)
+        actual_balances = [
+            tx.runningBalance for tx in transactions if tx.runningBalance is not None
+        ]
+        print(f"Running balances: {actual_balances}")
+
+        # The balances should be valid intermediate states regardless of execution order
+        # Starting balance: 150, spending 10+20+30=60, so final should be 90
+        # The intermediate balances depend on execution order but should all be valid
+        expected_possible_balances = {
+            # If order is 10, 20, 30: [140, 120, 90]
+            # If order is 10, 30, 20: [140, 110, 90]
+            # If order is 20, 10, 30: [130, 120, 90]
+            # If order is 20, 30, 10: [130, 100, 90]
+            # If order is 30, 10, 20: [120, 110, 90]
+            # If order is 30, 20, 10: [120, 100, 90]
+            90,
+            100,
+            110,
+            120,
+            130,
+            140,  # All possible intermediate balances
+        }
+
+        # Verify all balances are valid intermediate states
+        for balance in actual_balances:
+            assert (
+                balance in expected_possible_balances
+            ), f"Invalid balance {balance}, expected one of {expected_possible_balances}"
+
+        # Final balance should always be 90 (150 - 60)
+        assert (
+            min(actual_balances) == 90
+        ), f"Final balance should be 90, got {min(actual_balances)}"
+
+        # The final transaction should always have balance 90
+        # The other transactions should have valid intermediate balances
+        assert (
+            90 in actual_balances
+        ), f"Final balance 90 should be in actual_balances: {actual_balances}"
+
+        # All balances should be >= 90 (the final state)
+        assert all(
+            balance >= 90 for balance in actual_balances
+        ), f"All balances should be >= 90, got {actual_balances}"
+
+        # CRITICAL: Transactions are atomic but can complete in any order
+        # What matters is that all running balances are valid intermediate states
+        # Each balance should be between 90 (final) and 140 (after first transaction)
+        for balance in actual_balances:
+            assert (
+                90 <= balance <= 140
+            ), f"Balance {balance} is outside valid range [90, 140]"
+
+        # Final balance (minimum) should always be 90
+        assert (
+            min(actual_balances) == 90
+        ), f"Final balance should be 90, got {min(actual_balances)}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_prove_database_locking_behavior(server: SpinTestServer):
+    """Definitively prove whether database locking causes waiting vs failures."""
+    user_id = f"locking-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Set balance to exact amount that can handle all spends using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=60,  # Exactly 10+20+30
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "exact_amount_test"}),
+        )
+
+        async def spend_with_precise_timing(amount: int, label: str):
+            request_start = asyncio.get_event_loop().time()
+            db_operation_start = asyncio.get_event_loop().time()
+            try:
+                # Add a small delay to increase chance of true concurrency
+                await asyncio.sleep(0.001)
+
+                db_operation_start = asyncio.get_event_loop().time()
+                await credit_system.spend_credits(
+                    user_id,
+                    amount,
+                    UsageTransactionMetadata(
+                        graph_exec_id=f"locking-{label}",
+                        reason=f"Locking test {label}",
+                    ),
+                )
+                db_operation_end = asyncio.get_event_loop().time()
+
+                return {
+                    "label": label,
+                    "status": "SUCCESS",
+                    "request_start": request_start,
+                    "db_start": db_operation_start,
+                    "db_end": db_operation_end,
+                    "db_duration": db_operation_end - db_operation_start,
+                }
+            except Exception as e:
+                db_operation_end = asyncio.get_event_loop().time()
+                return {
+                    "label": label,
+                    "status": "FAILED",
+                    "error": str(e),
+                    "request_start": request_start,
+                    "db_start": db_operation_start,
+                    "db_end": db_operation_end,
+                    "db_duration": db_operation_end - db_operation_start,
+                }
+
+        # Launch all requests simultaneously
+        results = await asyncio.gather(
+            spend_with_precise_timing(10, "A"),
+            spend_with_precise_timing(20, "B"),
+            spend_with_precise_timing(30, "C"),
+            return_exceptions=True,
+        )
+
+        print("\n🔍 LOCKING BEHAVIOR ANALYSIS:")
+        print("=" * 50)
+
+        successful = [
+            r for r in results if isinstance(r, dict) and r.get("status") == "SUCCESS"
+        ]
+        failed = [
+            r for r in results if isinstance(r, dict) and r.get("status") == "FAILED"
+        ]
+
+        print(f"✅ Successful operations: {len(successful)}")
+        print(f"❌ Failed operations: {len(failed)}")
+
+        if len(failed) > 0:
+            print(
+                "\n🚫 CONCURRENT FAILURES - Some requests failed due to insufficient balance:"
+            )
+            for result in failed:
+                if isinstance(result, dict):
+                    print(
+                        f"   {result['label']}: {result.get('error', 'Unknown error')}"
+                    )
+
+        if len(successful) == 3:
+            print(
+                "\n🔒 SERIALIZATION CONFIRMED - All requests succeeded, indicating they were queued:"
+            )
+
+            # Sort by actual execution time to see order
+            dict_results = [r for r in results if isinstance(r, dict)]
+            sorted_results = sorted(dict_results, key=lambda x: x["db_start"])
+
+            for i, result in enumerate(sorted_results):
+                print(
+                    f"   {i+1}. {result['label']}: DB operation took {result['db_duration']:.4f}s"
+                )
+
+            # Check if any operations overlapped at the database level
+            print("\n⏱️  Database operation timeline:")
+            for result in sorted_results:
+                print(
+                    f"   {result['label']}: {result['db_start']:.4f} -> {result['db_end']:.4f}"
+                )
+
+        # Verify final state
+        final_balance = await credit_system.get_credits(user_id)
+        print(f"\n💰 Final balance: {final_balance}")
+
+        if len(successful) == 3:
+            assert (
+                final_balance == 0
+            ), f"If all succeeded, balance should be 0, got {final_balance}"
+            print(
+                "✅ CONCLUSION: Database row locking causes requests to WAIT and execute serially"
+            )
+        else:
+            print(
+                "❌ CONCLUSION: Some requests failed, indicating different concurrency behavior"
+            )
+
+    finally:
+        await cleanup_test_user(user_id)

autogpt_platform/backend/backend/data/credit_integration_test.py

@@ -0,0 +1,277 @@
+"""
+Integration tests for credit system to catch SQL enum casting issues.
+
+These tests run actual database operations to ensure SQL queries work correctly,
+which would have caught the CreditTransactionType enum casting bug.
+"""
+
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.models import CreditTransaction, User, UserBalance
+
+from backend.data.credit import (
+    AutoTopUpConfig,
+    BetaUserCredit,
+    UsageTransactionMetadata,
+    get_auto_top_up,
+    set_auto_top_up,
+)
+from backend.util.json import SafeJson
+
+
+@pytest.fixture
+async def cleanup_test_user():
+    """Clean up test user data before and after tests."""
+    import uuid
+
+    user_id = str(uuid.uuid4())  # Use unique user ID for each test
+
+    # Create the user first
+    try:
+        await User.prisma().create(
+            data={
+                "id": user_id,
+                "email": f"test-{user_id}@example.com",
+                "topUpConfig": SafeJson({}),
+                "timezone": "UTC",
+            }
+        )
+    except Exception:
+        # User might already exist, that's fine
+        pass
+
+    yield user_id
+
+    # Cleanup after test
+    await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+    await UserBalance.prisma().delete_many(where={"userId": user_id})
+    # Clear auto-top-up config before deleting user
+    await User.prisma().update(
+        where={"id": user_id}, data={"topUpConfig": SafeJson({})}
+    )
+    await User.prisma().delete(where={"id": user_id})
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_credit_transaction_enum_casting_integration(cleanup_test_user):
+    """
+    Integration test to verify CreditTransactionType enum casting works in SQL queries.
+
+    This test would have caught the enum casting bug where PostgreSQL expected
+    platform."CreditTransactionType" but got "CreditTransactionType".
+    """
+    user_id = cleanup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # Test each transaction type to ensure enum casting works
+    test_cases = [
+        (CreditTransactionType.TOP_UP, 100, "Test top-up"),
+        (CreditTransactionType.USAGE, -50, "Test usage"),
+        (CreditTransactionType.GRANT, 200, "Test grant"),
+        (CreditTransactionType.REFUND, -25, "Test refund"),
+        (CreditTransactionType.CARD_CHECK, 0, "Test card check"),
+    ]
+
+    for transaction_type, amount, reason in test_cases:
+        metadata = SafeJson({"reason": reason, "test": "enum_casting"})
+
+        # This call would fail with enum casting error before the fix
+        balance, tx_key = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=amount,
+            transaction_type=transaction_type,
+            metadata=metadata,
+            is_active=True,
+        )
+
+        # Verify transaction was created with correct type
+        transaction = await CreditTransaction.prisma().find_first(
+            where={"userId": user_id, "transactionKey": tx_key}
+        )
+
+        assert transaction is not None
+        assert transaction.type == transaction_type
+        assert transaction.amount == amount
+        assert transaction.metadata is not None
+
+        # Verify metadata content
+        assert transaction.metadata["reason"] == reason
+        assert transaction.metadata["test"] == "enum_casting"
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_auto_top_up_integration(cleanup_test_user, monkeypatch):
+    """
+    Integration test for auto-top-up functionality that triggers enum casting.
+
+    This tests the complete auto-top-up flow which involves SQL queries with
+    CreditTransactionType enums, ensuring enum casting works end-to-end.
+    """
+    # Enable credits for this test
+    from backend.data.credit import settings
+
+    monkeypatch.setattr(settings.config, "enable_credit", True)
+    monkeypatch.setattr(settings.config, "enable_beta_monthly_credit", True)
+    monkeypatch.setattr(settings.config, "num_user_credits_refill", 1000)
+
+    user_id = cleanup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # First add some initial credits so we can test the configuration and subsequent behavior
+    balance, _ = await credit_system._add_transaction(
+        user_id=user_id,
+        amount=50,  # Below threshold that we'll set
+        transaction_type=CreditTransactionType.GRANT,
+        metadata=SafeJson({"reason": "Initial credits before auto top-up config"}),
+    )
+    assert balance == 50
+
+    # Configure auto top-up with threshold above current balance
+    config = AutoTopUpConfig(threshold=100, amount=500)
+    await set_auto_top_up(user_id, config)
+
+    # Verify configuration was saved but no immediate top-up occurred
+    current_balance = await credit_system.get_credits(user_id)
+    assert current_balance == 50  # Balance should be unchanged
+
+    # Simulate spending credits that would trigger auto top-up
+    # This involves multiple SQL operations with enum casting
+    try:
+        metadata = UsageTransactionMetadata(reason="Test spend to trigger auto top-up")
+        await credit_system.spend_credits(user_id=user_id, cost=10, metadata=metadata)
+
+        # The auto top-up mechanism should have been triggered
+        # Verify the transaction types were handled correctly
+        transactions = await CreditTransaction.prisma().find_many(
+            where={"userId": user_id}, order={"createdAt": "desc"}
+        )
+
+        # Should have at least: GRANT (initial), USAGE (spend), and TOP_UP (auto top-up)
+        assert len(transactions) >= 3
+
+        # Verify different transaction types exist and enum casting worked
+        transaction_types = {t.type for t in transactions}
+        assert CreditTransactionType.GRANT in transaction_types
+        assert CreditTransactionType.USAGE in transaction_types
+        assert (
+            CreditTransactionType.TOP_UP in transaction_types
+        )  # Auto top-up should have triggered
+
+    except Exception as e:
+        # If this fails with enum casting error, the test successfully caught the bug
+        if "CreditTransactionType" in str(e) and (
+            "cast" in str(e).lower() or "type" in str(e).lower()
+        ):
+            pytest.fail(f"Enum casting error detected: {e}")
+        else:
+            # Re-raise other unexpected errors
+            raise
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_enable_transaction_enum_casting_integration(cleanup_test_user):
+    """
+    Integration test for _enable_transaction with enum casting.
+
+    Tests the scenario where inactive transactions are enabled, which also
+    involves SQL queries with CreditTransactionType enum casting.
+    """
+    user_id = cleanup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # Create an inactive transaction
+    balance, tx_key = await credit_system._add_transaction(
+        user_id=user_id,
+        amount=100,
+        transaction_type=CreditTransactionType.TOP_UP,
+        metadata=SafeJson({"reason": "Inactive transaction test"}),
+        is_active=False,  # Create as inactive
+    )
+
+    # Balance should be 0 since transaction is inactive
+    assert balance == 0
+
+    # Enable the transaction with new metadata
+    enable_metadata = SafeJson(
+        {
+            "payment_method": "test_payment",
+            "activation_reason": "Integration test activation",
+        }
+    )
+
+    # This would fail with enum casting error before the fix
+    final_balance = await credit_system._enable_transaction(
+        transaction_key=tx_key,
+        user_id=user_id,
+        metadata=enable_metadata,
+    )
+
+    # Now balance should reflect the activated transaction
+    assert final_balance == 100
+
+    # Verify transaction was properly enabled with correct enum type
+    transaction = await CreditTransaction.prisma().find_first(
+        where={"userId": user_id, "transactionKey": tx_key}
+    )
+
+    assert transaction is not None
+    assert transaction.isActive is True
+    assert transaction.type == CreditTransactionType.TOP_UP
+    assert transaction.runningBalance == 100
+
+    # Verify metadata was updated
+    assert transaction.metadata is not None
+    assert transaction.metadata["payment_method"] == "test_payment"
+    assert transaction.metadata["activation_reason"] == "Integration test activation"
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_auto_top_up_configuration_storage(cleanup_test_user, monkeypatch):
+    """
+    Test that auto-top-up configuration is properly stored and retrieved.
+
+    The immediate top-up logic is handled by the API routes, not the core
+    set_auto_top_up function. This test verifies the configuration is correctly
+    saved and can be retrieved.
+    """
+    # Enable credits for this test
+    from backend.data.credit import settings
+
+    monkeypatch.setattr(settings.config, "enable_credit", True)
+    monkeypatch.setattr(settings.config, "enable_beta_monthly_credit", True)
+    monkeypatch.setattr(settings.config, "num_user_credits_refill", 1000)
+
+    user_id = cleanup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # Set initial balance
+    balance, _ = await credit_system._add_transaction(
+        user_id=user_id,
+        amount=50,
+        transaction_type=CreditTransactionType.GRANT,
+        metadata=SafeJson({"reason": "Initial balance for config test"}),
+    )
+
+    assert balance == 50
+
+    # Configure auto top-up
+    config = AutoTopUpConfig(threshold=100, amount=200)
+    await set_auto_top_up(user_id, config)
+
+    # Verify the configuration was saved
+    retrieved_config = await get_auto_top_up(user_id)
+    assert retrieved_config.threshold == config.threshold
+    assert retrieved_config.amount == config.amount
+
+    # Verify balance is unchanged (no immediate top-up from set_auto_top_up)
+    final_balance = await credit_system.get_credits(user_id)
+    assert final_balance == 50  # Should be unchanged
+
+    # Verify no immediate auto-top-up transaction was created by set_auto_top_up
+    transactions = await CreditTransaction.prisma().find_many(
+        where={"userId": user_id}, order={"createdAt": "desc"}
+    )
+
+    # Should only have the initial GRANT transaction
+    assert len(transactions) == 1
+    assert transactions[0].type == CreditTransactionType.GRANT

autogpt_platform/backend/backend/data/credit_metadata_test.py

@@ -0,0 +1,141 @@
+"""
+Tests for credit system metadata handling to ensure JSON casting works correctly.
+
+This test verifies that metadata parameters are properly serialized when passed
+to raw SQL queries with JSONB columns.
+"""
+
+# type: ignore
+
+from typing import Any
+
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.models import CreditTransaction, UserBalance
+
+from backend.data.credit import BetaUserCredit
+from backend.data.user import DEFAULT_USER_ID
+from backend.util.json import SafeJson
+
+
+@pytest.fixture
+async def setup_test_user():
+    """Setup test user and cleanup after test."""
+    user_id = DEFAULT_USER_ID
+
+    # Cleanup before test
+    await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+    await UserBalance.prisma().delete_many(where={"userId": user_id})
+
+    yield user_id
+
+    # Cleanup after test
+    await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+    await UserBalance.prisma().delete_many(where={"userId": user_id})
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_metadata_json_serialization(setup_test_user):
+    """Test that metadata is properly serialized for JSONB column in raw SQL."""
+    user_id = setup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # Test with complex metadata that would fail if not properly serialized
+    complex_metadata = SafeJson(
+        {
+            "graph_exec_id": "test-12345",
+            "reason": "Testing metadata serialization",
+            "nested_data": {
+                "key1": "value1",
+                "key2": ["array", "of", "values"],
+                "key3": {"deeply": {"nested": "object"}},
+            },
+            "special_chars": "Testing 'quotes' and \"double quotes\" and unicode: 🚀",
+        }
+    )
+
+    # This should work without throwing a JSONB casting error
+    balance, tx_key = await credit_system._add_transaction(
+        user_id=user_id,
+        amount=500,  # $5 top-up
+        transaction_type=CreditTransactionType.TOP_UP,
+        metadata=complex_metadata,
+        is_active=True,
+    )
+
+    # Verify the transaction was created successfully
+    assert balance == 500
+
+    # Verify the metadata was stored correctly in the database
+    transaction = await CreditTransaction.prisma().find_first(
+        where={"userId": user_id, "transactionKey": tx_key}
+    )
+
+    assert transaction is not None
+    assert transaction.metadata is not None
+
+    # Verify the metadata contains our complex data
+    metadata_dict: dict[str, Any] = dict(transaction.metadata)  # type: ignore
+    assert metadata_dict["graph_exec_id"] == "test-12345"
+    assert metadata_dict["reason"] == "Testing metadata serialization"
+    assert metadata_dict["nested_data"]["key1"] == "value1"
+    assert metadata_dict["nested_data"]["key3"]["deeply"]["nested"] == "object"
+    assert (
+        metadata_dict["special_chars"]
+        == "Testing 'quotes' and \"double quotes\" and unicode: 🚀"
+    )
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_enable_transaction_metadata_serialization(setup_test_user):
+    """Test that _enable_transaction also handles metadata JSON serialization correctly."""
+    user_id = setup_test_user
+    credit_system = BetaUserCredit(1000)
+
+    # First create an inactive transaction
+    balance, tx_key = await credit_system._add_transaction(
+        user_id=user_id,
+        amount=300,
+        transaction_type=CreditTransactionType.TOP_UP,
+        metadata=SafeJson({"initial": "inactive_transaction"}),
+        is_active=False,  # Create as inactive
+    )
+
+    # Initial balance should be 0 because transaction is inactive
+    assert balance == 0
+
+    # Now enable the transaction with new metadata
+    enable_metadata = SafeJson(
+        {
+            "payment_method": "stripe",
+            "payment_intent": "pi_test_12345",
+            "activation_reason": "Payment confirmed",
+            "complex_data": {"array": [1, 2, 3], "boolean": True, "null_value": None},
+        }
+    )
+
+    # This should work without JSONB casting errors
+    final_balance = await credit_system._enable_transaction(
+        transaction_key=tx_key,
+        user_id=user_id,
+        metadata=enable_metadata,
+    )
+
+    # Now balance should reflect the activated transaction
+    assert final_balance == 300
+
+    # Verify the metadata was updated correctly
+    transaction = await CreditTransaction.prisma().find_first(
+        where={"userId": user_id, "transactionKey": tx_key}
+    )
+
+    assert transaction is not None
+    assert transaction.isActive is True
+
+    # Verify the metadata was updated with enable_metadata
+    metadata_dict: dict[str, Any] = dict(transaction.metadata)  # type: ignore
+    assert metadata_dict["payment_method"] == "stripe"
+    assert metadata_dict["payment_intent"] == "pi_test_12345"
+    assert metadata_dict["complex_data"]["array"] == [1, 2, 3]
+    assert metadata_dict["complex_data"]["boolean"] is True
+    assert metadata_dict["complex_data"]["null_value"] is None

autogpt_platform/backend/backend/data/credit_refund_test.py

@@ -0,0 +1,372 @@
+"""
+Tests for credit system refund and dispute operations.
+
+These tests ensure that refund operations (deduct_credits, handle_dispute)
+are atomic and maintain data consistency.
+"""
+
+from datetime import datetime, timezone
+from unittest.mock import MagicMock, patch
+
+import pytest
+import stripe
+from prisma.enums import CreditTransactionType
+from prisma.models import CreditRefundRequest, CreditTransaction, User, UserBalance
+
+from backend.data.credit import UserCredit
+from backend.util.json import SafeJson
+from backend.util.test import SpinTestServer
+
+credit_system = UserCredit()
+
+# Test user ID for refund tests
+REFUND_TEST_USER_ID = "refund-test-user"
+
+
+async def setup_test_user_with_topup():
+    """Create a test user with initial balance and a top-up transaction."""
+    # Clean up any existing data
+    await CreditRefundRequest.prisma().delete_many(
+        where={"userId": REFUND_TEST_USER_ID}
+    )
+    await CreditTransaction.prisma().delete_many(where={"userId": REFUND_TEST_USER_ID})
+    await UserBalance.prisma().delete_many(where={"userId": REFUND_TEST_USER_ID})
+    await User.prisma().delete_many(where={"id": REFUND_TEST_USER_ID})
+
+    # Create user
+    await User.prisma().create(
+        data={
+            "id": REFUND_TEST_USER_ID,
+            "email": f"{REFUND_TEST_USER_ID}@example.com",
+            "name": "Refund Test User",
+        }
+    )
+
+    # Create user balance
+    await UserBalance.prisma().create(
+        data={
+            "userId": REFUND_TEST_USER_ID,
+            "balance": 1000,  # $10
+        }
+    )
+
+    # Create a top-up transaction that can be refunded
+    topup_tx = await CreditTransaction.prisma().create(
+        data={
+            "userId": REFUND_TEST_USER_ID,
+            "amount": 1000,
+            "type": CreditTransactionType.TOP_UP,
+            "transactionKey": "pi_test_12345",
+            "runningBalance": 1000,
+            "isActive": True,
+            "metadata": SafeJson({"stripe_payment_intent": "pi_test_12345"}),
+        }
+    )
+
+    return topup_tx
+
+
+async def cleanup_test_user():
+    """Clean up test data."""
+    await CreditRefundRequest.prisma().delete_many(
+        where={"userId": REFUND_TEST_USER_ID}
+    )
+    await CreditTransaction.prisma().delete_many(where={"userId": REFUND_TEST_USER_ID})
+    await UserBalance.prisma().delete_many(where={"userId": REFUND_TEST_USER_ID})
+    await User.prisma().delete_many(where={"id": REFUND_TEST_USER_ID})
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_deduct_credits_atomic(server: SpinTestServer):
+    """Test that deduct_credits is atomic and creates transaction correctly."""
+    topup_tx = await setup_test_user_with_topup()
+
+    try:
+        # Create a mock refund object
+        refund = MagicMock(spec=stripe.Refund)
+        refund.id = "re_test_refund_123"
+        refund.payment_intent = topup_tx.transactionKey
+        refund.amount = 500  # Refund $5 of the $10 top-up
+        refund.status = "succeeded"
+        refund.reason = "requested_by_customer"
+        refund.created = int(datetime.now(timezone.utc).timestamp())
+
+        # Create refund request record (simulating webhook flow)
+        await CreditRefundRequest.prisma().create(
+            data={
+                "userId": REFUND_TEST_USER_ID,
+                "amount": 500,
+                "transactionKey": topup_tx.transactionKey,  # Should match the original transaction
+                "reason": "Test refund",
+            }
+        )
+
+        # Call deduct_credits
+        await credit_system.deduct_credits(refund)
+
+        # Verify the user's balance was deducted
+        user_balance = await UserBalance.prisma().find_unique(
+            where={"userId": REFUND_TEST_USER_ID}
+        )
+        assert user_balance is not None
+        assert (
+            user_balance.balance == 500
+        ), f"Expected balance 500, got {user_balance.balance}"
+
+        # Verify refund transaction was created
+        refund_tx = await CreditTransaction.prisma().find_first(
+            where={
+                "userId": REFUND_TEST_USER_ID,
+                "type": CreditTransactionType.REFUND,
+                "transactionKey": refund.id,
+            }
+        )
+        assert refund_tx is not None
+        assert refund_tx.amount == -500
+        assert refund_tx.runningBalance == 500
+        assert refund_tx.isActive
+
+        # Verify refund request was updated
+        refund_request = await CreditRefundRequest.prisma().find_first(
+            where={
+                "userId": REFUND_TEST_USER_ID,
+                "transactionKey": topup_tx.transactionKey,
+            }
+        )
+        assert refund_request is not None
+        assert (
+            refund_request.result
+            == "The refund request has been approved, the amount will be credited back to your account."
+        )
+
+    finally:
+        await cleanup_test_user()
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_deduct_credits_user_not_found(server: SpinTestServer):
+    """Test that deduct_credits raises error if transaction not found (which means user doesn't exist)."""
+    # Create a mock refund object that references a non-existent payment intent
+    refund = MagicMock(spec=stripe.Refund)
+    refund.id = "re_test_refund_nonexistent"
+    refund.payment_intent = "pi_test_nonexistent"  # This payment intent doesn't exist
+    refund.amount = 500
+    refund.status = "succeeded"
+    refund.reason = "requested_by_customer"
+    refund.created = int(datetime.now(timezone.utc).timestamp())
+
+    # Should raise error for missing transaction
+    with pytest.raises(Exception):  # Should raise NotFoundError for missing transaction
+        await credit_system.deduct_credits(refund)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+@patch("backend.data.credit.settings")
+@patch("stripe.Dispute.modify")
+@patch("backend.data.credit.get_user_by_id")
+async def test_handle_dispute_with_sufficient_balance(
+    mock_get_user, mock_stripe_modify, mock_settings, server: SpinTestServer
+):
+    """Test handling dispute when user has sufficient balance (dispute gets closed)."""
+    topup_tx = await setup_test_user_with_topup()
+
+    try:
+        # Mock settings to have a low tolerance threshold
+        mock_settings.config.refund_credit_tolerance_threshold = 0
+
+        # Mock the user lookup
+        mock_user = MagicMock()
+        mock_user.email = f"{REFUND_TEST_USER_ID}@example.com"
+        mock_get_user.return_value = mock_user
+
+        # Create a mock dispute object for small amount (user has 1000, disputing 100)
+        dispute = MagicMock(spec=stripe.Dispute)
+        dispute.id = "dp_test_dispute_123"
+        dispute.payment_intent = topup_tx.transactionKey
+        dispute.amount = 100  # Small dispute amount
+        dispute.status = "pending"
+        dispute.reason = "fraudulent"
+        dispute.created = int(datetime.now(timezone.utc).timestamp())
+
+        # Mock the close method to prevent real API calls
+        dispute.close = MagicMock()
+
+        # Handle the dispute
+        await credit_system.handle_dispute(dispute)
+
+        # Verify dispute.close() was called (since user has sufficient balance)
+        dispute.close.assert_called_once()
+
+        # Verify no stripe evidence was added since dispute was closed
+        mock_stripe_modify.assert_not_called()
+
+        # Verify the user's balance was NOT deducted (dispute was closed)
+        user_balance = await UserBalance.prisma().find_unique(
+            where={"userId": REFUND_TEST_USER_ID}
+        )
+        assert user_balance is not None
+        assert (
+            user_balance.balance == 1000
+        ), f"Balance should remain 1000, got {user_balance.balance}"
+
+    finally:
+        await cleanup_test_user()
+
+
+@pytest.mark.asyncio(loop_scope="session")
+@patch("backend.data.credit.settings")
+@patch("stripe.Dispute.modify")
+@patch("backend.data.credit.get_user_by_id")
+async def test_handle_dispute_with_insufficient_balance(
+    mock_get_user, mock_stripe_modify, mock_settings, server: SpinTestServer
+):
+    """Test handling dispute when user has insufficient balance (evidence gets added)."""
+    topup_tx = await setup_test_user_with_topup()
+
+    # Save original method for restoration before any try blocks
+    original_get_history = credit_system.get_transaction_history
+
+    try:
+        # Mock settings to have a high tolerance threshold so dispute isn't closed
+        mock_settings.config.refund_credit_tolerance_threshold = 2000
+
+        # Mock the user lookup
+        mock_user = MagicMock()
+        mock_user.email = f"{REFUND_TEST_USER_ID}@example.com"
+        mock_get_user.return_value = mock_user
+
+        # Mock the transaction history method to return an async result
+        from unittest.mock import AsyncMock
+
+        mock_history = MagicMock()
+        mock_history.transactions = []
+        credit_system.get_transaction_history = AsyncMock(return_value=mock_history)
+
+        # Create a mock dispute object for full amount (user has 1000, disputing 1000)
+        dispute = MagicMock(spec=stripe.Dispute)
+        dispute.id = "dp_test_dispute_pending"
+        dispute.payment_intent = topup_tx.transactionKey
+        dispute.amount = 1000
+        dispute.status = "warning_needs_response"
+        dispute.created = int(datetime.now(timezone.utc).timestamp())
+
+        # Mock the close method to prevent real API calls
+        dispute.close = MagicMock()
+
+        # Handle the dispute (evidence should be added)
+        await credit_system.handle_dispute(dispute)
+
+        # Verify dispute.close() was NOT called (insufficient balance after tolerance)
+        dispute.close.assert_not_called()
+
+        # Verify stripe evidence was added since dispute wasn't closed
+        mock_stripe_modify.assert_called_once()
+
+        # Verify the user's balance was NOT deducted (handle_dispute doesn't deduct credits)
+        user_balance = await UserBalance.prisma().find_unique(
+            where={"userId": REFUND_TEST_USER_ID}
+        )
+        assert user_balance is not None
+        assert user_balance.balance == 1000, "Balance should remain unchanged"
+
+    finally:
+        credit_system.get_transaction_history = original_get_history
+        await cleanup_test_user()
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_refunds(server: SpinTestServer):
+    """Test that concurrent refunds are handled atomically."""
+    import asyncio
+
+    topup_tx = await setup_test_user_with_topup()
+
+    try:
+        # Create multiple refund requests
+        refund_requests = []
+        for i in range(5):
+            req = await CreditRefundRequest.prisma().create(
+                data={
+                    "userId": REFUND_TEST_USER_ID,
+                    "amount": 100,  # $1 each
+                    "transactionKey": topup_tx.transactionKey,
+                    "reason": f"Test refund {i}",
+                }
+            )
+            refund_requests.append(req)
+
+        # Create refund tasks to run concurrently
+        async def process_refund(index: int):
+            refund = MagicMock(spec=stripe.Refund)
+            refund.id = f"re_test_concurrent_{index}"
+            refund.payment_intent = topup_tx.transactionKey
+            refund.amount = 100  # $1 refund
+            refund.status = "succeeded"
+            refund.reason = "requested_by_customer"
+            refund.created = int(datetime.now(timezone.utc).timestamp())
+
+            try:
+                await credit_system.deduct_credits(refund)
+                return "success"
+            except Exception as e:
+                return f"error: {e}"
+
+        # Run refunds concurrently
+        results = await asyncio.gather(
+            *[process_refund(i) for i in range(5)], return_exceptions=True
+        )
+
+        # All should succeed
+        assert all(r == "success" for r in results), f"Some refunds failed: {results}"
+
+        # Verify final balance - with non-atomic implementation, this will demonstrate race condition
+        # EXPECTED BEHAVIOR: Due to race conditions, not all refunds will be properly processed
+        # The balance will be incorrect (higher than expected) showing lost updates
+        user_balance = await UserBalance.prisma().find_unique(
+            where={"userId": REFUND_TEST_USER_ID}
+        )
+        assert user_balance is not None
+
+        # With atomic implementation, this should be 500 (1000 - 5*100)
+        # With current non-atomic implementation, this will likely be wrong due to race conditions
+        print(f"DEBUG: Final balance = {user_balance.balance}, expected = 500")
+
+        # With atomic implementation, all 5 refunds should process correctly
+        assert (
+            user_balance.balance == 500
+        ), f"Expected balance 500 after 5 refunds of 100 each, got {user_balance.balance}"
+
+        # Verify all refund transactions exist
+        refund_txs = await CreditTransaction.prisma().find_many(
+            where={
+                "userId": REFUND_TEST_USER_ID,
+                "type": CreditTransactionType.REFUND,
+            }
+        )
+        assert (
+            len(refund_txs) == 5
+        ), f"Expected 5 refund transactions, got {len(refund_txs)}"
+
+        running_balances: set[int] = {
+            tx.runningBalance for tx in refund_txs if tx.runningBalance is not None
+        }
+
+        # Verify all balances are valid intermediate states
+        for balance in running_balances:
+            assert (
+                500 <= balance <= 1000
+            ), f"Invalid balance {balance}, should be between 500 and 1000"
+
+        # Final balance should be present
+        assert (
+            500 in running_balances
+        ), f"Final balance 500 should be in {running_balances}"
+
+        # All balances should be unique and form a valid sequence
+        sorted_balances = sorted(running_balances, reverse=True)
+        assert (
+            len(sorted_balances) == 5
+        ), f"Expected 5 unique balances, got {len(sorted_balances)}"
+
+    finally:
+        await cleanup_test_user()

autogpt_platform/backend/backend/data/credit_test.py

@@ -1,8 +1,8 @@
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 
 import pytest
 from prisma.enums import CreditTransactionType
-from prisma.models import CreditTransaction
+from prisma.models import CreditTransaction, UserBalance
 
 from backend.blocks.llm import AITextGeneratorBlock
 from backend.data.block import get_block
@@ -19,14 +19,24 @@ user_credit = BetaUserCredit(REFILL_VALUE)
 
 async def disable_test_user_transactions():
     await CreditTransaction.prisma().delete_many(where={"userId": DEFAULT_USER_ID})
+    # Also reset the balance to 0 and set updatedAt to old date to trigger monthly refill
+    old_date = datetime.now(timezone.utc) - timedelta(days=35)  # More than a month ago
+    await UserBalance.prisma().upsert(
+        where={"userId": DEFAULT_USER_ID},
+        data={
+            "create": {"userId": DEFAULT_USER_ID, "balance": 0},
+            "update": {"balance": 0, "updatedAt": old_date},
+        },
+    )
 
 
 async def top_up(amount: int):
-    await user_credit._add_transaction(
+    balance, _ = await user_credit._add_transaction(
         DEFAULT_USER_ID,
         amount,
         CreditTransactionType.TOP_UP,
     )
+    return balance
 
 
 async def spend_credits(entry: NodeExecutionEntry) -> int:
@@ -111,29 +121,90 @@ async def test_block_credit_top_up(server: SpinTestServer):
 
 @pytest.mark.asyncio(loop_scope="session")
 async def test_block_credit_reset(server: SpinTestServer):
+    """Test that BetaUserCredit provides monthly refills correctly."""
     await disable_test_user_transactions()
-    month1 = 1
-    month2 = 2
 
-    # set the calendar to month 2 but use current time from now
-    user_credit.time_now = lambda: datetime.now(timezone.utc).replace(
-        month=month2, day=1
-    )
-    month2credit = await user_credit.get_credits(DEFAULT_USER_ID)
+    # Save original time_now function for restoration
+    original_time_now = user_credit.time_now
 
-    # Month 1 result should only affect month 1
-    user_credit.time_now = lambda: datetime.now(timezone.utc).replace(
-        month=month1, day=1
-    )
-    month1credit = await user_credit.get_credits(DEFAULT_USER_ID)
-    await top_up(100)
-    assert await user_credit.get_credits(DEFAULT_USER_ID) == month1credit + 100
+    try:
+        # Test month 1 behavior
+        month1 = datetime.now(timezone.utc).replace(month=1, day=1)
+        user_credit.time_now = lambda: month1
 
-    # Month 2 balance is unaffected
-    user_credit.time_now = lambda: datetime.now(timezone.utc).replace(
-        month=month2, day=1
-    )
-    assert await user_credit.get_credits(DEFAULT_USER_ID) == month2credit
+        # First call in month 1 should trigger refill
+        balance = await user_credit.get_credits(DEFAULT_USER_ID)
+        assert balance == REFILL_VALUE  # Should get 1000 credits
+
+        # Manually create a transaction with month 1 timestamp to establish history
+        await CreditTransaction.prisma().create(
+            data={
+                "userId": DEFAULT_USER_ID,
+                "amount": 100,
+                "type": CreditTransactionType.TOP_UP,
+                "runningBalance": 1100,
+                "isActive": True,
+                "createdAt": month1,  # Set specific timestamp
+            }
+        )
+
+        # Update user balance to match
+        await UserBalance.prisma().upsert(
+            where={"userId": DEFAULT_USER_ID},
+            data={
+                "create": {"userId": DEFAULT_USER_ID, "balance": 1100},
+                "update": {"balance": 1100},
+            },
+        )
+
+        # Now test month 2 behavior
+        month2 = datetime.now(timezone.utc).replace(month=2, day=1)
+        user_credit.time_now = lambda: month2
+
+        # In month 2, since balance (1100) > refill (1000), no refill should happen
+        month2_balance = await user_credit.get_credits(DEFAULT_USER_ID)
+        assert month2_balance == 1100  # Balance persists, no reset
+
+        # Now test the refill behavior when balance is low
+        # Set balance below refill threshold
+        await UserBalance.prisma().update(
+            where={"userId": DEFAULT_USER_ID}, data={"balance": 400}
+        )
+
+        # Create a month 2 transaction to update the last transaction time
+        await CreditTransaction.prisma().create(
+            data={
+                "userId": DEFAULT_USER_ID,
+                "amount": -700,  # Spent 700 to get to 400
+                "type": CreditTransactionType.USAGE,
+                "runningBalance": 400,
+                "isActive": True,
+                "createdAt": month2,
+            }
+        )
+
+        # Move to month 3
+        month3 = datetime.now(timezone.utc).replace(month=3, day=1)
+        user_credit.time_now = lambda: month3
+
+        # Should get refilled since balance (400) < refill value (1000)
+        month3_balance = await user_credit.get_credits(DEFAULT_USER_ID)
+        assert month3_balance == REFILL_VALUE  # Should be refilled to 1000
+
+        # Verify the refill transaction was created
+        refill_tx = await CreditTransaction.prisma().find_first(
+            where={
+                "userId": DEFAULT_USER_ID,
+                "type": CreditTransactionType.GRANT,
+                "transactionKey": {"contains": "MONTHLY-CREDIT-TOP-UP"},
+            },
+            order={"createdAt": "desc"},
+        )
+        assert refill_tx is not None, "Monthly refill transaction should be created"
+        assert refill_tx.amount == 600, "Refill should be 600 (1000 - 400)"
+    finally:
+        # Restore original time_now function
+        user_credit.time_now = original_time_now
 
 
 @pytest.mark.asyncio(loop_scope="session")

autogpt_platform/backend/backend/data/credit_underflow_test.py

@@ -0,0 +1,361 @@
+"""
+Test underflow protection for cumulative refunds and negative transactions.
+
+This test ensures that when multiple large refunds are processed, the user balance
+doesn't underflow below POSTGRES_INT_MIN, which could cause integer wraparound issues.
+"""
+
+import asyncio
+from uuid import uuid4
+
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.errors import UniqueViolationError
+from prisma.models import CreditTransaction, User, UserBalance
+
+from backend.data.credit import POSTGRES_INT_MIN, UserCredit
+from backend.util.test import SpinTestServer
+
+
+async def create_test_user(user_id: str) -> None:
+    """Create a test user for underflow tests."""
+    try:
+        await User.prisma().create(
+            data={
+                "id": user_id,
+                "email": f"test-{user_id}@example.com",
+                "name": f"Test User {user_id[:8]}",
+            }
+        )
+    except UniqueViolationError:
+        # User already exists, continue
+        pass
+
+    await UserBalance.prisma().upsert(
+        where={"userId": user_id},
+        data={"create": {"userId": user_id, "balance": 0}, "update": {"balance": 0}},
+    )
+
+
+async def cleanup_test_user(user_id: str) -> None:
+    """Clean up test user and their transactions."""
+    try:
+        await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+        await UserBalance.prisma().delete_many(where={"userId": user_id})
+        await User.prisma().delete_many(where={"id": user_id})
+    except Exception as e:
+        # Log cleanup failures but don't fail the test
+        print(f"Warning: Failed to cleanup test user {user_id}: {e}")
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_debug_underflow_step_by_step(server: SpinTestServer):
+    """Debug underflow behavior step by step."""
+    credit_system = UserCredit()
+    user_id = f"debug-underflow-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        print(f"POSTGRES_INT_MIN: {POSTGRES_INT_MIN}")
+
+        # Test 1: Set up balance close to underflow threshold
+        print("\n=== Test 1: Setting up balance close to underflow threshold ===")
+        # First, manually set balance to a value very close to POSTGRES_INT_MIN
+        # We'll set it to POSTGRES_INT_MIN + 100, then try to subtract 200
+        # This should trigger underflow protection: (POSTGRES_INT_MIN + 100) + (-200) = POSTGRES_INT_MIN - 100
+        initial_balance_target = POSTGRES_INT_MIN + 100
+
+        # Use direct database update to set the balance close to underflow
+        from prisma.models import UserBalance
+
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": initial_balance_target},
+                "update": {"balance": initial_balance_target},
+            },
+        )
+
+        current_balance = await credit_system.get_credits(user_id)
+        print(f"Set balance to: {current_balance}")
+        assert current_balance == initial_balance_target
+
+        # Test 2: Apply amount that should cause underflow
+        print("\n=== Test 2: Testing underflow protection ===")
+        test_amount = (
+            -200
+        )  # This should cause underflow: (POSTGRES_INT_MIN + 100) + (-200) = POSTGRES_INT_MIN - 100
+        expected_without_protection = current_balance + test_amount
+        print(f"Current balance: {current_balance}")
+        print(f"Test amount: {test_amount}")
+        print(f"Without protection would be: {expected_without_protection}")
+        print(f"Should be clamped to POSTGRES_INT_MIN: {POSTGRES_INT_MIN}")
+
+        # Apply the amount that should trigger underflow protection
+        balance_result, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=test_amount,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,
+        )
+        print(f"Actual result: {balance_result}")
+
+        # Check if underflow protection worked
+        assert (
+            balance_result == POSTGRES_INT_MIN
+        ), f"Expected underflow protection to clamp balance to {POSTGRES_INT_MIN}, got {balance_result}"
+
+        # Test 3: Edge case - exactly at POSTGRES_INT_MIN
+        print("\n=== Test 3: Testing exact POSTGRES_INT_MIN boundary ===")
+        # Set balance to exactly POSTGRES_INT_MIN
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": POSTGRES_INT_MIN},
+                "update": {"balance": POSTGRES_INT_MIN},
+            },
+        )
+
+        edge_balance = await credit_system.get_credits(user_id)
+        print(f"Balance set to exactly POSTGRES_INT_MIN: {edge_balance}")
+
+        # Try to subtract 1 - should stay at POSTGRES_INT_MIN
+        edge_result, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=-1,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,
+        )
+        print(f"After subtracting 1: {edge_result}")
+
+        assert (
+            edge_result == POSTGRES_INT_MIN
+        ), f"Expected balance to remain clamped at {POSTGRES_INT_MIN}, got {edge_result}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_underflow_protection_large_refunds(server: SpinTestServer):
+    """Test that large cumulative refunds don't cause integer underflow."""
+    credit_system = UserCredit()
+    user_id = f"underflow-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Set up balance close to underflow threshold to test the protection
+        # Set balance to POSTGRES_INT_MIN + 1000, then try to subtract 2000
+        # This should trigger underflow protection
+        from prisma.models import UserBalance
+
+        test_balance = POSTGRES_INT_MIN + 1000
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": test_balance},
+                "update": {"balance": test_balance},
+            },
+        )
+
+        current_balance = await credit_system.get_credits(user_id)
+        assert current_balance == test_balance
+
+        # Try to deduct amount that would cause underflow: test_balance + (-2000) = POSTGRES_INT_MIN - 1000
+        underflow_amount = -2000
+        expected_without_protection = (
+            current_balance + underflow_amount
+        )  # Should be POSTGRES_INT_MIN - 1000
+
+        # Use _add_transaction directly with amount that would cause underflow
+        final_balance, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=underflow_amount,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,  # Allow going negative for refunds
+        )
+
+        # Balance should be clamped to POSTGRES_INT_MIN, not the calculated underflow value
+        assert (
+            final_balance == POSTGRES_INT_MIN
+        ), f"Balance should be clamped to {POSTGRES_INT_MIN}, got {final_balance}"
+        assert (
+            final_balance > expected_without_protection
+        ), f"Balance should be greater than underflow result {expected_without_protection}, got {final_balance}"
+
+        # Verify with get_credits too
+        stored_balance = await credit_system.get_credits(user_id)
+        assert (
+            stored_balance == POSTGRES_INT_MIN
+        ), f"Stored balance should be {POSTGRES_INT_MIN}, got {stored_balance}"
+
+        # Verify transaction was created with the underflow-protected balance
+        transactions = await CreditTransaction.prisma().find_many(
+            where={"userId": user_id, "type": CreditTransactionType.REFUND},
+            order={"createdAt": "desc"},
+        )
+        assert len(transactions) > 0, "Refund transaction should be created"
+        assert (
+            transactions[0].runningBalance == POSTGRES_INT_MIN
+        ), f"Transaction should show clamped balance {POSTGRES_INT_MIN}, got {transactions[0].runningBalance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_multiple_large_refunds_cumulative_underflow(server: SpinTestServer):
+    """Test that multiple large refunds applied sequentially don't cause underflow."""
+    credit_system = UserCredit()
+    user_id = f"cumulative-underflow-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Set up balance close to underflow threshold
+        from prisma.models import UserBalance
+
+        initial_balance = POSTGRES_INT_MIN + 500  # Close to minimum but with some room
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": initial_balance},
+                "update": {"balance": initial_balance},
+            },
+        )
+
+        # Apply multiple refunds that would cumulatively underflow
+        refund_amount = -300  # Each refund that would cause underflow when cumulative
+
+        # First refund: (POSTGRES_INT_MIN + 500) + (-300) = POSTGRES_INT_MIN + 200 (still above minimum)
+        balance_1, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=refund_amount,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,
+        )
+
+        # Should be above minimum for first refund
+        expected_balance_1 = (
+            initial_balance + refund_amount
+        )  # Should be POSTGRES_INT_MIN + 200
+        assert (
+            balance_1 == expected_balance_1
+        ), f"First refund should result in {expected_balance_1}, got {balance_1}"
+        assert (
+            balance_1 >= POSTGRES_INT_MIN
+        ), f"First refund should not go below {POSTGRES_INT_MIN}, got {balance_1}"
+
+        # Second refund: (POSTGRES_INT_MIN + 200) + (-300) = POSTGRES_INT_MIN - 100 (would underflow)
+        balance_2, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=refund_amount,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,
+        )
+
+        # Should be clamped to minimum due to underflow protection
+        assert (
+            balance_2 == POSTGRES_INT_MIN
+        ), f"Second refund should be clamped to {POSTGRES_INT_MIN}, got {balance_2}"
+
+        # Third refund: Should stay at minimum
+        balance_3, _ = await credit_system._add_transaction(
+            user_id=user_id,
+            amount=refund_amount,
+            transaction_type=CreditTransactionType.REFUND,
+            fail_insufficient_credits=False,
+        )
+
+        # Should still be at minimum
+        assert (
+            balance_3 == POSTGRES_INT_MIN
+        ), f"Third refund should stay at {POSTGRES_INT_MIN}, got {balance_3}"
+
+        # Final balance check
+        final_balance = await credit_system.get_credits(user_id)
+        assert (
+            final_balance == POSTGRES_INT_MIN
+        ), f"Final balance should be {POSTGRES_INT_MIN}, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_large_refunds_no_underflow(server: SpinTestServer):
+    """Test that concurrent large refunds don't cause race condition underflow."""
+    credit_system = UserCredit()
+    user_id = f"concurrent-underflow-test-{uuid4()}"
+    await create_test_user(user_id)
+
+    try:
+        # Set up balance close to underflow threshold
+        from prisma.models import UserBalance
+
+        initial_balance = POSTGRES_INT_MIN + 1000  # Close to minimum
+        await UserBalance.prisma().upsert(
+            where={"userId": user_id},
+            data={
+                "create": {"userId": user_id, "balance": initial_balance},
+                "update": {"balance": initial_balance},
+            },
+        )
+
+        async def large_refund(amount: int, label: str):
+            try:
+                return await credit_system._add_transaction(
+                    user_id=user_id,
+                    amount=-amount,
+                    transaction_type=CreditTransactionType.REFUND,
+                    fail_insufficient_credits=False,
+                )
+            except Exception as e:
+                return f"FAILED-{label}: {e}"
+
+        # Run concurrent refunds that would cause underflow if not protected
+        # Each refund of 500 would cause underflow: initial_balance + (-500) could go below POSTGRES_INT_MIN
+        refund_amount = 500
+        results = await asyncio.gather(
+            large_refund(refund_amount, "A"),
+            large_refund(refund_amount, "B"),
+            large_refund(refund_amount, "C"),
+            return_exceptions=True,
+        )
+
+        # Check all results are valid and no underflow occurred
+        valid_results = []
+        for i, result in enumerate(results):
+            if isinstance(result, tuple):
+                balance, _ = result
+                assert (
+                    balance >= POSTGRES_INT_MIN
+                ), f"Result {i} balance {balance} underflowed below {POSTGRES_INT_MIN}"
+                valid_results.append(balance)
+            elif isinstance(result, str) and "FAILED" in result:
+                # Some operations might fail due to validation, that's okay
+                pass
+            else:
+                # Unexpected exception
+                assert not isinstance(
+                    result, Exception
+                ), f"Unexpected exception in result {i}: {result}"
+
+        # At least one operation should succeed
+        assert (
+            len(valid_results) > 0
+        ), f"At least one refund should succeed, got results: {results}"
+
+        # All successful results should be >= POSTGRES_INT_MIN
+        for balance in valid_results:
+            assert (
+                balance >= POSTGRES_INT_MIN
+            ), f"Balance {balance} should not be below {POSTGRES_INT_MIN}"
+
+        # Final balance should be valid and at or above POSTGRES_INT_MIN
+        final_balance = await credit_system.get_credits(user_id)
+        assert (
+            final_balance >= POSTGRES_INT_MIN
+        ), f"Final balance {final_balance} should not underflow below {POSTGRES_INT_MIN}"
+
+    finally:
+        await cleanup_test_user(user_id)

autogpt_platform/backend/backend/data/credit_user_balance_migration_test.py

@@ -0,0 +1,217 @@
+"""
+Integration test to verify complete migration from User.balance to UserBalance table.
+
+This test ensures that:
+1. No User.balance queries exist in the system
+2. All balance operations go through UserBalance table
+3. User and UserBalance stay synchronized properly
+"""
+
+import asyncio
+from datetime import datetime
+
+import pytest
+from prisma.enums import CreditTransactionType
+from prisma.errors import UniqueViolationError
+from prisma.models import CreditTransaction, User, UserBalance
+
+from backend.data.credit import UsageTransactionMetadata, UserCredit
+from backend.util.json import SafeJson
+from backend.util.test import SpinTestServer
+
+
+async def create_test_user(user_id: str) -> None:
+    """Create a test user for migration tests."""
+    try:
+        await User.prisma().create(
+            data={
+                "id": user_id,
+                "email": f"test-{user_id}@example.com",
+                "name": f"Test User {user_id[:8]}",
+            }
+        )
+    except UniqueViolationError:
+        # User already exists, continue
+        pass
+
+
+async def cleanup_test_user(user_id: str) -> None:
+    """Clean up test user and their data."""
+    try:
+        await CreditTransaction.prisma().delete_many(where={"userId": user_id})
+        await UserBalance.prisma().delete_many(where={"userId": user_id})
+        await User.prisma().delete_many(where={"id": user_id})
+    except Exception as e:
+        # Log cleanup failures but don't fail the test
+        print(f"Warning: Failed to cleanup test user {user_id}: {e}")
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_user_balance_migration_complete(server: SpinTestServer):
+    """Test that User table balance is never used and UserBalance is source of truth."""
+    credit_system = UserCredit()
+    user_id = f"migration-test-{datetime.now().timestamp()}"
+    await create_test_user(user_id)
+
+    try:
+        # 1. Verify User table does NOT have balance set initially
+        user = await User.prisma().find_unique(where={"id": user_id})
+        assert user is not None
+        # User.balance should not exist or should be None/0 if it exists
+        user_balance_attr = getattr(user, "balance", None)
+        if user_balance_attr is not None:
+            assert (
+                user_balance_attr == 0 or user_balance_attr is None
+            ), f"User.balance should be 0 or None, got {user_balance_attr}"
+
+        # 2. Perform various credit operations using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=1000,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "migration_test"}),
+        )
+        balance1 = await credit_system.get_credits(user_id)
+        assert balance1 == 1000
+
+        await credit_system.spend_credits(
+            user_id,
+            300,
+            UsageTransactionMetadata(
+                graph_exec_id="test", reason="Migration test spend"
+            ),
+        )
+        balance2 = await credit_system.get_credits(user_id)
+        assert balance2 == 700
+
+        # 3. Verify UserBalance table has correct values
+        user_balance = await UserBalance.prisma().find_unique(where={"userId": user_id})
+        assert user_balance is not None
+        assert (
+            user_balance.balance == 700
+        ), f"UserBalance should be 700, got {user_balance.balance}"
+
+        # 4. CRITICAL: Verify User.balance is NEVER updated during operations
+        user_after = await User.prisma().find_unique(where={"id": user_id})
+        assert user_after is not None
+        user_balance_after = getattr(user_after, "balance", None)
+        if user_balance_after is not None:
+            # If User.balance exists, it should still be 0 (never updated)
+            assert (
+                user_balance_after == 0 or user_balance_after is None
+            ), f"User.balance should remain 0/None after operations, got {user_balance_after}. This indicates User.balance is still being used!"
+
+        # 5. Verify get_credits always returns UserBalance value, not User.balance
+        final_balance = await credit_system.get_credits(user_id)
+        assert (
+            final_balance == user_balance.balance
+        ), f"get_credits should return UserBalance value {user_balance.balance}, got {final_balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_detect_stale_user_balance_queries(server: SpinTestServer):
+    """Test to detect if any operations are still using User.balance instead of UserBalance."""
+    credit_system = UserCredit()
+    user_id = f"stale-query-test-{datetime.now().timestamp()}"
+    await create_test_user(user_id)
+
+    try:
+        # Create UserBalance with specific value
+        await UserBalance.prisma().create(
+            data={"userId": user_id, "balance": 5000}  # $50
+        )
+
+        # Verify that get_credits returns UserBalance value (5000), not any stale User.balance value
+        balance = await credit_system.get_credits(user_id)
+        assert (
+            balance == 5000
+        ), f"Expected get_credits to return 5000 from UserBalance, got {balance}"
+
+        # Verify all operations use UserBalance using internal method (bypasses Stripe)
+        await credit_system._add_transaction(
+            user_id=user_id,
+            amount=1000,
+            transaction_type=CreditTransactionType.TOP_UP,
+            metadata=SafeJson({"test": "final_verification"}),
+        )
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 6000, f"Expected 6000, got {final_balance}"
+
+        # Verify UserBalance table has the correct value
+        user_balance = await UserBalance.prisma().find_unique(where={"userId": user_id})
+        assert user_balance is not None
+        assert (
+            user_balance.balance == 6000
+        ), f"UserBalance should be 6000, got {user_balance.balance}"
+
+    finally:
+        await cleanup_test_user(user_id)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_concurrent_operations_use_userbalance_only(server: SpinTestServer):
+    """Test that concurrent operations all use UserBalance locking, not User.balance."""
+    credit_system = UserCredit()
+    user_id = f"concurrent-userbalance-test-{datetime.now().timestamp()}"
+    await create_test_user(user_id)
+
+    try:
+        # Set initial balance in UserBalance
+        await UserBalance.prisma().create(data={"userId": user_id, "balance": 1000})
+
+        # Run concurrent operations to ensure they all use UserBalance atomic operations
+        async def concurrent_spend(amount: int, label: str):
+            try:
+                await credit_system.spend_credits(
+                    user_id,
+                    amount,
+                    UsageTransactionMetadata(
+                        graph_exec_id=f"concurrent-{label}",
+                        reason=f"Concurrent test {label}",
+                    ),
+                )
+                return f"{label}-SUCCESS"
+            except Exception as e:
+                return f"{label}-FAILED: {e}"
+
+        # Run concurrent operations
+        results = await asyncio.gather(
+            concurrent_spend(100, "A"),
+            concurrent_spend(200, "B"),
+            concurrent_spend(300, "C"),
+            return_exceptions=True,
+        )
+
+        # All should succeed (1000 >= 100+200+300)
+        successful = [r for r in results if "SUCCESS" in str(r)]
+        assert len(successful) == 3, f"All operations should succeed, got {results}"
+
+        # Final balance should be 1000 - 600 = 400
+        final_balance = await credit_system.get_credits(user_id)
+        assert final_balance == 400, f"Expected final balance 400, got {final_balance}"
+
+        # Verify UserBalance has correct value
+        user_balance = await UserBalance.prisma().find_unique(where={"userId": user_id})
+        assert user_balance is not None
+        assert (
+            user_balance.balance == 400
+        ), f"UserBalance should be 400, got {user_balance.balance}"
+
+        # Critical: If User.balance exists and was used, it might have wrong value
+        try:
+            user = await User.prisma().find_unique(where={"id": user_id})
+            user_balance_attr = getattr(user, "balance", None)
+            if user_balance_attr is not None:
+                # If User.balance exists, it should NOT be used for operations
+                # The fact that our final balance is correct from UserBalance proves the system is working
+                print(
+                    f"✅ User.balance exists ({user_balance_attr}) but UserBalance ({user_balance.balance}) is being used correctly"
+                )
+        except Exception:
+            print("✅ User.balance column doesn't exist - migration is complete")
+
+    finally:
+        await cleanup_test_user(user_id)

autogpt_platform/backend/backend/data/db.py

@@ -98,42 +98,6 @@ async def transaction(timeout: int = TRANSACTION_TIMEOUT):
         yield tx
 
 
-@asynccontextmanager
-async def locked_transaction(key: str, timeout: int = TRANSACTION_TIMEOUT):
-    """
-    Create a transaction and take a per-key advisory *transaction* lock.
-
-    - Uses a 64-bit lock id via hashtextextended(key, 0) to avoid 32-bit collisions.
-    - Bound by lock_timeout and statement_timeout so it won't block indefinitely.
-    - Lock is held for the duration of the transaction and auto-released on commit/rollback.
-
-    Args:
-        key: String lock key (e.g., "usr_trx_<uuid>").
-        timeout: Transaction/lock/statement timeout in milliseconds.
-    """
-    async with transaction(timeout=timeout) as tx:
-        # Ensure we don't wait longer than desired
-        # Note: SET LOCAL doesn't support parameterized queries, must use string interpolation
-        await tx.execute_raw(f"SET LOCAL statement_timeout = '{int(timeout)}ms'")  # type: ignore[arg-type]
-        await tx.execute_raw(f"SET LOCAL lock_timeout = '{int(timeout)}ms'")  # type: ignore[arg-type]
-
-        # Block until acquired or lock_timeout hits
-        try:
-            await tx.execute_raw(
-                "SELECT pg_advisory_xact_lock(hashtextextended($1, 0))",
-                key,
-            )
-        except Exception as e:
-            # Normalize PG's lock timeout error to TimeoutError for callers
-            if "lock timeout" in str(e).lower():
-                raise TimeoutError(
-                    f"Could not acquire lock for key={key!r} within {timeout}ms"
-                ) from e
-            raise
-
-        yield tx
-
-
 def get_database_schema() -> str:
     """Extract database schema from DATABASE_URL."""
     parsed_url = urlparse(DATABASE_URL)

autogpt_platform/backend/backend/data/execution.py

@@ -38,8 +38,8 @@ from prisma.types import (
 from pydantic import BaseModel, ConfigDict, JsonValue, ValidationError
 from pydantic.fields import Field
 
-from backend.server.v2.store.exceptions import DatabaseError
 from backend.util import type as type_utils
+from backend.util.exceptions import DatabaseError
 from backend.util.json import SafeJson
 from backend.util.models import Pagination
 from backend.util.retry import func_retry
@@ -478,6 +478,48 @@ async def get_graph_executions(
     return [GraphExecutionMeta.from_db(execution) for execution in executions]
 
 
+async def get_graph_executions_count(
+    user_id: Optional[str] = None,
+    graph_id: Optional[str] = None,
+    statuses: Optional[list[ExecutionStatus]] = None,
+    created_time_gte: Optional[datetime] = None,
+    created_time_lte: Optional[datetime] = None,
+) -> int:
+    """
+    Get count of graph executions with optional filters.
+
+    Args:
+        user_id: Optional user ID to filter by
+        graph_id: Optional graph ID to filter by
+        statuses: Optional list of execution statuses to filter by
+        created_time_gte: Optional minimum creation time
+        created_time_lte: Optional maximum creation time
+
+    Returns:
+        Count of matching graph executions
+    """
+    where_filter: AgentGraphExecutionWhereInput = {
+        "isDeleted": False,
+    }
+
+    if user_id:
+        where_filter["userId"] = user_id
+
+    if graph_id:
+        where_filter["agentGraphId"] = graph_id
+
+    if created_time_gte or created_time_lte:
+        where_filter["createdAt"] = {
+            "gte": created_time_gte or datetime.min.replace(tzinfo=timezone.utc),
+            "lte": created_time_lte or datetime.max.replace(tzinfo=timezone.utc),
+        }
+    if statuses:
+        where_filter["OR"] = [{"executionStatus": status} for status in statuses]
+
+    count = await AgentGraphExecution.prisma().count(where=where_filter)
+    return count
+
+
 class GraphExecutionsPaginated(BaseModel):
     """Response schema for paginated graph executions."""
 

autogpt_platform/backend/backend/data/generate_data.py

@@ -7,7 +7,7 @@ from prisma.enums import AgentExecutionStatus
 from backend.data.execution import get_graph_executions
 from backend.data.graph import get_graph_metadata
 from backend.data.model import UserExecutionSummaryStats
-from backend.server.v2.store.exceptions import DatabaseError
+from backend.util.exceptions import DatabaseError
 from backend.util.logging import TruncatedLogger
 
 logger = TruncatedLogger(logging.getLogger(__name__), prefix="[SummaryData]")

autogpt_platform/backend/backend/data/graph.py

@@ -129,17 +129,20 @@ class NodeModel(Node):
         Returns a copy of the node model, stripped of any non-transferable properties
         """
         stripped_node = self.model_copy(deep=True)
-        # Remove credentials from node input
+
+        # Remove credentials and other (possible) secrets from node input
         if stripped_node.input_default:
             stripped_node.input_default = NodeModel._filter_secrets_from_node_input(
                 stripped_node.input_default, self.block.input_schema.jsonschema()
             )
 
+        # Remove default secret value from secret input nodes
         if (
             stripped_node.block.block_type == BlockType.INPUT
+            and stripped_node.input_default.get("secret", False) is True
             and "value" in stripped_node.input_default
         ):
-            stripped_node.input_default["value"] = ""
+            del stripped_node.input_default["value"]
 
         # Remove webhook info
         stripped_node.webhook_id = None
@@ -156,8 +159,10 @@ class NodeModel(Node):
         result = {}
         for key, value in input_data.items():
             field_schema: dict | None = field_schemas.get(key)
-            if (field_schema and field_schema.get("secret", False)) or any(
-                sensitive_key in key.lower() for sensitive_key in sensitive_keys
+            if (field_schema and field_schema.get("secret", False)) or (
+                any(sensitive_key in key.lower() for sensitive_key in sensitive_keys)
+                # Prevent removing `secret` flag on input nodes
+                and type(value) is not bool
             ):
                 # This is a secret value -> filter this key-value pair out
                 continue

autogpt_platform/backend/backend/data/graph_test.py

@@ -201,25 +201,56 @@ async def test_get_input_schema(server: SpinTestServer, snapshot: Snapshot):
 @pytest.mark.asyncio(loop_scope="session")
 async def test_clean_graph(server: SpinTestServer):
     """
-    Test the clean_graph function that:
-    1. Clears input block values
-    2. Removes credentials from nodes
+    Test the stripped_for_export function that:
+    1. Removes sensitive/secret fields from node inputs
+    2. Removes webhook information
+    3. Preserves non-sensitive data including input block values
     """
-    # Create a graph with input blocks and credentials
+    # Create a graph with input blocks containing both sensitive and normal data
     graph = Graph(
         id="test_clean_graph",
         name="Test Clean Graph",
         description="Test graph cleaning",
         nodes=[
             Node(
-                id="input_node",
                 block_id=AgentInputBlock().id,
                 input_default={
+                    "_test_id": "input_node",
                     "name": "test_input",
-                    "value": "test value",
+                    "value": "test value",  # This should be preserved
                     "description": "Test input description",
                 },
             ),
+            Node(
+                block_id=AgentInputBlock().id,
+                input_default={
+                    "_test_id": "input_node_secret",
+                    "name": "secret_input",
+                    "value": "another value",
+                    "secret": True,  # This makes the input secret
+                },
+            ),
+            Node(
+                block_id=StoreValueBlock().id,
+                input_default={
+                    "_test_id": "node_with_secrets",
+                    "input": "normal_value",
+                    "control_test_input": "should be preserved",
+                    "api_key": "secret_api_key_123",  # Should be filtered
+                    "password": "secret_password_456",  # Should be filtered
+                    "token": "secret_token_789",  # Should be filtered
+                    "credentials": {  # Should be filtered
+                        "id": "fake-github-credentials-id",
+                        "provider": "github",
+                        "type": "api_key",
+                    },
+                    "anthropic_credentials": {  # Should be filtered
+                        "id": "fake-anthropic-credentials-id",
+                        "provider": "anthropic",
+                        "type": "api_key",
+                    },
+                },
+            ),
         ],
         links=[],
     )
@@ -231,15 +262,54 @@ async def test_clean_graph(server: SpinTestServer):
     )
 
     # Clean the graph
-    created_graph = await server.agent_server.test_get_graph(
+    cleaned_graph = await server.agent_server.test_get_graph(
         created_graph.id, created_graph.version, DEFAULT_USER_ID, for_export=True
     )
 
-    # # Verify input block value is cleared
+    # Verify sensitive fields are removed but normal fields are preserved
     input_node = next(
-        n for n in created_graph.nodes if n.block_id == AgentInputBlock().id
+        n for n in cleaned_graph.nodes if n.input_default["_test_id"] == "input_node"
     )
-    assert input_node.input_default["value"] == ""
+
+    # Non-sensitive fields should be preserved
+    assert input_node.input_default["name"] == "test_input"
+    assert input_node.input_default["value"] == "test value"  # Should be preserved now
+    assert input_node.input_default["description"] == "Test input description"
+
+    # Sensitive fields should be filtered out
+    assert "api_key" not in input_node.input_default
+    assert "password" not in input_node.input_default
+
+    # Verify secret input node preserves non-sensitive fields but removes secret value
+    secret_node = next(
+        n
+        for n in cleaned_graph.nodes
+        if n.input_default["_test_id"] == "input_node_secret"
+    )
+    assert secret_node.input_default["name"] == "secret_input"
+    assert "value" not in secret_node.input_default  # Secret default should be removed
+    assert secret_node.input_default["secret"] is True
+
+    # Verify sensitive fields are filtered from nodes with secrets
+    secrets_node = next(
+        n
+        for n in cleaned_graph.nodes
+        if n.input_default["_test_id"] == "node_with_secrets"
+    )
+    # Normal fields should be preserved
+    assert secrets_node.input_default["input"] == "normal_value"
+    assert secrets_node.input_default["control_test_input"] == "should be preserved"
+    # Sensitive fields should be filtered out
+    assert "api_key" not in secrets_node.input_default
+    assert "password" not in secrets_node.input_default
+    assert "token" not in secrets_node.input_default
+    assert "credentials" not in secrets_node.input_default
+    assert "anthropic_credentials" not in secrets_node.input_default
+
+    # Verify webhook info is removed (if any nodes had it)
+    for node in cleaned_graph.nodes:
+        assert node.webhook_id is None
+        assert node.webhook is None
 
 
 @pytest.mark.asyncio(loop_scope="session")

autogpt_platform/backend/backend/data/model.py

@@ -347,6 +347,9 @@ class APIKeyCredentials(_BaseCredentials):
     """Unix timestamp (seconds) indicating when the API key expires (if at all)"""
 
     def auth_header(self) -> str:
+        # Linear API keys should not have Bearer prefix
+        if self.provider == "linear":
+            return self.api_key.get_secret_value()
         return f"Bearer {self.api_key.get_secret_value()}"
 
 

autogpt_platform/backend/backend/data/notifications.py

@@ -15,7 +15,7 @@ from prisma.types import (
 # from backend.notifications.models import NotificationEvent
 from pydantic import BaseModel, ConfigDict, EmailStr, Field, field_validator
 
-from backend.server.v2.store.exceptions import DatabaseError
+from backend.util.exceptions import DatabaseError
 from backend.util.json import SafeJson
 from backend.util.logging import TruncatedLogger
 
@@ -235,6 +235,7 @@ class BaseEventModel(BaseModel):
 
 
 class NotificationEventModel(BaseEventModel, Generic[NotificationDataType_co]):
+    id: Optional[str] = None  # None when creating, populated when reading from DB
     data: NotificationDataType_co
 
     @property
@@ -378,6 +379,7 @@ class NotificationPreference(BaseModel):
 
 
 class UserNotificationEventDTO(BaseModel):
+    id: str  # Added to track notifications for removal
     type: NotificationType
     data: dict
     created_at: datetime
@@ -386,6 +388,7 @@ class UserNotificationEventDTO(BaseModel):
     @staticmethod
     def from_db(model: NotificationEvent) -> "UserNotificationEventDTO":
         return UserNotificationEventDTO(
+            id=model.id,
             type=model.type,
             data=dict(model.data),
             created_at=model.createdAt,
@@ -541,6 +544,79 @@ async def empty_user_notification_batch(
         ) from e
 
 
+async def clear_all_user_notification_batches(user_id: str) -> None:
+    """Clear ALL notification batches for a user across all types.
+
+    Used when user's email is bounced/inactive and we should stop
+    trying to send them ANY emails.
+    """
+    try:
+        # Delete all notification events for this user
+        await NotificationEvent.prisma().delete_many(
+            where={"UserNotificationBatch": {"is": {"userId": user_id}}}
+        )
+
+        # Delete all batches for this user
+        await UserNotificationBatch.prisma().delete_many(where={"userId": user_id})
+
+        logger.info(f"Cleared all notification batches for user {user_id}")
+    except Exception as e:
+        raise DatabaseError(
+            f"Failed to clear all notification batches for user {user_id}: {e}"
+        ) from e
+
+
+async def remove_notifications_from_batch(
+    user_id: str, notification_type: NotificationType, notification_ids: list[str]
+) -> None:
+    """Remove specific notifications from a user's batch by their IDs.
+
+    This is used after successful sending to remove only the
+    sent notifications, preventing duplicates on retry.
+    """
+    if not notification_ids:
+        return
+
+    try:
+        # Delete the specific notification events
+        deleted_count = await NotificationEvent.prisma().delete_many(
+            where={
+                "id": {"in": notification_ids},
+                "UserNotificationBatch": {
+                    "is": {"userId": user_id, "type": notification_type}
+                },
+            }
+        )
+
+        logger.info(
+            f"Removed {deleted_count} notifications from batch for user {user_id}"
+        )
+
+        # Check if batch is now empty and delete it if so
+        remaining = await NotificationEvent.prisma().count(
+            where={
+                "UserNotificationBatch": {
+                    "is": {"userId": user_id, "type": notification_type}
+                }
+            }
+        )
+
+        if remaining == 0:
+            await UserNotificationBatch.prisma().delete_many(
+                where=UserNotificationBatchWhereInput(
+                    userId=user_id,
+                    type=notification_type,
+                )
+            )
+            logger.info(
+                f"Deleted empty batch for user {user_id} and type {notification_type}"
+            )
+    except Exception as e:
+        raise DatabaseError(
+            f"Failed to remove notifications from batch for user {user_id} and type {notification_type}: {e}"
+        ) from e
+
+
 async def get_user_notification_batch(
     user_id: str,
     notification_type: NotificationType,

autogpt_platform/backend/backend/data/onboarding.py

@@ -4,16 +4,15 @@ from typing import Any, Optional
 
 import prisma
 import pydantic
-from autogpt_libs.utils.cache import cached
 from prisma.enums import OnboardingStep
 from prisma.models import UserOnboarding
 from prisma.types import UserOnboardingCreateInput, UserOnboardingUpdateInput
 
 from backend.data.block import get_blocks
 from backend.data.credit import get_user_credit_model
-from backend.data.graph import GraphModel
 from backend.data.model import CredentialsMetaInput
 from backend.server.v2.store.model import StoreAgentDetails
+from backend.util.cache import cached
 from backend.util.json import SafeJson
 
 # Mapping from user reason id to categories to search for when choosing agent to show
@@ -27,8 +26,6 @@ REASON_MAPPING: dict[str, list[str]] = {
 POINTS_AGENT_COUNT = 50  # Number of agents to calculate points for
 MIN_AGENT_COUNT = 2  # Minimum number of marketplace agents to enable onboarding
 
-user_credit = get_user_credit_model()
-
 
 class UserOnboardingUpdate(pydantic.BaseModel):
     completedSteps: Optional[list[OnboardingStep]] = None
@@ -148,7 +145,8 @@ async def reward_user(user_id: str, step: OnboardingStep):
         return
 
     onboarding.rewardedFor.append(step)
-    await user_credit.onboarding_reward(user_id, reward, step)
+    user_credit_model = await get_user_credit_model(user_id)
+    await user_credit_model.onboarding_reward(user_id, reward, step)
     await UserOnboarding.prisma().update(
         where={"userId": user_id},
         data={
@@ -278,8 +276,14 @@ async def get_recommended_agents(user_id: str) -> list[StoreAgentDetails]:
         for word in user_onboarding.integrations
     ]
 
+    where_clause["is_available"] = True
+
+    # Try to take only agents that are available and allowed for onboarding
     storeAgents = await prisma.models.StoreAgent.prisma().find_many(
-        where=prisma.types.StoreAgentWhereInput(**where_clause),
+        where={
+            "is_available": True,
+            "useForOnboarding": True,
+        },
         order=[
             {"featured": "desc"},
             {"runs": "desc"},
@@ -288,59 +292,16 @@ async def get_recommended_agents(user_id: str) -> list[StoreAgentDetails]:
         take=100,
     )
 
-    agentListings = await prisma.models.StoreListingVersion.prisma().find_many(
-        where={
-            "id": {"in": [agent.storeListingVersionId for agent in storeAgents]},
-        },
-        include={"AgentGraph": True},
-    )
-
-    for listing in agentListings:
-        agent = listing.AgentGraph
-        if agent is None:
-            continue
-        graph = GraphModel.from_db(agent)
-        # Remove agents with empty input schema
-        if not graph.input_schema:
-            storeAgents = [
-                a for a in storeAgents if a.storeListingVersionId != listing.id
-            ]
-            continue
-
-        # Remove agents with empty credentials
-        # Get nodes from this agent that have credentials
-        nodes = await prisma.models.AgentNode.prisma().find_many(
-            where={
-                "agentGraphId": agent.id,
-                "agentBlockId": {"in": list(CREDENTIALS_FIELDS.keys())},
-            },
-        )
-        for node in nodes:
-            block_id = node.agentBlockId
-            field_name = CREDENTIALS_FIELDS[block_id]
-            # If there are no credentials or they are empty, remove the agent
-            # FIXME ignores default values
-            if (
-                field_name not in node.constantInput
-                or node.constantInput[field_name] is None
-            ):
-                storeAgents = [
-                    a for a in storeAgents if a.storeListingVersionId != listing.id
-                ]
-                break
-
-    # If there are less than 2 agents, add more agents to the list
+    # If not enough agents found, relax the useForOnboarding filter
     if len(storeAgents) < 2:
-        storeAgents += await prisma.models.StoreAgent.prisma().find_many(
-            where={
-                "listing_id": {"not_in": [agent.listing_id for agent in storeAgents]},
-            },
+        storeAgents = await prisma.models.StoreAgent.prisma().find_many(
+            where=prisma.types.StoreAgentWhereInput(**where_clause),
             order=[
                 {"featured": "desc"},
                 {"runs": "desc"},
                 {"rating": "desc"},
             ],
-            take=2 - len(storeAgents),
+            take=100,
         )
 
     # Calculate points for the first X agents and choose the top 2

autogpt_platform/backend/backend/data/partial_types.py

@@ -0,0 +1,5 @@
+import prisma.models
+
+
+class StoreAgentWithRank(prisma.models.StoreAgent):
+    rank: float

autogpt_platform/backend/backend/data/redis_client.py

@@ -1,11 +1,11 @@
 import logging
 import os
 
-from autogpt_libs.utils.cache import cached, thread_cached
 from dotenv import load_dotenv
 from redis import Redis
 from redis.asyncio import Redis as AsyncRedis
 
+from backend.util.cache import cached, thread_cached
 from backend.util.retry import conn_retry
 
 load_dotenv()
@@ -34,7 +34,7 @@ def disconnect():
     get_redis().close()
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def get_redis() -> Redis:
     return connect()
 

autogpt_platform/backend/backend/data/user.py

@@ -7,7 +7,6 @@ from typing import Optional, cast
 from urllib.parse import quote_plus
 
 from autogpt_libs.auth.models import DEFAULT_USER_ID
-from autogpt_libs.utils.cache import cached
 from fastapi import HTTPException
 from prisma.enums import NotificationType
 from prisma.models import User as PrismaUser
@@ -16,8 +15,9 @@ from prisma.types import JsonFilter, UserCreateInput, UserUpdateInput
 from backend.data.db import prisma
 from backend.data.model import User, UserIntegrations, UserMetadata
 from backend.data.notifications import NotificationPreference, NotificationPreferenceDTO
-from backend.server.v2.store.exceptions import DatabaseError
+from backend.util.cache import cached
 from backend.util.encryption import JSONCryptor
+from backend.util.exceptions import DatabaseError
 from backend.util.json import SafeJson
 from backend.util.settings import Settings
 
@@ -354,6 +354,36 @@ async def set_user_email_verification(user_id: str, verified: bool) -> None:
         ) from e
 
 
+async def disable_all_user_notifications(user_id: str) -> None:
+    """Disable all notification preferences for a user.
+
+    Used when user's email bounces/is inactive to prevent any future notifications.
+    """
+    try:
+        await PrismaUser.prisma().update(
+            where={"id": user_id},
+            data={
+                "notifyOnAgentRun": False,
+                "notifyOnZeroBalance": False,
+                "notifyOnLowBalance": False,
+                "notifyOnBlockExecutionFailed": False,
+                "notifyOnContinuousAgentError": False,
+                "notifyOnDailySummary": False,
+                "notifyOnWeeklySummary": False,
+                "notifyOnMonthlySummary": False,
+                "notifyOnAgentApproved": False,
+                "notifyOnAgentRejected": False,
+            },
+        )
+        # Invalidate cache for this user
+        get_user_by_id.cache_delete(user_id)
+        logger.info(f"Disabled all notification preferences for user {user_id}")
+    except Exception as e:
+        raise DatabaseError(
+            f"Failed to disable notifications for user {user_id}: {e}"
+        ) from e
+
+
 async def get_user_email_verification(user_id: str) -> bool:
     """Get the email verification status for a user."""
     try:

autogpt_platform/backend/backend/executor/database.py

@@ -1,5 +1,6 @@
 import logging
-from typing import Callable, Concatenate, ParamSpec, TypeVar, cast
+from contextlib import asynccontextmanager
+from typing import TYPE_CHECKING, Callable, Concatenate, ParamSpec, TypeVar, cast
 
 from backend.data import db
 from backend.data.credit import UsageTransactionMetadata, get_user_credit_model
@@ -9,6 +10,7 @@ from backend.data.execution import (
     get_execution_kv_data,
     get_graph_execution_meta,
     get_graph_executions,
+    get_graph_executions_count,
     get_latest_node_execution,
     get_node_execution,
     get_node_executions,
@@ -28,14 +30,17 @@ from backend.data.graph import (
     get_node,
 )
 from backend.data.notifications import (
+    clear_all_user_notification_batches,
     create_or_add_to_user_notification_batch,
     empty_user_notification_batch,
     get_all_batches_by_type,
     get_user_notification_batch,
     get_user_notification_oldest_message_in_batch,
+    remove_notifications_from_batch,
 )
 from backend.data.user import (
     get_active_user_ids_in_timerange,
+    get_user_by_id,
     get_user_email_by_id,
     get_user_email_verification,
     get_user_integrations,
@@ -53,8 +58,10 @@ from backend.util.service import (
 )
 from backend.util.settings import Config
 
+if TYPE_CHECKING:
+    from fastapi import FastAPI
+
 config = Config()
-_user_credit_model = get_user_credit_model()
 logger = logging.getLogger(__name__)
 P = ParamSpec("P")
 R = TypeVar("R")
@@ -63,24 +70,27 @@ R = TypeVar("R")
 async def _spend_credits(
     user_id: str, cost: int, metadata: UsageTransactionMetadata
 ) -> int:
-    return await _user_credit_model.spend_credits(user_id, cost, metadata)
+    user_credit_model = await get_user_credit_model(user_id)
+    return await user_credit_model.spend_credits(user_id, cost, metadata)
 
 
 async def _get_credits(user_id: str) -> int:
-    return await _user_credit_model.get_credits(user_id)
+    user_credit_model = await get_user_credit_model(user_id)
+    return await user_credit_model.get_credits(user_id)
 
 
 class DatabaseManager(AppService):
+    @asynccontextmanager
+    async def lifespan(self, app: "FastAPI"):
+        async with super().lifespan(app):
+            logger.info(f"[{self.service_name}] ⏳ Connecting to Database...")
+            await db.connect()
 
-    def run_service(self) -> None:
-        logger.info(f"[{self.service_name}] ⏳ Connecting to Database...")
-        self.run_and_wait(db.connect())
-        super().run_service()
+            logger.info(f"[{self.service_name}] ✅ Ready")
+            yield
 
-    def cleanup(self):
-        super().cleanup()
-        logger.info(f"[{self.service_name}] ⏳ Disconnecting Database...")
-        self.run_and_wait(db.disconnect())
+            logger.info(f"[{self.service_name}] ⏳ Disconnecting Database...")
+            await db.disconnect()
 
     async def health_check(self) -> str:
         if not db.is_connected():
@@ -111,6 +121,7 @@ class DatabaseManager(AppService):
 
     # Executions
     get_graph_executions = _(get_graph_executions)
+    get_graph_executions_count = _(get_graph_executions_count)
     get_graph_execution_meta = _(get_graph_execution_meta)
     create_graph_execution = _(create_graph_execution)
     get_node_execution = _(get_node_execution)
@@ -142,15 +153,18 @@ class DatabaseManager(AppService):
 
     # User Comms - async
     get_active_user_ids_in_timerange = _(get_active_user_ids_in_timerange)
+    get_user_by_id = _(get_user_by_id)
     get_user_email_by_id = _(get_user_email_by_id)
     get_user_email_verification = _(get_user_email_verification)
     get_user_notification_preference = _(get_user_notification_preference)
 
     # Notifications - async
+    clear_all_user_notification_batches = _(clear_all_user_notification_batches)
     create_or_add_to_user_notification_batch = _(
         create_or_add_to_user_notification_batch
     )
     empty_user_notification_batch = _(empty_user_notification_batch)
+    remove_notifications_from_batch = _(remove_notifications_from_batch)
     get_all_batches_by_type = _(get_all_batches_by_type)
     get_user_notification_batch = _(get_user_notification_batch)
     get_user_notification_oldest_message_in_batch = _(
@@ -179,6 +193,7 @@ class DatabaseManagerClient(AppServiceClient):
 
     # Executions
     get_graph_executions = _(d.get_graph_executions)
+    get_graph_executions_count = _(d.get_graph_executions_count)
     get_graph_execution_meta = _(d.get_graph_execution_meta)
     get_node_executions = _(d.get_node_executions)
     update_node_execution_status = _(d.update_node_execution_status)
@@ -224,6 +239,7 @@ class DatabaseManagerAsyncClient(AppServiceClient):
     get_node = d.get_node
     get_node_execution = d.get_node_execution
     get_node_executions = d.get_node_executions
+    get_user_by_id = d.get_user_by_id
     get_user_integrations = d.get_user_integrations
     upsert_execution_input = d.upsert_execution_input
     upsert_execution_output = d.upsert_execution_output
@@ -241,10 +257,12 @@ class DatabaseManagerAsyncClient(AppServiceClient):
     get_user_notification_preference = d.get_user_notification_preference
 
     # Notifications
+    clear_all_user_notification_batches = d.clear_all_user_notification_batches
     create_or_add_to_user_notification_batch = (
         d.create_or_add_to_user_notification_batch
     )
     empty_user_notification_batch = d.empty_user_notification_batch
+    remove_notifications_from_batch = d.remove_notifications_from_batch
     get_all_batches_by_type = d.get_all_batches_by_type
     get_user_notification_batch = d.get_user_notification_batch
     get_user_notification_oldest_message_in_batch = (

autogpt_platform/backend/backend/executor/manager.py

@@ -7,8 +7,10 @@ import uuid
 from collections import defaultdict
 from concurrent.futures import Future, ThreadPoolExecutor
 from contextlib import asynccontextmanager
+from datetime import datetime, timedelta, timezone
 from typing import TYPE_CHECKING, Any, Optional, TypeVar, cast
 
+import sentry_sdk
 from pika.adapters.blocking_connection import BlockingChannel
 from pika.spec import Basic, BasicProperties
 from prometheus_client import Gauge, start_http_server
@@ -84,7 +86,11 @@ from backend.util.file import clean_exec_files
 from backend.util.logging import TruncatedLogger, configure_logging
 from backend.util.metrics import DiscordChannel
 from backend.util.process import AppProcess, set_service_name
-from backend.util.retry import continuous_retry, func_retry
+from backend.util.retry import (
+    continuous_retry,
+    func_retry,
+    send_rate_limited_discord_alert,
+)
 from backend.util.settings import Settings
 
 from .cluster_lock import ClusterLock
@@ -184,6 +190,7 @@ async def execute_node(
         _input_data.inputs = input_data
         if nodes_input_masks:
             _input_data.nodes_input_masks = nodes_input_masks
+        _input_data.user_id = user_id
         input_data = _input_data.model_dump()
     data.inputs = input_data
 
@@ -218,14 +225,37 @@ async def execute_node(
         extra_exec_kwargs[field_name] = credentials
 
     output_size = 0
+
+    # sentry tracking nonsense to get user counts for blocks because isolation scopes don't work :(
+    scope = sentry_sdk.get_current_scope()
+
+    # save the tags
+    original_user = scope._user
+    original_tags = dict(scope._tags) if scope._tags else {}
+    # Set user ID for error tracking
+    scope.set_user({"id": user_id})
+
+    scope.set_tag("graph_id", graph_id)
+    scope.set_tag("node_id", node_id)
+    scope.set_tag("block_name", node_block.name)
+    scope.set_tag("block_id", node_block.id)
+    for k, v in (data.user_context or UserContext(timezone="UTC")).model_dump().items():
+        scope.set_tag(f"user_context.{k}", v)
+
     try:
         async for output_name, output_data in node_block.execute(
             input_data, **extra_exec_kwargs
         ):
-            output_data = json.convert_pydantic_to_json(output_data)
+            output_data = json.to_dict(output_data)
             output_size += len(json.dumps(output_data))
             log_metadata.debug("Node produced output", **{output_name: output_data})
             yield output_name, output_data
+    except Exception:
+        # Capture exception WITH context still set before restoring scope
+        sentry_sdk.capture_exception(scope=scope)
+        sentry_sdk.flush()  # Ensure it's sent before we restore scope
+        # Re-raise to maintain normal error flow
+        raise
     finally:
         # Ensure credentials are released even if execution fails
         if creds_lock and (await creds_lock.locked()) and (await creds_lock.owned()):
@@ -240,6 +270,10 @@ async def execute_node(
             execution_stats.input_size = input_size
             execution_stats.output_size = output_size
 
+        # Restore scope AFTER error has been captured
+        scope._user = original_user
+        scope._tags = original_tags
+
 
 async def _enqueue_next_nodes(
     db_client: "DatabaseManagerAsyncClient",
@@ -564,7 +598,6 @@ class ExecutionProcessor:
                 await persist_output(
                     "error", str(stats.error) or type(stats.error).__name__
                 )
-
         return status
 
     @func_retry
@@ -979,16 +1012,31 @@ class ExecutionProcessor:
                 if isinstance(e, Exception)
                 else Exception(f"{e.__class__.__name__}: {e}")
             )
+            if not execution_stats.error:
+                execution_stats.error = str(error)
 
             known_errors = (InsufficientBalanceError, ModerationError)
             if isinstance(error, known_errors):
-                execution_stats.error = str(error)
                 return ExecutionStatus.FAILED
 
             execution_status = ExecutionStatus.FAILED
             log_metadata.exception(
                 f"Failed graph execution {graph_exec.graph_exec_id}: {error}"
             )
+
+            # Send rate-limited Discord alert for unknown/unexpected errors
+            send_rate_limited_discord_alert(
+                "graph_execution",
+                error,
+                "unknown_error",
+                f"🚨 **Unknown Graph Execution Error**\n"
+                f"User: {graph_exec.user_id}\n"
+                f"Graph ID: {graph_exec.graph_id}\n"
+                f"Execution ID: {graph_exec.graph_exec_id}\n"
+                f"Error Type: {type(error).__name__}\n"
+                f"Error: {str(error)[:200]}{'...' if len(str(error)) > 200 else ''}\n",
+            )
+
             raise
 
         finally:
@@ -1163,9 +1211,9 @@ class ExecutionProcessor:
                 f"❌ **Insufficient Funds Alert**\n"
                 f"User: {user_email or user_id}\n"
                 f"Agent: {metadata.name if metadata else 'Unknown Agent'}\n"
-                f"Current balance: ${e.balance/100:.2f}\n"
-                f"Attempted cost: ${abs(e.amount)/100:.2f}\n"
-                f"Shortfall: ${abs(shortfall)/100:.2f}\n"
+                f"Current balance: ${e.balance / 100:.2f}\n"
+                f"Attempted cost: ${abs(e.amount) / 100:.2f}\n"
+                f"Shortfall: ${abs(shortfall) / 100:.2f}\n"
                 f"[View User Details]({base_url}/admin/spending?search={user_email})"
             )
 
@@ -1212,9 +1260,9 @@ class ExecutionProcessor:
                 alert_message = (
                     f"⚠️ **Low Balance Alert**\n"
                     f"User: {user_email or user_id}\n"
-                    f"Balance dropped below ${LOW_BALANCE_THRESHOLD/100:.2f}\n"
-                    f"Current balance: ${current_balance/100:.2f}\n"
-                    f"Transaction cost: ${transaction_cost/100:.2f}\n"
+                    f"Balance dropped below ${LOW_BALANCE_THRESHOLD / 100:.2f}\n"
+                    f"Current balance: ${current_balance / 100:.2f}\n"
+                    f"Transaction cost: ${transaction_cost / 100:.2f}\n"
                     f"[View User Details]({base_url}/admin/spending?search={user_email})"
                 )
                 get_notification_manager_client().discord_system_alert(
@@ -1445,10 +1493,39 @@ class ExecutionManager(AppProcess):
             return
 
         graph_exec_id = graph_exec_entry.graph_exec_id
+        user_id = graph_exec_entry.user_id
+        graph_id = graph_exec_entry.graph_id
         logger.info(
-            f"[{self.service_name}] Received RUN for graph_exec_id={graph_exec_id}"
+            f"[{self.service_name}] Received RUN for graph_exec_id={graph_exec_id}, user_id={user_id}"
         )
 
+        # Check user rate limit before processing
+        try:
+            # Only check executions from the last 24 hours for performance
+            current_running_count = get_db_client().get_graph_executions_count(
+                user_id=user_id,
+                graph_id=graph_id,
+                statuses=[ExecutionStatus.RUNNING],
+                created_time_gte=datetime.now(timezone.utc) - timedelta(hours=24),
+            )
+
+            if (
+                current_running_count
+                >= settings.config.max_concurrent_graph_executions_per_user
+            ):
+                logger.warning(
+                    f"[{self.service_name}] Rate limit exceeded for user {user_id} on graph {graph_id}: "
+                    f"{current_running_count}/{settings.config.max_concurrent_graph_executions_per_user} running executions"
+                )
+                _ack_message(reject=True, requeue=True)
+                return
+
+        except Exception as e:
+            logger.error(
+                f"[{self.service_name}] Failed to check rate limit for user {user_id}: {e}, proceeding with execution"
+            )
+            # If rate limit check fails, proceed to avoid blocking executions
+
         # Check for local duplicate execution first
         if graph_exec_id in self.active_graph_runs:
             logger.warning(
@@ -1471,11 +1548,12 @@ class ExecutionManager(AppProcess):
                 logger.warning(
                     f"[{self.service_name}] Graph {graph_exec_id} already running on pod {current_owner}"
                 )
+                _ack_message(reject=True, requeue=False)
             else:
                 logger.warning(
                     f"[{self.service_name}] Could not acquire lock for {graph_exec_id} - Redis unavailable"
                 )
-            _ack_message(reject=True, requeue=True)
+                _ack_message(reject=True, requeue=True)
             return
         self._execution_locks[graph_exec_id] = cluster_lock
 
@@ -1636,6 +1714,8 @@ class ExecutionManager(AppProcess):
 
         logger.info(f"{prefix} ✅ Finished GraphExec cleanup")
 
+        super().cleanup()
+
 
 # ------- UTILITIES ------- #
 

autogpt_platform/backend/backend/executor/scheduler.py

@@ -248,7 +248,7 @@ class Scheduler(AppService):
             raise UnhealthyServiceError("Scheduler is still initializing")
 
         # Check if we're in the middle of cleanup
-        if self.cleaned_up:
+        if self._shutting_down:
             return await super().health_check()
 
         # Normal operation - check if scheduler is running
@@ -375,7 +375,6 @@ class Scheduler(AppService):
         super().run_service()
 
     def cleanup(self):
-        super().cleanup()
         if self.scheduler:
             logger.info("⏳ Shutting down scheduler...")
             self.scheduler.shutdown(wait=True)
@@ -390,7 +389,7 @@ class Scheduler(AppService):
             logger.info("⏳ Waiting for event loop thread to finish...")
             _event_loop_thread.join(timeout=SCHEDULER_OPERATION_TIMEOUT_SECONDS)
 
-        logger.info("Scheduler cleanup complete.")
+        super().cleanup()
 
     @expose
     def add_graph_execution_schedule(

autogpt_platform/backend/backend/executor/utils.py

@@ -34,6 +34,7 @@ from backend.data.graph import GraphModel, Node
 from backend.data.model import CredentialsMetaInput
 from backend.data.rabbitmq import Exchange, ExchangeType, Queue, RabbitMQConfig
 from backend.data.user import get_user_by_id
+from backend.util.cache import cached
 from backend.util.clients import (
     get_async_execution_event_bus,
     get_async_execution_queue,
@@ -41,11 +42,12 @@ from backend.util.clients import (
     get_integration_credentials_store,
 )
 from backend.util.exceptions import GraphValidationError, NotFoundError
-from backend.util.logging import TruncatedLogger
+from backend.util.logging import TruncatedLogger, is_structured_logging_enabled
 from backend.util.settings import Config
 from backend.util.type import convert
 
 
+@cached(maxsize=1000, ttl_seconds=3600)
 async def get_user_context(user_id: str) -> UserContext:
     """
     Get UserContext for a user, always returns a valid context with timezone.
@@ -53,7 +55,11 @@ async def get_user_context(user_id: str) -> UserContext:
     """
     user_context = UserContext(timezone="UTC")  # Default to UTC
     try:
-        user = await get_user_by_id(user_id)
+        if prisma.is_connected():
+            user = await get_user_by_id(user_id)
+        else:
+            user = await get_database_manager_async_client().get_user_by_id(user_id)
+
         if user and user.timezone and user.timezone != "not-set":
             user_context.timezone = user.timezone
             logger.debug(f"Retrieved user context: timezone={user.timezone}")
@@ -93,7 +99,11 @@ class LogMetadata(TruncatedLogger):
             "node_id": node_id,
             "block_name": block_name,
         }
-        prefix = f"[ExecutionManager|uid:{user_id}|gid:{graph_id}|nid:{node_id}]|geid:{graph_eid}|neid:{node_eid}|{block_name}]"
+        prefix = (
+            "[ExecutionManager]"
+            if is_structured_logging_enabled()
+            else f"[ExecutionManager|uid:{user_id}|gid:{graph_id}|nid:{node_id}]|geid:{graph_eid}|neid:{node_eid}|{block_name}]"  # noqa
+        )
         super().__init__(
             logger,
             max_length=max_length,

autogpt_platform/backend/backend/integrations/webhooks/__init__.py

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from autogpt_libs.utils.cache import cached
+from backend.util.cache import cached
 
 if TYPE_CHECKING:
     from ..providers import ProviderName
@@ -8,7 +8,7 @@ if TYPE_CHECKING:
 
 
 # --8<-- [start:load_webhook_managers]
-@cached()
+@cached(ttl_seconds=3600)
 def load_webhook_managers() -> dict["ProviderName", type["BaseWebhooksManager"]]:
     webhook_managers = {}
 

autogpt_platform/backend/backend/notifications/notifications.py

@@ -25,7 +25,11 @@ from backend.data.notifications import (
     get_summary_params_type,
 )
 from backend.data.rabbitmq import Exchange, ExchangeType, Queue, RabbitMQConfig
-from backend.data.user import generate_unsubscribe_link
+from backend.data.user import (
+    disable_all_user_notifications,
+    generate_unsubscribe_link,
+    set_user_email_verification,
+)
 from backend.notifications.email import EmailSender
 from backend.util.clients import get_database_manager_async_client
 from backend.util.logging import TruncatedLogger
@@ -38,7 +42,7 @@ from backend.util.service import (
     endpoint_to_sync,
     expose,
 )
-from backend.util.settings import Settings
+from backend.util.settings import AppEnvironment, Settings
 
 logger = TruncatedLogger(logging.getLogger(__name__), "[NotificationManager]")
 settings = Settings()
@@ -124,6 +128,12 @@ def get_routing_key(event_type: NotificationType) -> str:
 
 def queue_notification(event: NotificationEventModel) -> NotificationResult:
     """Queue a notification - exposed method for other services to call"""
+    # Disable in production
+    if settings.config.app_env == AppEnvironment.PRODUCTION:
+        return NotificationResult(
+            success=True,
+            message="Queueing notifications is disabled in production",
+        )
     try:
         logger.debug(f"Received Request to queue {event=}")
 
@@ -151,6 +161,12 @@ def queue_notification(event: NotificationEventModel) -> NotificationResult:
 
 async def queue_notification_async(event: NotificationEventModel) -> NotificationResult:
     """Queue a notification - exposed method for other services to call"""
+    # Disable in production
+    if settings.config.app_env == AppEnvironment.PRODUCTION:
+        return NotificationResult(
+            success=True,
+            message="Queueing notifications is disabled in production",
+        )
     try:
         logger.debug(f"Received Request to queue {event=}")
 
@@ -213,6 +229,9 @@ class NotificationManager(AppService):
 
     @expose
     async def queue_weekly_summary(self):
+        # disable in prod
+        if settings.config.app_env == AppEnvironment.PRODUCTION:
+            return
         # Use the existing event loop instead of creating a new one with asyncio.run()
         asyncio.create_task(self._queue_weekly_summary())
 
@@ -226,7 +245,9 @@ class NotificationManager(AppService):
             logger.info(
                 f"Querying for active users between {start_time} and {current_time}"
             )
-            users = await get_database_manager_async_client().get_active_user_ids_in_timerange(
+            users = await get_database_manager_async_client(
+                should_retry=False
+            ).get_active_user_ids_in_timerange(
                 end_time=current_time.isoformat(),
                 start_time=start_time.isoformat(),
             )
@@ -253,6 +274,9 @@ class NotificationManager(AppService):
     async def process_existing_batches(
         self, notification_types: list[NotificationType]
     ):
+        # disable in prod
+        if settings.config.app_env == AppEnvironment.PRODUCTION:
+            return
         # Use the existing event loop instead of creating a new process
         asyncio.create_task(self._process_existing_batches(notification_types))
 
@@ -266,15 +290,15 @@ class NotificationManager(AppService):
 
             for notification_type in notification_types:
                 # Get all batches for this notification type
-                batches = (
-                    await get_database_manager_async_client().get_all_batches_by_type(
-                        notification_type
-                    )
-                )
+                batches = await get_database_manager_async_client(
+                    should_retry=False
+                ).get_all_batches_by_type(notification_type)
 
                 for batch in batches:
                     # Check if batch has aged out
-                    oldest_message = await get_database_manager_async_client().get_user_notification_oldest_message_in_batch(
+                    oldest_message = await get_database_manager_async_client(
+                        should_retry=False
+                    ).get_user_notification_oldest_message_in_batch(
                         batch.user_id, notification_type
                     )
 
@@ -289,9 +313,9 @@ class NotificationManager(AppService):
 
                     # If batch has aged out, process it
                     if oldest_message.created_at + max_delay < current_time:
-                        recipient_email = await get_database_manager_async_client().get_user_email_by_id(
-                            batch.user_id
-                        )
+                        recipient_email = await get_database_manager_async_client(
+                            should_retry=False
+                        ).get_user_email_by_id(batch.user_id)
 
                         if not recipient_email:
                             logger.error(
@@ -308,21 +332,25 @@ class NotificationManager(AppService):
                                 f"User {batch.user_id} does not want to receive {notification_type} notifications"
                             )
                             # Clear the batch
-                            await get_database_manager_async_client().empty_user_notification_batch(
+                            await get_database_manager_async_client(
+                                should_retry=False
+                            ).empty_user_notification_batch(
                                 batch.user_id, notification_type
                             )
                             continue
 
-                        batch_data = await get_database_manager_async_client().get_user_notification_batch(
-                            batch.user_id, notification_type
-                        )
+                        batch_data = await get_database_manager_async_client(
+                            should_retry=False
+                        ).get_user_notification_batch(batch.user_id, notification_type)
 
                         if not batch_data or not batch_data.notifications:
                             logger.error(
                                 f"Batch data not found for user {batch.user_id}"
                             )
                             # Clear the batch
-                            await get_database_manager_async_client().empty_user_notification_batch(
+                            await get_database_manager_async_client(
+                                should_retry=False
+                            ).empty_user_notification_batch(
                                 batch.user_id, notification_type
                             )
                             continue
@@ -358,7 +386,9 @@ class NotificationManager(AppService):
                         )
 
                         # Clear the batch
-                        await get_database_manager_async_client().empty_user_notification_batch(
+                        await get_database_manager_async_client(
+                            should_retry=False
+                        ).empty_user_notification_batch(
                             batch.user_id, notification_type
                         )
 
@@ -413,15 +443,13 @@ class NotificationManager(AppService):
         self, user_id: str, event_type: NotificationType
     ) -> bool:
         """Check if a user wants to receive a notification based on their preferences and email verification status"""
-        validated_email = (
-            await get_database_manager_async_client().get_user_email_verification(
-                user_id
-            )
-        )
+        validated_email = await get_database_manager_async_client(
+            should_retry=False
+        ).get_user_email_verification(user_id)
         preference = (
-            await get_database_manager_async_client().get_user_notification_preference(
-                user_id
-            )
+            await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_notification_preference(user_id)
         ).preferences.get(event_type, True)
         # only if both are true, should we email this person
         return validated_email and preference
@@ -437,7 +465,9 @@ class NotificationManager(AppService):
 
         try:
             # Get summary data from the database
-            summary_data = await get_database_manager_async_client().get_user_execution_summary_data(
+            summary_data = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_execution_summary_data(
                 user_id=user_id,
                 start_time=params.start_date,
                 end_time=params.end_date,
@@ -524,13 +554,13 @@ class NotificationManager(AppService):
         self, user_id: str, event_type: NotificationType, event: NotificationEventModel
     ) -> bool:
 
-        await get_database_manager_async_client().create_or_add_to_user_notification_batch(
-            user_id, event_type, event
-        )
+        await get_database_manager_async_client(
+            should_retry=False
+        ).create_or_add_to_user_notification_batch(user_id, event_type, event)
 
-        oldest_message = await get_database_manager_async_client().get_user_notification_oldest_message_in_batch(
-            user_id, event_type
-        )
+        oldest_message = await get_database_manager_async_client(
+            should_retry=False
+        ).get_user_notification_oldest_message_in_batch(user_id, event_type)
         if not oldest_message:
             logger.error(
                 f"Batch for user {user_id} and type {event_type} has no oldest message whichshould never happen!!!!!!!!!!!!!!!!"
@@ -580,11 +610,9 @@ class NotificationManager(AppService):
                 return False
             logger.debug(f"Processing immediate notification: {event}")
 
-            recipient_email = (
-                await get_database_manager_async_client().get_user_email_by_id(
-                    event.user_id
-                )
-            )
+            recipient_email = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_email_by_id(event.user_id)
             if not recipient_email:
                 logger.error(f"User email not found for user {event.user_id}")
                 return False
@@ -619,11 +647,9 @@ class NotificationManager(AppService):
                 return False
             logger.info(f"Processing batch notification: {event}")
 
-            recipient_email = (
-                await get_database_manager_async_client().get_user_email_by_id(
-                    event.user_id
-                )
-            )
+            recipient_email = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_email_by_id(event.user_id)
             if not recipient_email:
                 logger.error(f"User email not found for user {event.user_id}")
                 return False
@@ -642,11 +668,9 @@ class NotificationManager(AppService):
             if not should_send:
                 logger.info("Batch not old enough to send")
                 return False
-            batch = (
-                await get_database_manager_async_client().get_user_notification_batch(
-                    event.user_id, event.type
-                )
-            )
+            batch = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_notification_batch(event.user_id, event.type)
             if not batch or not batch.notifications:
                 logger.error(f"Batch not found for user {event.user_id}")
                 return False
@@ -657,6 +681,7 @@ class NotificationManager(AppService):
                     get_notif_data_type(db_event.type)
                 ].model_validate(
                     {
+                        "id": db_event.id,  # Include ID from database
                         "user_id": event.user_id,
                         "type": db_event.type,
                         "data": db_event.data,
@@ -679,6 +704,9 @@ class NotificationManager(AppService):
                 chunk_sent = False
                 for attempt_size in [chunk_size, 50, 25, 10, 5, 1]:
                     chunk = batch_messages[i : i + attempt_size]
+                    chunk_ids = [
+                        msg.id for msg in chunk if msg.id
+                    ]  # Extract IDs for removal
 
                     try:
                         # Try to render the email to check its size
@@ -705,6 +733,23 @@ class NotificationManager(AppService):
                                 user_unsub_link=unsub_link,
                             )
 
+                            # Remove successfully sent notifications immediately
+                            if chunk_ids:
+                                try:
+                                    await get_database_manager_async_client(
+                                        should_retry=False
+                                    ).remove_notifications_from_batch(
+                                        event.user_id, event.type, chunk_ids
+                                    )
+                                    logger.info(
+                                        f"Removed {len(chunk_ids)} sent notifications from batch"
+                                    )
+                                except Exception as e:
+                                    logger.error(
+                                        f"Failed to remove sent notifications: {e}"
+                                    )
+                                    # Continue anyway - better to risk duplicates than lose emails
+
                             # Track successful sends
                             successfully_sent_count += len(chunk)
 
@@ -722,13 +767,137 @@ class NotificationManager(AppService):
                             i += len(chunk)
                             chunk_sent = True
                             break
+                        else:
+                            # Message is too large even after size reduction
+                            if attempt_size == 1:
+                                logger.error(
+                                    f"Failed to send notification at index {i}: "
+                                    f"Single notification exceeds email size limit "
+                                    f"({len(test_message):,} chars > {MAX_EMAIL_SIZE:,} chars). "
+                                    f"Removing permanently from batch - will not retry."
+                                )
+
+                                # Remove the oversized notification permanently - it will NEVER fit
+                                if chunk_ids:
+                                    try:
+                                        await get_database_manager_async_client(
+                                            should_retry=False
+                                        ).remove_notifications_from_batch(
+                                            event.user_id, event.type, chunk_ids
+                                        )
+                                        logger.info(
+                                            f"Removed oversized notification {chunk_ids[0]} from batch permanently"
+                                        )
+                                    except Exception as e:
+                                        logger.error(
+                                            f"Failed to remove oversized notification: {e}"
+                                        )
+
+                                failed_indices.append(i)
+                                i += 1
+                                chunk_sent = True
+                                break
+                            # Try smaller chunk size
+                            continue
                     except Exception as e:
+                        # Check if it's a Postmark API error
                         if attempt_size == 1:
-                            # Even single notification is too large
-                            logger.error(
-                                f"Single notification too large to send: {e}. "
-                                f"Skipping notification at index {i}"
-                            )
+                            # Single notification failed - determine the actual cause
+                            error_message = str(e).lower()
+                            error_type = type(e).__name__
+
+                            # Check for HTTP 406 - Inactive recipient (common in Postmark errors)
+                            if "406" in error_message or "inactive" in error_message:
+                                logger.warning(
+                                    f"Failed to send notification at index {i}: "
+                                    f"Recipient marked as inactive by Postmark. "
+                                    f"Error: {e}. Disabling ALL notifications for this user."
+                                )
+
+                                # 1. Mark email as unverified
+                                try:
+                                    await set_user_email_verification(
+                                        event.user_id, False
+                                    )
+                                    logger.info(
+                                        f"Set email verification to false for user {event.user_id}"
+                                    )
+                                except Exception as deactivation_error:
+                                    logger.error(
+                                        f"Failed to deactivate email for user {event.user_id}: "
+                                        f"{deactivation_error}"
+                                    )
+
+                                # 2. Disable all notification preferences
+                                try:
+                                    await disable_all_user_notifications(event.user_id)
+                                    logger.info(
+                                        f"Disabled all notification preferences for user {event.user_id}"
+                                    )
+                                except Exception as disable_error:
+                                    logger.error(
+                                        f"Failed to disable notification preferences: {disable_error}"
+                                    )
+
+                                # 3. Clear ALL notification batches for this user
+                                try:
+                                    await get_database_manager_async_client(
+                                        should_retry=False
+                                    ).clear_all_user_notification_batches(event.user_id)
+                                    logger.info(
+                                        f"Cleared ALL notification batches for user {event.user_id}"
+                                    )
+                                except Exception as remove_error:
+                                    logger.error(
+                                        f"Failed to clear batches for inactive recipient: {remove_error}"
+                                    )
+
+                                # Stop processing - we've nuked everything for this user
+                                return True
+                            # Check for HTTP 422 - Malformed data
+                            elif (
+                                "422" in error_message
+                                or "unprocessable" in error_message
+                            ):
+                                logger.error(
+                                    f"Failed to send notification at index {i}: "
+                                    f"Malformed notification data rejected by Postmark. "
+                                    f"Error: {e}. Removing from batch permanently."
+                                )
+
+                                # Remove from batch - 422 means bad data that won't fix itself
+                                if chunk_ids:
+                                    try:
+                                        await get_database_manager_async_client(
+                                            should_retry=False
+                                        ).remove_notifications_from_batch(
+                                            event.user_id, event.type, chunk_ids
+                                        )
+                                        logger.info(
+                                            "Removed malformed notification from batch permanently"
+                                        )
+                                    except Exception as remove_error:
+                                        logger.error(
+                                            f"Failed to remove malformed notification: {remove_error}"
+                                        )
+                            # Check if it's a ValueError for size limit
+                            elif (
+                                isinstance(e, ValueError)
+                                and "too large" in error_message
+                            ):
+                                logger.error(
+                                    f"Failed to send notification at index {i}: "
+                                    f"Notification size exceeds email limit. "
+                                    f"Error: {e}. Skipping this notification."
+                                )
+                            # Other API errors
+                            else:
+                                logger.error(
+                                    f"Failed to send notification at index {i}: "
+                                    f"Email API error ({error_type}): {e}. "
+                                    f"Skipping this notification."
+                                )
+
                             failed_indices.append(i)
                             i += 1
                             chunk_sent = True
@@ -742,18 +911,20 @@ class NotificationManager(AppService):
                     failed_indices.append(i)
                     i += 1
 
-            # Only empty the batch if ALL notifications were sent successfully
-            if successfully_sent_count == len(batch_messages):
+            # Check what remains in the batch (notifications are removed as sent)
+            remaining_batch = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_notification_batch(event.user_id, event.type)
+
+            if not remaining_batch or not remaining_batch.notifications:
                 logger.info(
-                    f"Successfully sent all {successfully_sent_count} notifications, clearing batch"
-                )
-                await get_database_manager_async_client().empty_user_notification_batch(
-                    event.user_id, event.type
+                    f"All {successfully_sent_count} notifications sent and removed from batch"
                 )
             else:
+                remaining_count = len(remaining_batch.notifications)
                 logger.warning(
-                    f"Only sent {successfully_sent_count} of {len(batch_messages)} notifications. "
-                    f"Failed indices: {failed_indices}. Batch will be retained for retry."
+                    f"Sent {successfully_sent_count} notifications. "
+                    f"{remaining_count} remain in batch for retry due to errors."
                 )
             return True
         except Exception as e:
@@ -771,11 +942,9 @@ class NotificationManager(AppService):
 
             logger.info(f"Processing summary notification: {model}")
 
-            recipient_email = (
-                await get_database_manager_async_client().get_user_email_by_id(
-                    event.user_id
-                )
-            )
+            recipient_email = await get_database_manager_async_client(
+                should_retry=False
+            ).get_user_email_by_id(event.user_id)
             if not recipient_email:
                 logger.error(f"User email not found for user {event.user_id}")
                 return False
@@ -848,10 +1017,14 @@ class NotificationManager(AppService):
             logger.exception(f"Fatal error in consumer for {queue_name}: {e}")
             raise
 
-    @continuous_retry()
     def run_service(self):
-        self.run_and_wait(self._run_service())
+        # Queue the main _run_service task
+        asyncio.run_coroutine_threadsafe(self._run_service(), self.shared_event_loop)
 
+        # Start the main event loop
+        super().run_service()
+
+    @continuous_retry()
     async def _run_service(self):
         logger.info(f"[{self.service_name}] ⏳ Configuring RabbitMQ...")
         self.rabbitmq_service = rabbitmq.AsyncRabbitMQ(self.rabbitmq_config)
@@ -917,10 +1090,11 @@ class NotificationManager(AppService):
     def cleanup(self):
         """Cleanup service resources"""
         self.running = False
-        super().cleanup()
-        logger.info(f"[{self.service_name}] ⏳ Disconnecting RabbitMQ...")
+        logger.info("⏳ Disconnecting RabbitMQ...")
         self.run_and_wait(self.rabbitmq_service.disconnect())
 
+        super().cleanup()
+
 
 class NotificationManagerClient(AppServiceClient):
     @classmethod

autogpt_platform/backend/backend/notifications/test_notifications.py

@@ -0,0 +1,598 @@
+"""Tests for notification error handling in NotificationManager."""
+
+from datetime import datetime, timezone
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
+
+import pytest
+from prisma.enums import NotificationType
+
+from backend.data.notifications import AgentRunData, NotificationEventModel
+from backend.notifications.notifications import NotificationManager
+
+
+class TestNotificationErrorHandling:
+    """Test cases for notification error handling in NotificationManager."""
+
+    @pytest.fixture
+    def notification_manager(self):
+        """Create a NotificationManager instance for testing."""
+        with patch("backend.notifications.notifications.AppService.__init__"):
+            manager = NotificationManager()
+            manager.email_sender = MagicMock()
+            # Mock the _get_template method used by _process_batch
+            template_mock = Mock()
+            template_mock.base_template = "base"
+            template_mock.subject_template = "subject"
+            template_mock.body_template = "body"
+            manager.email_sender._get_template = Mock(return_value=template_mock)
+            # Mock the formatter
+            manager.email_sender.formatter = Mock()
+            manager.email_sender.formatter.format_email = Mock(
+                return_value=("subject", "body content")
+            )
+            manager.email_sender.formatter.env = Mock()
+            manager.email_sender.formatter.env.globals = {
+                "base_url": "http://example.com"
+            }
+            return manager
+
+    @pytest.fixture
+    def sample_batch_event(self):
+        """Create a sample batch event for testing."""
+        return NotificationEventModel(
+            type=NotificationType.AGENT_RUN,
+            user_id="user_1",
+            created_at=datetime.now(timezone.utc),
+            data=AgentRunData(
+                agent_name="Test Agent",
+                credits_used=10.0,
+                execution_time=5.0,
+                node_count=3,
+                graph_id="graph_1",
+                outputs=[],
+            ),
+        )
+
+    @pytest.fixture
+    def sample_batch_notifications(self):
+        """Create sample batch notifications for testing."""
+        notifications = []
+        for i in range(3):
+            notification = Mock()
+            notification.type = NotificationType.AGENT_RUN
+            notification.data = {
+                "agent_name": f"Test Agent {i}",
+                "credits_used": 10.0 * (i + 1),
+                "execution_time": 5.0 * (i + 1),
+                "node_count": 3 + i,
+                "graph_id": f"graph_{i}",
+                "outputs": [],
+            }
+            notification.created_at = datetime.now(timezone.utc)
+            notifications.append(notification)
+        return notifications
+
+    @pytest.mark.asyncio
+    async def test_406_stops_all_processing_for_user(
+        self, notification_manager, sample_batch_event
+    ):
+        """Test that 406 inactive recipient error stops ALL processing for that user."""
+        with patch("backend.notifications.notifications.logger"), patch(
+            "backend.notifications.notifications.set_user_email_verification",
+            new_callable=AsyncMock,
+        ) as mock_set_verification, patch(
+            "backend.notifications.notifications.disable_all_user_notifications",
+            new_callable=AsyncMock,
+        ) as mock_disable_all, patch(
+            "backend.notifications.notifications.get_database_manager_async_client"
+        ) as mock_db_client, patch(
+            "backend.notifications.notifications.generate_unsubscribe_link"
+        ) as mock_unsub_link:
+
+            # Create batch of 5 notifications
+            notifications = []
+            for i in range(5):
+                notification = Mock()
+                notification.id = f"notif_{i}"
+                notification.type = NotificationType.AGENT_RUN
+                notification.data = {
+                    "agent_name": f"Test Agent {i}",
+                    "credits_used": 10.0 * (i + 1),
+                    "execution_time": 5.0 * (i + 1),
+                    "node_count": 3 + i,
+                    "graph_id": f"graph_{i}",
+                    "outputs": [],
+                }
+                notification.created_at = datetime.now(timezone.utc)
+                notifications.append(notification)
+
+            # Setup mocks
+            mock_db = mock_db_client.return_value
+            mock_db.get_user_email_by_id = AsyncMock(return_value="test@example.com")
+            mock_db.get_user_notification_batch = AsyncMock(
+                return_value=Mock(notifications=notifications)
+            )
+            mock_db.clear_all_user_notification_batches = AsyncMock()
+            mock_db.remove_notifications_from_batch = AsyncMock()
+            mock_unsub_link.return_value = "http://example.com/unsub"
+
+            # Mock internal methods
+            notification_manager._should_email_user_based_on_preference = AsyncMock(
+                return_value=True
+            )
+            notification_manager._should_batch = AsyncMock(return_value=True)
+            notification_manager._parse_message = Mock(return_value=sample_batch_event)
+
+            # Track calls
+            call_count = [0]
+
+            def send_side_effect(*args, **kwargs):
+                data = kwargs.get("data", [])
+                if isinstance(data, list) and len(data) == 1:
+                    current_call = call_count[0]
+                    call_count[0] += 1
+
+                    # First two succeed, third hits 406
+                    if current_call < 2:
+                        return None
+                    else:
+                        raise Exception("Recipient marked as inactive (406)")
+                # Force single processing
+                raise Exception("Force single processing")
+
+            notification_manager.email_sender.send_templated.side_effect = (
+                send_side_effect
+            )
+
+            # Act
+            result = await notification_manager._process_batch(
+                sample_batch_event.model_dump_json()
+            )
+
+            # Assert
+            assert result is True
+
+            # Only 3 calls should have been made (2 successful, 1 failed with 406)
+            assert call_count[0] == 3
+
+            # User should be deactivated
+            mock_set_verification.assert_called_once_with("user_1", False)
+            mock_disable_all.assert_called_once_with("user_1")
+            mock_db.clear_all_user_notification_batches.assert_called_once_with(
+                "user_1"
+            )
+
+            # No further processing should occur after 406
+
+    @pytest.mark.asyncio
+    async def test_422_permanently_removes_malformed_notification(
+        self, notification_manager, sample_batch_event
+    ):
+        """Test that 422 error permanently removes the malformed notification from batch and continues with others."""
+        with patch("backend.notifications.notifications.logger") as mock_logger, patch(
+            "backend.notifications.notifications.get_database_manager_async_client"
+        ) as mock_db_client, patch(
+            "backend.notifications.notifications.generate_unsubscribe_link"
+        ) as mock_unsub_link:
+
+            # Create batch of 5 notifications
+            notifications = []
+            for i in range(5):
+                notification = Mock()
+                notification.id = f"notif_{i}"
+                notification.type = NotificationType.AGENT_RUN
+                notification.data = {
+                    "agent_name": f"Test Agent {i}",
+                    "credits_used": 10.0 * (i + 1),
+                    "execution_time": 5.0 * (i + 1),
+                    "node_count": 3 + i,
+                    "graph_id": f"graph_{i}",
+                    "outputs": [],
+                }
+                notification.created_at = datetime.now(timezone.utc)
+                notifications.append(notification)
+
+            # Setup mocks
+            mock_db = mock_db_client.return_value
+            mock_db.get_user_email_by_id = AsyncMock(return_value="test@example.com")
+            mock_db.get_user_notification_batch = AsyncMock(
+                side_effect=[
+                    Mock(notifications=notifications),
+                    Mock(notifications=[]),  # Empty after processing
+                ]
+            )
+            mock_db.remove_notifications_from_batch = AsyncMock()
+            mock_unsub_link.return_value = "http://example.com/unsub"
+
+            # Mock internal methods
+            notification_manager._should_email_user_based_on_preference = AsyncMock(
+                return_value=True
+            )
+            notification_manager._should_batch = AsyncMock(return_value=True)
+            notification_manager._parse_message = Mock(return_value=sample_batch_event)
+
+            # Track calls
+            call_count = [0]
+            successful_indices = []
+            removed_notification_ids = []
+
+            # Capture what gets removed
+            def remove_side_effect(user_id, notif_type, notif_ids):
+                removed_notification_ids.extend(notif_ids)
+                return None
+
+            mock_db.remove_notifications_from_batch.side_effect = remove_side_effect
+
+            def send_side_effect(*args, **kwargs):
+                data = kwargs.get("data", [])
+                if isinstance(data, list) and len(data) == 1:
+                    current_call = call_count[0]
+                    call_count[0] += 1
+
+                    # Index 2 has malformed data (422)
+                    if current_call == 2:
+                        raise Exception(
+                            "Unprocessable entity (422): Malformed email data"
+                        )
+                    else:
+                        successful_indices.append(current_call)
+                        return None
+                # Force single processing
+                raise Exception("Force single processing")
+
+            notification_manager.email_sender.send_templated.side_effect = (
+                send_side_effect
+            )
+
+            # Act
+            result = await notification_manager._process_batch(
+                sample_batch_event.model_dump_json()
+            )
+
+            # Assert
+            assert result is True
+            assert call_count[0] == 5  # All 5 attempted
+            assert len(successful_indices) == 4  # 4 succeeded (all except index 2)
+            assert 2 not in successful_indices  # Index 2 failed
+
+            # Verify 422 error was logged
+            error_calls = [call[0][0] for call in mock_logger.error.call_args_list]
+            assert any(
+                "422" in call or "malformed" in call.lower() for call in error_calls
+            )
+
+            # Verify all notifications were removed (4 successful + 1 malformed)
+            assert mock_db.remove_notifications_from_batch.call_count == 5
+            assert (
+                "notif_2" in removed_notification_ids
+            )  # Malformed one was removed permanently
+
+    @pytest.mark.asyncio
+    async def test_oversized_notification_permanently_removed(
+        self, notification_manager, sample_batch_event
+    ):
+        """Test that oversized notifications are permanently removed from batch but others continue."""
+        with patch("backend.notifications.notifications.logger") as mock_logger, patch(
+            "backend.notifications.notifications.get_database_manager_async_client"
+        ) as mock_db_client, patch(
+            "backend.notifications.notifications.generate_unsubscribe_link"
+        ) as mock_unsub_link:
+
+            # Create batch of 5 notifications
+            notifications = []
+            for i in range(5):
+                notification = Mock()
+                notification.id = f"notif_{i}"
+                notification.type = NotificationType.AGENT_RUN
+                notification.data = {
+                    "agent_name": f"Test Agent {i}",
+                    "credits_used": 10.0 * (i + 1),
+                    "execution_time": 5.0 * (i + 1),
+                    "node_count": 3 + i,
+                    "graph_id": f"graph_{i}",
+                    "outputs": [],
+                }
+                notification.created_at = datetime.now(timezone.utc)
+                notifications.append(notification)
+
+            # Setup mocks
+            mock_db = mock_db_client.return_value
+            mock_db.get_user_email_by_id = AsyncMock(return_value="test@example.com")
+            mock_db.get_user_notification_batch = AsyncMock(
+                side_effect=[
+                    Mock(notifications=notifications),
+                    Mock(notifications=[]),  # Empty after processing
+                ]
+            )
+            mock_db.remove_notifications_from_batch = AsyncMock()
+            mock_unsub_link.return_value = "http://example.com/unsub"
+
+            # Mock internal methods
+            notification_manager._should_email_user_based_on_preference = AsyncMock(
+                return_value=True
+            )
+            notification_manager._should_batch = AsyncMock(return_value=True)
+            notification_manager._parse_message = Mock(return_value=sample_batch_event)
+
+            # Override formatter to simulate oversized on index 3
+            # original_format = notification_manager.email_sender.formatter.format_email
+
+            def format_side_effect(*args, **kwargs):
+                # Check if we're formatting index 3
+                data = kwargs.get("data", {}).get("notifications", [])
+                if data and len(data) == 1:
+                    # Check notification content to identify index 3
+                    if any(
+                        "Test Agent 3" in str(n.data)
+                        for n in data
+                        if hasattr(n, "data")
+                    ):
+                        # Return oversized message for index 3
+                        return ("subject", "x" * 5_000_000)  # Over 4.5MB limit
+                return ("subject", "normal sized content")
+
+            notification_manager.email_sender.formatter.format_email = Mock(
+                side_effect=format_side_effect
+            )
+
+            # Track calls
+            successful_indices = []
+
+            def send_side_effect(*args, **kwargs):
+                data = kwargs.get("data", [])
+                if isinstance(data, list) and len(data) == 1:
+                    # Track which notification was sent based on content
+                    for i, notif in enumerate(notifications):
+                        if any(
+                            f"Test Agent {i}" in str(n.data)
+                            for n in data
+                            if hasattr(n, "data")
+                        ):
+                            successful_indices.append(i)
+                            return None
+                    return None
+                # Force single processing
+                raise Exception("Force single processing")
+
+            notification_manager.email_sender.send_templated.side_effect = (
+                send_side_effect
+            )
+
+            # Act
+            result = await notification_manager._process_batch(
+                sample_batch_event.model_dump_json()
+            )
+
+            # Assert
+            assert result is True
+            assert (
+                len(successful_indices) == 4
+            )  # Only 4 sent (index 3 skipped due to size)
+            assert 3 not in successful_indices  # Index 3 was not sent
+
+            # Verify oversized error was logged
+            error_calls = [call[0][0] for call in mock_logger.error.call_args_list]
+            assert any(
+                "exceeds email size limit" in call or "oversized" in call.lower()
+                for call in error_calls
+            )
+
+    @pytest.mark.asyncio
+    async def test_generic_api_error_keeps_notification_for_retry(
+        self, notification_manager, sample_batch_event
+    ):
+        """Test that generic API errors keep notifications in batch for retry while others continue."""
+        with patch("backend.notifications.notifications.logger") as mock_logger, patch(
+            "backend.notifications.notifications.get_database_manager_async_client"
+        ) as mock_db_client, patch(
+            "backend.notifications.notifications.generate_unsubscribe_link"
+        ) as mock_unsub_link:
+
+            # Create batch of 5 notifications
+            notifications = []
+            for i in range(5):
+                notification = Mock()
+                notification.id = f"notif_{i}"
+                notification.type = NotificationType.AGENT_RUN
+                notification.data = {
+                    "agent_name": f"Test Agent {i}",
+                    "credits_used": 10.0 * (i + 1),
+                    "execution_time": 5.0 * (i + 1),
+                    "node_count": 3 + i,
+                    "graph_id": f"graph_{i}",
+                    "outputs": [],
+                }
+                notification.created_at = datetime.now(timezone.utc)
+                notifications.append(notification)
+
+            # Notification that failed with generic error
+            failed_notifications = [notifications[1]]  # Only index 1 remains for retry
+
+            # Setup mocks
+            mock_db = mock_db_client.return_value
+            mock_db.get_user_email_by_id = AsyncMock(return_value="test@example.com")
+            mock_db.get_user_notification_batch = AsyncMock(
+                side_effect=[
+                    Mock(notifications=notifications),
+                    Mock(
+                        notifications=failed_notifications
+                    ),  # Failed ones remain for retry
+                ]
+            )
+            mock_db.remove_notifications_from_batch = AsyncMock()
+            mock_unsub_link.return_value = "http://example.com/unsub"
+
+            # Mock internal methods
+            notification_manager._should_email_user_based_on_preference = AsyncMock(
+                return_value=True
+            )
+            notification_manager._should_batch = AsyncMock(return_value=True)
+            notification_manager._parse_message = Mock(return_value=sample_batch_event)
+
+            # Track calls
+            successful_indices = []
+            failed_indices = []
+            removed_notification_ids = []
+
+            # Capture what gets removed
+            def remove_side_effect(user_id, notif_type, notif_ids):
+                removed_notification_ids.extend(notif_ids)
+                return None
+
+            mock_db.remove_notifications_from_batch.side_effect = remove_side_effect
+
+            def send_side_effect(*args, **kwargs):
+                data = kwargs.get("data", [])
+                if isinstance(data, list) and len(data) == 1:
+                    # Track which notification based on content
+                    for i, notif in enumerate(notifications):
+                        if any(
+                            f"Test Agent {i}" in str(n.data)
+                            for n in data
+                            if hasattr(n, "data")
+                        ):
+                            # Index 1 has generic API error
+                            if i == 1:
+                                failed_indices.append(i)
+                                raise Exception("Network timeout - temporary failure")
+                            else:
+                                successful_indices.append(i)
+                                return None
+                    return None
+                # Force single processing
+                raise Exception("Force single processing")
+
+            notification_manager.email_sender.send_templated.side_effect = (
+                send_side_effect
+            )
+
+            # Act
+            result = await notification_manager._process_batch(
+                sample_batch_event.model_dump_json()
+            )
+
+            # Assert
+            assert result is True
+            assert len(successful_indices) == 4  # 4 succeeded (0, 2, 3, 4)
+            assert len(failed_indices) == 1  # 1 failed
+            assert 1 in failed_indices  # Index 1 failed
+
+            # Verify generic error was logged
+            error_calls = [call[0][0] for call in mock_logger.error.call_args_list]
+            assert any(
+                "api error" in call.lower() or "skipping" in call.lower()
+                for call in error_calls
+            )
+
+            # Only successful ones should be removed from batch (failed one stays for retry)
+            assert mock_db.remove_notifications_from_batch.call_count == 4
+            assert (
+                "notif_1" not in removed_notification_ids
+            )  # Failed one NOT removed (stays for retry)
+            assert "notif_0" in removed_notification_ids  # Successful one removed
+            assert "notif_2" in removed_notification_ids  # Successful one removed
+            assert "notif_3" in removed_notification_ids  # Successful one removed
+            assert "notif_4" in removed_notification_ids  # Successful one removed
+
+    @pytest.mark.asyncio
+    async def test_batch_all_notifications_sent_successfully(
+        self, notification_manager, sample_batch_event
+    ):
+        """Test successful batch processing where all notifications are sent without errors."""
+        with patch("backend.notifications.notifications.logger") as mock_logger, patch(
+            "backend.notifications.notifications.get_database_manager_async_client"
+        ) as mock_db_client, patch(
+            "backend.notifications.notifications.generate_unsubscribe_link"
+        ) as mock_unsub_link:
+
+            # Create batch of 5 notifications
+            notifications = []
+            for i in range(5):
+                notification = Mock()
+                notification.id = f"notif_{i}"
+                notification.type = NotificationType.AGENT_RUN
+                notification.data = {
+                    "agent_name": f"Test Agent {i}",
+                    "credits_used": 10.0 * (i + 1),
+                    "execution_time": 5.0 * (i + 1),
+                    "node_count": 3 + i,
+                    "graph_id": f"graph_{i}",
+                    "outputs": [],
+                }
+                notification.created_at = datetime.now(timezone.utc)
+                notifications.append(notification)
+
+            # Setup mocks
+            mock_db = mock_db_client.return_value
+            mock_db.get_user_email_by_id = AsyncMock(return_value="test@example.com")
+            mock_db.get_user_notification_batch = AsyncMock(
+                side_effect=[
+                    Mock(notifications=notifications),
+                    Mock(notifications=[]),  # Empty after all sent successfully
+                ]
+            )
+            mock_db.remove_notifications_from_batch = AsyncMock()
+            mock_unsub_link.return_value = "http://example.com/unsub"
+
+            # Mock internal methods
+            notification_manager._should_email_user_based_on_preference = AsyncMock(
+                return_value=True
+            )
+            notification_manager._should_batch = AsyncMock(return_value=True)
+            notification_manager._parse_message = Mock(return_value=sample_batch_event)
+
+            # Track successful sends
+            successful_indices = []
+            removed_notification_ids = []
+
+            # Capture what gets removed
+            def remove_side_effect(user_id, notif_type, notif_ids):
+                removed_notification_ids.extend(notif_ids)
+                return None
+
+            mock_db.remove_notifications_from_batch.side_effect = remove_side_effect
+
+            def send_side_effect(*args, **kwargs):
+                data = kwargs.get("data", [])
+                if isinstance(data, list) and len(data) == 1:
+                    # Track which notification was sent
+                    for i, notif in enumerate(notifications):
+                        if any(
+                            f"Test Agent {i}" in str(n.data)
+                            for n in data
+                            if hasattr(n, "data")
+                        ):
+                            successful_indices.append(i)
+                            return None
+                    return None  # Success
+                # Force single processing
+                raise Exception("Force single processing")
+
+            notification_manager.email_sender.send_templated.side_effect = (
+                send_side_effect
+            )
+
+            # Act
+            result = await notification_manager._process_batch(
+                sample_batch_event.model_dump_json()
+            )
+
+            # Assert
+            assert result is True
+
+            # All 5 notifications should be sent successfully
+            assert len(successful_indices) == 5
+            assert successful_indices == [0, 1, 2, 3, 4]
+
+            # All notifications should be removed from batch
+            assert mock_db.remove_notifications_from_batch.call_count == 5
+            assert len(removed_notification_ids) == 5
+            for i in range(5):
+                assert f"notif_{i}" in removed_notification_ids
+
+            # No errors should be logged
+            assert mock_logger.error.call_count == 0
+
+            # Info message about successful sends should be logged
+            info_calls = [call[0][0] for call in mock_logger.info.call_args_list]
+            assert any("sent and removed" in call.lower() for call in info_calls)

autogpt_platform/backend/backend/server/conftest.py

@@ -14,19 +14,49 @@ def configured_snapshot(snapshot: Snapshot) -> Snapshot:
 @pytest.fixture
 def test_user_id() -> str:
     """Test user ID fixture."""
-    return "test-user-id"
+    return "3e53486c-cf57-477e-ba2a-cb02dc828e1a"
 
 
 @pytest.fixture
 def admin_user_id() -> str:
     """Admin user ID fixture."""
-    return "admin-user-id"
+    return "4e53486c-cf57-477e-ba2a-cb02dc828e1b"
 
 
 @pytest.fixture
 def target_user_id() -> str:
     """Target user ID fixture."""
-    return "target-user-id"
+    return "5e53486c-cf57-477e-ba2a-cb02dc828e1c"
+
+
+@pytest.fixture
+async def setup_test_user(test_user_id):
+    """Create test user in database before tests."""
+    from backend.data.user import get_or_create_user
+
+    # Create the test user in the database using JWT token format
+    user_data = {
+        "sub": test_user_id,
+        "email": "test@example.com",
+        "user_metadata": {"name": "Test User"},
+    }
+    await get_or_create_user(user_data)
+    return test_user_id
+
+
+@pytest.fixture
+async def setup_admin_user(admin_user_id):
+    """Create admin user in database before tests."""
+    from backend.data.user import get_or_create_user
+
+    # Create the admin user in the database using JWT token format
+    user_data = {
+        "sub": admin_user_id,
+        "email": "test-admin@example.com",
+        "user_metadata": {"name": "Test Admin"},
+    }
+    await get_or_create_user(user_data)
+    return admin_user_id
 
 
 @pytest.fixture

autogpt_platform/backend/backend/server/integrations/router.py

@@ -64,7 +64,7 @@ class LoginResponse(BaseModel):
     state_token: str
 
 
-@router.get("/{provider}/login")
+@router.get("/{provider}/login", summary="Initiate OAuth flow")
 async def login(
     provider: Annotated[
         ProviderName, Path(title="The provider to initiate an OAuth flow for")
@@ -102,7 +102,7 @@ class CredentialsMetaResponse(BaseModel):
     )
 
 
-@router.post("/{provider}/callback")
+@router.post("/{provider}/callback", summary="Exchange OAuth code for tokens")
 async def callback(
     provider: Annotated[
         ProviderName, Path(title="The target provider for this OAuth exchange")

autogpt_platform/backend/backend/server/rest_api.py

@@ -321,10 +321,6 @@ class AgentServer(backend.util.service.AppProcess):
 
         uvicorn.run(**uvicorn_config)
 
-    def cleanup(self):
-        super().cleanup()
-        logger.info(f"[{self.service_name}] ⏳ Shutting down Agent Server...")
-
     @staticmethod
     async def test_execute_graph(
         graph_id: str,

autogpt_platform/backend/backend/server/routers/v1.py

@@ -11,7 +11,6 @@ import pydantic
 import stripe
 from autogpt_libs.auth import get_user_id, requires_user
 from autogpt_libs.auth.jwt_utils import get_jwt_payload
-from autogpt_libs.utils.cache import cached
 from fastapi import (
     APIRouter,
     Body,
@@ -40,6 +39,7 @@ from backend.data.credit import (
     AutoTopUpConfig,
     RefundRequest,
     TransactionHistory,
+    UserCredit,
     get_auto_top_up,
     get_user_credit_model,
     set_auto_top_up,
@@ -84,6 +84,7 @@ from backend.server.model import (
     UpdateTimezoneRequest,
     UploadFileResponse,
 )
+from backend.util.cache import cached
 from backend.util.clients import get_scheduler_client
 from backend.util.cloud_storage import get_cloud_storage_handler
 from backend.util.exceptions import GraphValidationError, NotFoundError
@@ -107,9 +108,6 @@ def _create_file_size_error(size_bytes: int, max_size_mb: int) -> HTTPException:
 settings = Settings()
 logger = logging.getLogger(__name__)
 
-
-_user_credit_model = get_user_credit_model()
-
 # Define the API routes
 v1_router = APIRouter()
 
@@ -291,7 +289,7 @@ def _compute_blocks_sync() -> str:
     return dumps(result)
 
 
-@cached()
+@cached(ttl_seconds=3600)
 async def _get_cached_blocks() -> str:
     """
     Async cached function with thundering herd protection.
@@ -478,7 +476,8 @@ async def upload_file(
 async def get_user_credits(
     user_id: Annotated[str, Security(get_user_id)],
 ) -> dict[str, int]:
-    return {"credits": await _user_credit_model.get_credits(user_id)}
+    user_credit_model = await get_user_credit_model(user_id)
+    return {"credits": await user_credit_model.get_credits(user_id)}
 
 
 @v1_router.post(
@@ -490,9 +489,8 @@ async def get_user_credits(
 async def request_top_up(
     request: RequestTopUp, user_id: Annotated[str, Security(get_user_id)]
 ):
-    checkout_url = await _user_credit_model.top_up_intent(
-        user_id, request.credit_amount
-    )
+    user_credit_model = await get_user_credit_model(user_id)
+    checkout_url = await user_credit_model.top_up_intent(user_id, request.credit_amount)
     return {"checkout_url": checkout_url}
 
 
@@ -507,7 +505,8 @@ async def refund_top_up(
     transaction_key: str,
     metadata: dict[str, str],
 ) -> int:
-    return await _user_credit_model.top_up_refund(user_id, transaction_key, metadata)
+    user_credit_model = await get_user_credit_model(user_id)
+    return await user_credit_model.top_up_refund(user_id, transaction_key, metadata)
 
 
 @v1_router.patch(
@@ -517,7 +516,8 @@ async def refund_top_up(
     dependencies=[Security(requires_user)],
 )
 async def fulfill_checkout(user_id: Annotated[str, Security(get_user_id)]):
-    await _user_credit_model.fulfill_checkout(user_id=user_id)
+    user_credit_model = await get_user_credit_model(user_id)
+    await user_credit_model.fulfill_checkout(user_id=user_id)
     return Response(status_code=200)
 
 
@@ -531,18 +531,23 @@ async def configure_user_auto_top_up(
     request: AutoTopUpConfig, user_id: Annotated[str, Security(get_user_id)]
 ) -> str:
     if request.threshold < 0:
-        raise ValueError("Threshold must be greater than 0")
+        raise HTTPException(status_code=422, detail="Threshold must be greater than 0")
     if request.amount < 500 and request.amount != 0:
-        raise ValueError("Amount must be greater than or equal to 500")
-    if request.amount < request.threshold:
-        raise ValueError("Amount must be greater than or equal to threshold")
+        raise HTTPException(
+            status_code=422, detail="Amount must be greater than or equal to 500"
+        )
+    if request.amount != 0 and request.amount < request.threshold:
+        raise HTTPException(
+            status_code=422, detail="Amount must be greater than or equal to threshold"
+        )
 
-    current_balance = await _user_credit_model.get_credits(user_id)
+    user_credit_model = await get_user_credit_model(user_id)
+    current_balance = await user_credit_model.get_credits(user_id)
 
     if current_balance < request.threshold:
-        await _user_credit_model.top_up_credits(user_id, request.amount)
+        await user_credit_model.top_up_credits(user_id, request.amount)
     else:
-        await _user_credit_model.top_up_credits(user_id, 0)
+        await user_credit_model.top_up_credits(user_id, 0)
 
     await set_auto_top_up(
         user_id, AutoTopUpConfig(threshold=request.threshold, amount=request.amount)
@@ -590,15 +595,13 @@ async def stripe_webhook(request: Request):
         event["type"] == "checkout.session.completed"
         or event["type"] == "checkout.session.async_payment_succeeded"
     ):
-        await _user_credit_model.fulfill_checkout(
-            session_id=event["data"]["object"]["id"]
-        )
+        await UserCredit().fulfill_checkout(session_id=event["data"]["object"]["id"])
 
     if event["type"] == "charge.dispute.created":
-        await _user_credit_model.handle_dispute(event["data"]["object"])
+        await UserCredit().handle_dispute(event["data"]["object"])
 
     if event["type"] == "refund.created" or event["type"] == "charge.dispute.closed":
-        await _user_credit_model.deduct_credits(event["data"]["object"])
+        await UserCredit().deduct_credits(event["data"]["object"])
 
     return Response(status_code=200)
 
@@ -612,7 +615,8 @@ async def stripe_webhook(request: Request):
 async def manage_payment_method(
     user_id: Annotated[str, Security(get_user_id)],
 ) -> dict[str, str]:
-    return {"url": await _user_credit_model.create_billing_portal_session(user_id)}
+    user_credit_model = await get_user_credit_model(user_id)
+    return {"url": await user_credit_model.create_billing_portal_session(user_id)}
 
 
 @v1_router.get(
@@ -630,7 +634,8 @@ async def get_credit_history(
     if transaction_count_limit < 1 or transaction_count_limit > 1000:
         raise ValueError("Transaction count limit must be between 1 and 1000")
 
-    return await _user_credit_model.get_transaction_history(
+    user_credit_model = await get_user_credit_model(user_id)
+    return await user_credit_model.get_transaction_history(
         user_id=user_id,
         transaction_time_ceiling=transaction_time,
         transaction_count_limit=transaction_count_limit,
@@ -647,7 +652,8 @@ async def get_credit_history(
 async def get_refund_requests(
     user_id: Annotated[str, Security(get_user_id)],
 ) -> list[RefundRequest]:
-    return await _user_credit_model.get_refund_requests(user_id)
+    user_credit_model = await get_user_credit_model(user_id)
+    return await user_credit_model.get_refund_requests(user_id)
 
 
 ########################################################
@@ -869,7 +875,8 @@ async def execute_graph(
     graph_version: Optional[int] = None,
     preset_id: Optional[str] = None,
 ) -> execution_db.GraphExecutionMeta:
-    current_balance = await _user_credit_model.get_credits(user_id)
+    user_credit_model = await get_user_credit_model(user_id)
+    current_balance = await user_credit_model.get_credits(user_id)
     if current_balance <= 0:
         raise HTTPException(
             status_code=402,

autogpt_platform/backend/backend/server/routers/v1_test.py

@@ -23,10 +23,13 @@ client = fastapi.testclient.TestClient(app)
 
 
 @pytest.fixture(autouse=True)
-def setup_app_auth(mock_jwt_user):
+def setup_app_auth(mock_jwt_user, setup_test_user):
     """Setup auth overrides for all tests in this module"""
     from autogpt_libs.auth.jwt_utils import get_jwt_payload
 
+    # setup_test_user fixture already executed and user is created in database
+    # It returns the user_id which we don't need to await
+
     app.dependency_overrides[get_jwt_payload] = mock_jwt_user["get_jwt_payload"]
     yield
     app.dependency_overrides.clear()
@@ -194,8 +197,12 @@ def test_get_user_credits(
     snapshot: Snapshot,
 ) -> None:
     """Test get user credits endpoint"""
-    mock_credit_model = mocker.patch("backend.server.routers.v1._user_credit_model")
+    mock_credit_model = Mock()
     mock_credit_model.get_credits = AsyncMock(return_value=1000)
+    mocker.patch(
+        "backend.server.routers.v1.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
 
     response = client.get("/credits")
 
@@ -215,10 +222,14 @@ def test_request_top_up(
     snapshot: Snapshot,
 ) -> None:
     """Test request top up endpoint"""
-    mock_credit_model = mocker.patch("backend.server.routers.v1._user_credit_model")
+    mock_credit_model = Mock()
     mock_credit_model.top_up_intent = AsyncMock(
         return_value="https://checkout.example.com/session123"
     )
+    mocker.patch(
+        "backend.server.routers.v1.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
 
     request_data = {"credit_amount": 500}
 
@@ -261,6 +272,74 @@ def test_get_auto_top_up(
     )
 
 
+def test_configure_auto_top_up(
+    mocker: pytest_mock.MockFixture,
+    snapshot: Snapshot,
+) -> None:
+    """Test configure auto top-up endpoint - this test would have caught the enum casting bug"""
+    # Mock the set_auto_top_up function to avoid database operations
+    mocker.patch(
+        "backend.server.routers.v1.set_auto_top_up",
+        return_value=None,
+    )
+
+    # Mock credit model to avoid Stripe API calls
+    mock_credit_model = mocker.AsyncMock()
+    mock_credit_model.get_credits.return_value = 50  # Current balance below threshold
+    mock_credit_model.top_up_credits.return_value = None
+
+    mocker.patch(
+        "backend.server.routers.v1.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
+
+    # Test data
+    request_data = {
+        "threshold": 100,
+        "amount": 500,
+    }
+
+    response = client.post("/credits/auto-top-up", json=request_data)
+
+    # This should succeed with our fix, but would have failed before with the enum casting error
+    assert response.status_code == 200
+    assert response.json() == "Auto top-up settings updated"
+
+
+def test_configure_auto_top_up_validation_errors(
+    mocker: pytest_mock.MockFixture,
+) -> None:
+    """Test configure auto top-up endpoint validation"""
+    # Mock set_auto_top_up to avoid database operations for successful case
+    mocker.patch("backend.server.routers.v1.set_auto_top_up")
+
+    # Mock credit model to avoid Stripe API calls for the successful case
+    mock_credit_model = mocker.AsyncMock()
+    mock_credit_model.get_credits.return_value = 50
+    mock_credit_model.top_up_credits.return_value = None
+
+    mocker.patch(
+        "backend.server.routers.v1.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
+
+    # Test negative threshold
+    response = client.post(
+        "/credits/auto-top-up", json={"threshold": -1, "amount": 500}
+    )
+    assert response.status_code == 422  # Validation error
+
+    # Test amount too small (but not 0)
+    response = client.post(
+        "/credits/auto-top-up", json={"threshold": 100, "amount": 100}
+    )
+    assert response.status_code == 422  # Validation error
+
+    # Test amount = 0 (should be allowed)
+    response = client.post("/credits/auto-top-up", json={"threshold": 100, "amount": 0})
+    assert response.status_code == 200  # Should succeed
+
+
 # Graphs endpoints tests
 def test_get_graphs(
     mocker: pytest_mock.MockFixture,

autogpt_platform/backend/backend/server/v2/admin/credit_admin_routes.py

@@ -11,8 +11,6 @@ from backend.util.json import SafeJson
 
 logger = logging.getLogger(__name__)
 
-_user_credit_model = get_user_credit_model()
-
 
 router = APIRouter(
     prefix="/admin",
@@ -33,7 +31,8 @@ async def add_user_credits(
     logger.info(
         f"Admin user {admin_user_id} is adding {amount} credits to user {user_id}"
     )
-    new_balance, transaction_key = await _user_credit_model._add_transaction(
+    user_credit_model = await get_user_credit_model(user_id)
+    new_balance, transaction_key = await user_credit_model._add_transaction(
         user_id,
         amount,
         transaction_type=CreditTransactionType.GRANT,

autogpt_platform/backend/backend/server/v2/admin/credit_admin_routes_test.py

@@ -1,5 +1,5 @@
 import json
-from unittest.mock import AsyncMock
+from unittest.mock import AsyncMock, Mock
 
 import fastapi
 import fastapi.testclient
@@ -7,12 +7,12 @@ import prisma.enums
 import pytest
 import pytest_mock
 from autogpt_libs.auth.jwt_utils import get_jwt_payload
-from prisma import Json
 from pytest_snapshot.plugin import Snapshot
 
 import backend.server.v2.admin.credit_admin_routes as credit_admin_routes
 import backend.server.v2.admin.model as admin_model
 from backend.data.model import UserTransaction
+from backend.util.json import SafeJson
 from backend.util.models import Pagination
 
 app = fastapi.FastAPI()
@@ -37,12 +37,14 @@ def test_add_user_credits_success(
 ) -> None:
     """Test successful credit addition by admin"""
     # Mock the credit model
-    mock_credit_model = mocker.patch(
-        "backend.server.v2.admin.credit_admin_routes._user_credit_model"
-    )
+    mock_credit_model = Mock()
     mock_credit_model._add_transaction = AsyncMock(
         return_value=(1500, "transaction-123-uuid")
     )
+    mocker.patch(
+        "backend.server.v2.admin.credit_admin_routes.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
 
     request_data = {
         "user_id": target_user_id,
@@ -62,11 +64,17 @@ def test_add_user_credits_success(
     call_args = mock_credit_model._add_transaction.call_args
     assert call_args[0] == (target_user_id, 500)
     assert call_args[1]["transaction_type"] == prisma.enums.CreditTransactionType.GRANT
-    # Check that metadata is a Json object with the expected content
-    assert isinstance(call_args[1]["metadata"], Json)
-    assert call_args[1]["metadata"] == Json(
-        {"admin_id": admin_user_id, "reason": "Test credit grant for debugging"}
-    )
+    # Check that metadata is a SafeJson object with the expected content
+    assert isinstance(call_args[1]["metadata"], SafeJson)
+    actual_metadata = call_args[1]["metadata"]
+    expected_data = {
+        "admin_id": admin_user_id,
+        "reason": "Test credit grant for debugging",
+    }
+
+    # SafeJson inherits from Json which stores parsed data in the .data attribute
+    assert actual_metadata.data["admin_id"] == expected_data["admin_id"]
+    assert actual_metadata.data["reason"] == expected_data["reason"]
 
     # Snapshot test the response
     configured_snapshot.assert_match(
@@ -81,12 +89,14 @@ def test_add_user_credits_negative_amount(
 ) -> None:
     """Test credit deduction by admin (negative amount)"""
     # Mock the credit model
-    mock_credit_model = mocker.patch(
-        "backend.server.v2.admin.credit_admin_routes._user_credit_model"
-    )
+    mock_credit_model = Mock()
     mock_credit_model._add_transaction = AsyncMock(
         return_value=(200, "transaction-456-uuid")
     )
+    mocker.patch(
+        "backend.server.v2.admin.credit_admin_routes.get_user_credit_model",
+        return_value=mock_credit_model,
+    )
 
     request_data = {
         "user_id": "target-user-id",

autogpt_platform/backend/backend/server/v2/admin/store_admin_routes.py

@@ -7,6 +7,7 @@ import fastapi
 import fastapi.responses
 import prisma.enums
 
+import backend.server.v2.store.cache as store_cache
 import backend.server.v2.store.db
 import backend.server.v2.store.model
 import backend.util.json
@@ -86,6 +87,11 @@ async def review_submission(
         StoreSubmission with updated review information
     """
     try:
+        already_approved = (
+            await backend.server.v2.store.db.check_submission_already_approved(
+                store_listing_version_id=store_listing_version_id,
+            )
+        )
         submission = await backend.server.v2.store.db.review_store_submission(
             store_listing_version_id=store_listing_version_id,
             is_approved=request.is_approved,
@@ -93,6 +99,11 @@ async def review_submission(
             internal_comments=request.internal_comments or "",
             reviewer_id=user_id,
         )
+
+        state_changed = already_approved != request.is_approved
+        # Clear caches when the request is approved as it updates what is shown on the store
+        if state_changed:
+            store_cache.clear_all_caches()
         return submission
     except Exception as e:
         logger.exception("Error reviewing submission: %s", e)

autogpt_platform/backend/backend/server/v2/builder/db.py

@@ -2,7 +2,6 @@ import logging
 from datetime import datetime, timedelta, timezone
 
 import prisma
-from autogpt_libs.utils.cache import cached
 
 import backend.data.block
 from backend.blocks import load_all_blocks
@@ -18,6 +17,7 @@ from backend.server.v2.builder.model import (
     ProviderResponse,
     SearchBlocksResponse,
 )
+from backend.util.cache import cached
 from backend.util.models import Pagination
 
 logger = logging.getLogger(__name__)
@@ -118,6 +118,17 @@ def get_blocks(
     )
 
 
+def get_block_by_id(block_id: str) -> BlockInfo | None:
+    """
+    Get a specific block by its ID.
+    """
+    for block_type in load_all_blocks().values():
+        block: Block[BlockSchema, BlockSchema] = block_type()
+        if block.id == block_id:
+            return block.get_info()
+    return None
+
+
 def search_blocks(
     include_blocks: bool = True,
     include_integrations: bool = True,
@@ -296,7 +307,7 @@ def _matches_llm_model(schema_cls: type[BlockSchema], query: str) -> bool:
     return False
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def _get_all_providers() -> dict[ProviderName, Provider]:
     providers: dict[ProviderName, Provider] = {}
 

autogpt_platform/backend/backend/server/v2/builder/model.py

@@ -53,16 +53,6 @@ class ProviderResponse(BaseModel):
     pagination: Pagination
 
 
-# Search
-class SearchRequest(BaseModel):
-    search_query: str | None = None
-    filter: list[FilterType] | None = None
-    by_creator: list[str] | None = None
-    search_id: str | None = None
-    page: int | None = None
-    page_size: int | None = None
-
-
 class SearchBlocksResponse(BaseModel):
     blocks: BlockResponse
     total_block_count: int

autogpt_platform/backend/backend/server/v2/builder/routes.py

@@ -110,6 +110,25 @@ async def get_blocks(
     )
 
 
+@router.get(
+    "/blocks/batch",
+    summary="Get specific blocks",
+    response_model=list[builder_model.BlockInfo],
+)
+async def get_specific_blocks(
+    block_ids: Annotated[list[str], fastapi.Query()],
+) -> list[builder_model.BlockInfo]:
+    """
+    Get specific blocks by their IDs.
+    """
+    blocks = []
+    for block_id in block_ids:
+        block = builder_db.get_block_by_id(block_id)
+        if block:
+            blocks.append(block)
+    return blocks
+
+
 @router.get(
     "/providers",
     summary="Get Builder integration providers",
@@ -128,30 +147,34 @@ async def get_providers(
     )
 
 
-@router.post(
+# Not using post method because on frontend, orval doesn't support Infinite Query with POST method.
+@router.get(
     "/search",
     summary="Builder search",
     tags=["store", "private"],
     response_model=builder_model.SearchResponse,
 )
 async def search(
-    options: builder_model.SearchRequest,
     user_id: Annotated[str, fastapi.Security(get_user_id)],
+    search_query: Annotated[str | None, fastapi.Query()] = None,
+    filter: Annotated[list[str] | None, fastapi.Query()] = None,
+    search_id: Annotated[str | None, fastapi.Query()] = None,
+    by_creator: Annotated[list[str] | None, fastapi.Query()] = None,
+    page: Annotated[int, fastapi.Query()] = 1,
+    page_size: Annotated[int, fastapi.Query()] = 50,
 ) -> builder_model.SearchResponse:
     """
     Search for blocks (including integrations), marketplace agents, and user library agents.
     """
     # If no filters are provided, then we will return all types
-    if not options.filter:
-        options.filter = [
+    if not filter:
+        filter = [
             "blocks",
             "integrations",
             "marketplace_agents",
             "my_agents",
         ]
-    options.search_query = sanitize_query(options.search_query)
-    options.page = options.page or 1
-    options.page_size = options.page_size or 50
+    search_query = sanitize_query(search_query)
 
     # Blocks&Integrations
     blocks = builder_model.SearchBlocksResponse(
@@ -162,13 +185,13 @@ async def search(
         total_block_count=0,
         total_integration_count=0,
     )
-    if "blocks" in options.filter or "integrations" in options.filter:
+    if "blocks" in filter or "integrations" in filter:
         blocks = builder_db.search_blocks(
-            include_blocks="blocks" in options.filter,
-            include_integrations="integrations" in options.filter,
-            query=options.search_query or "",
-            page=options.page,
-            page_size=options.page_size,
+            include_blocks="blocks" in filter,
+            include_integrations="integrations" in filter,
+            query=search_query or "",
+            page=page,
+            page_size=page_size,
         )
 
     # Library Agents
@@ -176,12 +199,12 @@ async def search(
         agents=[],
         pagination=Pagination.empty(),
     )
-    if "my_agents" in options.filter:
+    if "my_agents" in filter:
         my_agents = await library_db.list_library_agents(
             user_id=user_id,
-            search_term=options.search_query,
-            page=options.page,
-            page_size=options.page_size,
+            search_term=search_query,
+            page=page,
+            page_size=page_size,
         )
 
     # Marketplace Agents
@@ -189,12 +212,12 @@ async def search(
         agents=[],
         pagination=Pagination.empty(),
     )
-    if "marketplace_agents" in options.filter:
+    if "marketplace_agents" in filter:
         marketplace_agents = await store_db.get_store_agents(
-            creators=options.by_creator,
-            search_query=options.search_query,
-            page=options.page,
-            page_size=options.page_size,
+            creators=by_creator,
+            search_query=search_query,
+            page=page,
+            page_size=page_size,
         )
 
     more_pages = False
@@ -214,7 +237,7 @@ async def search(
             "marketplace_agents": marketplace_agents.pagination.total_items,
             "my_agents": my_agents.pagination.total_items,
         },
-        page=options.page,
+        page=page,
         more_pages=more_pages,
     )
 

autogpt_platform/backend/backend/server/v2/library/db.py

@@ -20,7 +20,7 @@ from backend.data.includes import AGENT_PRESET_INCLUDE, library_agent_include
 from backend.data.model import CredentialsMetaInput
 from backend.integrations.creds_manager import IntegrationCredentialsManager
 from backend.integrations.webhooks.graph_lifecycle_hooks import on_graph_activate
-from backend.util.exceptions import NotFoundError
+from backend.util.exceptions import DatabaseError, NotFoundError
 from backend.util.json import SafeJson
 from backend.util.models import Pagination
 from backend.util.settings import Config
@@ -61,11 +61,11 @@ async def list_library_agents(
 
     if page < 1 or page_size < 1:
         logger.warning(f"Invalid pagination: page={page}, page_size={page_size}")
-        raise store_exceptions.DatabaseError("Invalid pagination input")
+        raise DatabaseError("Invalid pagination input")
 
     if search_term and len(search_term.strip()) > 100:
         logger.warning(f"Search term too long: {repr(search_term)}")
-        raise store_exceptions.DatabaseError("Search term is too long")
+        raise DatabaseError("Search term is too long")
 
     where_clause: prisma.types.LibraryAgentWhereInput = {
         "userId": user_id,
@@ -143,7 +143,7 @@ async def list_library_agents(
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error fetching library agents: {e}")
-        raise store_exceptions.DatabaseError("Failed to fetch library agents") from e
+        raise DatabaseError("Failed to fetch library agents") from e
 
 
 async def list_favorite_library_agents(
@@ -172,7 +172,7 @@ async def list_favorite_library_agents(
 
     if page < 1 or page_size < 1:
         logger.warning(f"Invalid pagination: page={page}, page_size={page_size}")
-        raise store_exceptions.DatabaseError("Invalid pagination input")
+        raise DatabaseError("Invalid pagination input")
 
     where_clause: prisma.types.LibraryAgentWhereInput = {
         "userId": user_id,
@@ -229,9 +229,7 @@ async def list_favorite_library_agents(
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error fetching favorite library agents: {e}")
-        raise store_exceptions.DatabaseError(
-            "Failed to fetch favorite library agents"
-        ) from e
+        raise DatabaseError("Failed to fetch favorite library agents") from e
 
 
 async def get_library_agent(id: str, user_id: str) -> library_model.LibraryAgent:
@@ -273,7 +271,7 @@ async def get_library_agent(id: str, user_id: str) -> library_model.LibraryAgent
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error fetching library agent: {e}")
-        raise store_exceptions.DatabaseError("Failed to fetch library agent") from e
+        raise DatabaseError("Failed to fetch library agent") from e
 
 
 async def get_library_agent_by_store_version_id(
@@ -338,7 +336,7 @@ async def get_library_agent_by_graph_id(
         return library_model.LibraryAgent.from_db(agent, sub_graphs=sub_graphs)
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error fetching library agent by graph ID: {e}")
-        raise store_exceptions.DatabaseError("Failed to fetch library agent") from e
+        raise DatabaseError("Failed to fetch library agent") from e
 
 
 async def add_generated_agent_image(
@@ -479,9 +477,7 @@ async def update_agent_version_in_library(
         )
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error updating agent version in library: {e}")
-        raise store_exceptions.DatabaseError(
-            "Failed to update agent version in library"
-        ) from e
+        raise DatabaseError("Failed to update agent version in library") from e
 
 
 async def update_library_agent(
@@ -544,7 +540,7 @@ async def update_library_agent(
         )
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error updating library agent: {str(e)}")
-        raise store_exceptions.DatabaseError("Failed to update library agent") from e
+        raise DatabaseError("Failed to update library agent") from e
 
 
 async def delete_library_agent(
@@ -572,7 +568,7 @@ async def delete_library_agent_by_graph_id(graph_id: str, user_id: str) -> None:
         )
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error deleting library agent: {e}")
-        raise store_exceptions.DatabaseError("Failed to delete library agent") from e
+        raise DatabaseError("Failed to delete library agent") from e
 
 
 async def add_store_agent_to_library(
@@ -663,7 +659,7 @@ async def add_store_agent_to_library(
         raise
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error adding agent to library: {e}")
-        raise store_exceptions.DatabaseError("Failed to add agent to library") from e
+        raise DatabaseError("Failed to add agent to library") from e
 
 
 ##############################################
@@ -697,7 +693,7 @@ async def list_presets(
         logger.warning(
             "Invalid pagination input: page=%d, page_size=%d", page, page_size
         )
-        raise store_exceptions.DatabaseError("Invalid pagination parameters")
+        raise DatabaseError("Invalid pagination parameters")
 
     query_filter: prisma.types.AgentPresetWhereInput = {
         "userId": user_id,
@@ -733,7 +729,7 @@ async def list_presets(
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error getting presets: {e}")
-        raise store_exceptions.DatabaseError("Failed to fetch presets") from e
+        raise DatabaseError("Failed to fetch presets") from e
 
 
 async def get_preset(
@@ -763,7 +759,7 @@ async def get_preset(
         return library_model.LibraryAgentPreset.from_db(preset)
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error getting preset: {e}")
-        raise store_exceptions.DatabaseError("Failed to fetch preset") from e
+        raise DatabaseError("Failed to fetch preset") from e
 
 
 async def create_preset(
@@ -813,7 +809,7 @@ async def create_preset(
         return library_model.LibraryAgentPreset.from_db(new_preset)
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error creating preset: {e}")
-        raise store_exceptions.DatabaseError("Failed to create preset") from e
+        raise DatabaseError("Failed to create preset") from e
 
 
 async def create_preset_from_graph_execution(
@@ -951,7 +947,7 @@ async def update_preset(
         return library_model.LibraryAgentPreset.from_db(updated)
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error updating preset: {e}")
-        raise store_exceptions.DatabaseError("Failed to update preset") from e
+        raise DatabaseError("Failed to update preset") from e
 
 
 async def set_preset_webhook(
@@ -997,7 +993,7 @@ async def delete_preset(user_id: str, preset_id: str) -> None:
         )
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error deleting preset: {e}")
-        raise store_exceptions.DatabaseError("Failed to delete preset") from e
+        raise DatabaseError("Failed to delete preset") from e
 
 
 async def fork_library_agent(
@@ -1025,7 +1021,7 @@ async def fork_library_agent(
         # TODO: once we have open/closed sourced agents this needs to be enabled ~kcze
         # + update library/agents/[id]/page.tsx agent actions
         # if not original_agent.can_access_graph:
-        #     raise store_exceptions.DatabaseError(
+        #     raise DatabaseError(
         #         f"User {user_id} cannot access library agent graph {library_agent_id}"
         #     )
 
@@ -1039,4 +1035,4 @@ async def fork_library_agent(
         return (await create_library_agent(new_graph, user_id))[0]
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error cloning library agent: {e}")
-        raise store_exceptions.DatabaseError("Failed to fork library agent") from e
+        raise DatabaseError("Failed to fork library agent") from e

autogpt_platform/backend/backend/server/v2/library/routes/agents.py

@@ -8,7 +8,7 @@ from fastapi.responses import Response
 import backend.server.v2.library.db as library_db
 import backend.server.v2.library.model as library_model
 import backend.server.v2.store.exceptions as store_exceptions
-from backend.util.exceptions import NotFoundError
+from backend.util.exceptions import DatabaseError, NotFoundError
 
 logger = logging.getLogger(__name__)
 
@@ -221,7 +221,7 @@ async def add_marketplace_agent_to_library(
             "to add to library"
         )
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(e))
-    except store_exceptions.DatabaseError as e:
+    except DatabaseError as e:
         logger.error(f"Database error while adding agent to library: {e}", e)
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
@@ -275,7 +275,7 @@ async def update_library_agent(
             status_code=status.HTTP_404_NOT_FOUND,
             detail=str(e),
         ) from e
-    except store_exceptions.DatabaseError as e:
+    except DatabaseError as e:
         logger.error(f"Database error while updating library agent: {e}")
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,

autogpt_platform/backend/backend/server/v2/store/cache.py

@@ -0,0 +1,75 @@
+import backend.server.v2.store.db
+from backend.util.cache import cached
+
+##############################################
+############### Caches #######################
+##############################################
+
+
+def clear_all_caches():
+    """Clear all caches."""
+    _get_cached_store_agents.cache_clear()
+    _get_cached_agent_details.cache_clear()
+    _get_cached_store_creators.cache_clear()
+    _get_cached_creator_details.cache_clear()
+
+
+# Cache store agents list for 5 minutes
+# Different cache entries for different query combinations
+@cached(maxsize=5000, ttl_seconds=300, shared_cache=True)
+async def _get_cached_store_agents(
+    featured: bool,
+    creator: str | None,
+    sorted_by: str | None,
+    search_query: str | None,
+    category: str | None,
+    page: int,
+    page_size: int,
+):
+    """Cached helper to get store agents."""
+    return await backend.server.v2.store.db.get_store_agents(
+        featured=featured,
+        creators=[creator] if creator else None,
+        sorted_by=sorted_by,
+        search_query=search_query,
+        category=category,
+        page=page,
+        page_size=page_size,
+    )
+
+
+# Cache individual agent details for 15 minutes
+@cached(maxsize=200, ttl_seconds=300, shared_cache=True)
+async def _get_cached_agent_details(username: str, agent_name: str):
+    """Cached helper to get agent details."""
+    return await backend.server.v2.store.db.get_store_agent_details(
+        username=username, agent_name=agent_name
+    )
+
+
+# Cache creators list for 5 minutes
+@cached(maxsize=200, ttl_seconds=300, shared_cache=True)
+async def _get_cached_store_creators(
+    featured: bool,
+    search_query: str | None,
+    sorted_by: str | None,
+    page: int,
+    page_size: int,
+):
+    """Cached helper to get store creators."""
+    return await backend.server.v2.store.db.get_store_creators(
+        featured=featured,
+        search_query=search_query,
+        sorted_by=sorted_by,
+        page=page,
+        page_size=page_size,
+    )
+
+
+# Cache individual creator details for 5 minutes
+@cached(maxsize=100, ttl_seconds=300, shared_cache=True)
+async def _get_cached_creator_details(username: str):
+    """Cached helper to get creator details."""
+    return await backend.server.v2.store.db.get_store_creator_details(
+        username=username.lower()
+    )

autogpt_platform/backend/backend/server/v2/store/db.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import typing
 from datetime import datetime, timezone
 
 import fastapi
@@ -25,6 +26,7 @@ from backend.data.notifications import (
     NotificationEventModel,
 )
 from backend.notifications.notifications import queue_notification_async
+from backend.util.exceptions import DatabaseError
 from backend.util.settings import Settings
 
 logger = logging.getLogger(__name__)
@@ -70,65 +72,199 @@ async def get_store_agents(
     logger.debug(
         f"Getting store agents. featured={featured}, creators={creators}, sorted_by={sorted_by}, search={search_query}, category={category}, page={page}"
     )
-    sanitized_query = sanitize_query(search_query)
 
-    where_clause: prisma.types.StoreAgentWhereInput = {"is_available": True}
-    if featured:
-        where_clause["featured"] = featured
+    sanitized_creators = []
     if creators:
-        where_clause["creator_username"] = {"in": creators}
+        for c in creators:
+            sanitized_creators.append(sanitize_query(c))
+
+    sanitized_category = None
     if category:
-        where_clause["categories"] = {"has": category}
-
-    if sanitized_query:
-        where_clause["OR"] = [
-            {"agent_name": {"contains": sanitized_query, "mode": "insensitive"}},
-            {"description": {"contains": sanitized_query, "mode": "insensitive"}},
-        ]
-
-    order_by = []
-    if sorted_by == "rating":
-        order_by.append({"rating": "desc"})
-    elif sorted_by == "runs":
-        order_by.append({"runs": "desc"})
-    elif sorted_by == "name":
-        order_by.append({"agent_name": "asc"})
+        sanitized_category = sanitize_query(category)
 
     try:
-        agents = await prisma.models.StoreAgent.prisma().find_many(
-            where=where_clause,
-            order=order_by,
-            skip=(page - 1) * page_size,
-            take=page_size,
-        )
-
-        total = await prisma.models.StoreAgent.prisma().count(where=where_clause)
-        total_pages = (total + page_size - 1) // page_size
-
-        store_agents: list[backend.server.v2.store.model.StoreAgent] = []
-        for agent in agents:
-            try:
-                # Create the StoreAgent object safely
-                store_agent = backend.server.v2.store.model.StoreAgent(
-                    slug=agent.slug,
-                    agent_name=agent.agent_name,
-                    agent_image=agent.agent_image[0] if agent.agent_image else "",
-                    creator=agent.creator_username or "Needs Profile",
-                    creator_avatar=agent.creator_avatar or "",
-                    sub_heading=agent.sub_heading,
-                    description=agent.description,
-                    runs=agent.runs,
-                    rating=agent.rating,
+        # If search_query is provided, use full-text search
+        if search_query:
+            search_term = sanitize_query(search_query)
+            if not search_term:
+                # Return empty results for invalid search query
+                return backend.server.v2.store.model.StoreAgentsResponse(
+                    agents=[],
+                    pagination=backend.server.v2.store.model.Pagination(
+                        current_page=page,
+                        total_items=0,
+                        total_pages=0,
+                        page_size=page_size,
+                    ),
                 )
-                # Add to the list only if creation was successful
-                store_agents.append(store_agent)
-            except Exception as e:
-                # Skip this agent if there was an error
-                # You could log the error here if needed
-                logger.error(
-                    f"Error parsing Store agent when getting store agents from db: {e}"
-                )
-                continue
+
+            offset = (page - 1) * page_size
+
+            # Whitelist allowed order_by columns
+            ALLOWED_ORDER_BY = {
+                "rating": "rating DESC, rank DESC",
+                "runs": "runs DESC, rank DESC",
+                "name": "agent_name ASC, rank DESC",
+                "updated_at": "updated_at DESC, rank DESC",
+            }
+
+            # Validate and get order clause
+            if sorted_by and sorted_by in ALLOWED_ORDER_BY:
+                order_by_clause = ALLOWED_ORDER_BY[sorted_by]
+            else:
+                order_by_clause = "updated_at DESC, rank DESC",
+
+            # Build WHERE conditions and parameters list
+            where_parts: list[str] = []
+            params: list[typing.Any] = [search_term]  # $1 - search term
+            param_index = 2  # Start at $2 for next parameter
+
+            # Always filter for available agents
+            where_parts.append("is_available = true")
+
+            if featured:
+                where_parts.append("featured = true")
+
+            if creators and sanitized_creators:
+                # Use ANY with array parameter
+                where_parts.append(f"creator_username = ANY(${param_index})")
+                params.append(sanitized_creators)
+                param_index += 1
+
+            if category and sanitized_category:
+                where_parts.append(f"${param_index} = ANY(categories)")
+                params.append(sanitized_category)
+                param_index += 1
+
+            sql_where_clause: str = " AND ".join(where_parts) if where_parts else "1=1"
+
+            # Add pagination params
+            params.extend([page_size, offset])
+            limit_param = f"${param_index}"
+            offset_param = f"${param_index + 1}"
+
+            # Execute full-text search query with parameterized values
+            sql_query = f"""
+                SELECT
+                    slug,
+                    agent_name,
+                    agent_image,
+                    creator_username,
+                    creator_avatar,
+                    sub_heading,
+                    description,
+                    runs,
+                    rating,
+                    categories,
+                    featured,
+                    is_available,
+                    updated_at,
+                    ts_rank_cd(search, query) AS rank
+                FROM "StoreAgent",
+                    plainto_tsquery('english', $1) AS query
+                WHERE {sql_where_clause}
+                    AND search @@ query
+                ORDER BY {order_by_clause}
+                LIMIT {limit_param} OFFSET {offset_param}
+            """
+
+            # Count query for pagination - only uses search term parameter
+            count_query = f"""
+                SELECT COUNT(*) as count
+                FROM "StoreAgent",
+                    plainto_tsquery('english', $1) AS query
+                WHERE {sql_where_clause}
+                    AND search @@ query
+            """
+
+            # Execute both queries with parameters
+            agents = await prisma.client.get_client().query_raw(
+                typing.cast(typing.LiteralString, sql_query), *params
+            )
+
+            # For count, use params without pagination (last 2 params)
+            count_params = params[:-2]
+            count_result = await prisma.client.get_client().query_raw(
+                typing.cast(typing.LiteralString, count_query), *count_params
+            )
+
+            total = count_result[0]["count"] if count_result else 0
+            total_pages = (total + page_size - 1) // page_size
+
+            # Convert raw results to StoreAgent models
+            store_agents: list[backend.server.v2.store.model.StoreAgent] = []
+            for agent in agents:
+                try:
+                    store_agent = backend.server.v2.store.model.StoreAgent(
+                        slug=agent["slug"],
+                        agent_name=agent["agent_name"],
+                        agent_image=(
+                            agent["agent_image"][0] if agent["agent_image"] else ""
+                        ),
+                        creator=agent["creator_username"] or "Needs Profile",
+                        creator_avatar=agent["creator_avatar"] or "",
+                        sub_heading=agent["sub_heading"],
+                        description=agent["description"],
+                        runs=agent["runs"],
+                        rating=agent["rating"],
+                    )
+                    store_agents.append(store_agent)
+                except Exception as e:
+                    logger.error(f"Error parsing Store agent from search results: {e}")
+                    continue
+
+        else:
+            # Non-search query path (original logic)
+            where_clause: prisma.types.StoreAgentWhereInput = {"is_available": True}
+            if featured:
+                where_clause["featured"] = featured
+            if creators:
+                where_clause["creator_username"] = {"in": sanitized_creators}
+            if sanitized_category:
+                where_clause["categories"] = {"has": sanitized_category}
+
+            order_by = []
+            if sorted_by == "rating":
+                order_by.append({"rating": "desc"})
+            elif sorted_by == "runs":
+                order_by.append({"runs": "desc"})
+            elif sorted_by == "name":
+                order_by.append({"agent_name": "asc"})
+
+            agents = await prisma.models.StoreAgent.prisma().find_many(
+                where=where_clause,
+                order=order_by,
+                skip=(page - 1) * page_size,
+                take=page_size,
+            )
+
+            total = await prisma.models.StoreAgent.prisma().count(where=where_clause)
+            total_pages = (total + page_size - 1) // page_size
+
+            store_agents: list[backend.server.v2.store.model.StoreAgent] = []
+            for agent in agents:
+                try:
+                    # Create the StoreAgent object safely
+                    store_agent = backend.server.v2.store.model.StoreAgent(
+                        slug=agent.slug,
+                        agent_name=agent.agent_name,
+                        agent_image=agent.agent_image[0] if agent.agent_image else "",
+                        creator=agent.creator_username or "Needs Profile",
+                        creator_avatar=agent.creator_avatar or "",
+                        sub_heading=agent.sub_heading,
+                        description=agent.description,
+                        runs=agent.runs,
+                        rating=agent.rating,
+                    )
+                    # Add to the list only if creation was successful
+                    store_agents.append(store_agent)
+                except Exception as e:
+                    # Skip this agent if there was an error
+                    # You could log the error here if needed
+                    logger.error(
+                        f"Error parsing Store agent when getting store agents from db: {e}"
+                    )
+                    continue
 
         logger.debug(f"Found {len(store_agents)} agents")
         return backend.server.v2.store.model.StoreAgentsResponse(
@@ -142,9 +278,25 @@ async def get_store_agents(
         )
     except Exception as e:
         logger.error(f"Error getting store agents: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch store agents"
-        ) from e
+        raise DatabaseError("Failed to fetch store agents") from e
+    # TODO: commenting this out as we concerned about potential db load issues
+    # finally:
+    #     if search_term:
+    #         await log_search_term(search_query=search_term)
+
+
+async def log_search_term(search_query: str):
+    """Log a search term to the database"""
+
+    # Anonymize the data by preventing correlation with other logs
+    date = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+    try:
+        await prisma.models.SearchTerms.prisma().create(
+            data={"searchTerm": search_query, "createdDate": date}
+        )
+    except Exception as e:
+        # Fail silently here so that logging search terms doesn't break the app
+        logger.error(f"Error logging search term: {e}")
 
 
 async def get_store_agent_details(
@@ -237,9 +389,7 @@ async def get_store_agent_details(
         raise
     except Exception as e:
         logger.error(f"Error getting store agent details: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch agent details"
-        ) from e
+        raise DatabaseError("Failed to fetch agent details") from e
 
 
 async def get_available_graph(store_listing_version_id: str) -> GraphMeta:
@@ -266,9 +416,7 @@ async def get_available_graph(store_listing_version_id: str) -> GraphMeta:
 
     except Exception as e:
         logger.error(f"Error getting agent: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch agent"
-        ) from e
+        raise DatabaseError("Failed to fetch agent") from e
 
 
 async def get_store_agent_by_version_id(
@@ -308,9 +456,7 @@ async def get_store_agent_by_version_id(
         raise
     except Exception as e:
         logger.error(f"Error getting store agent details: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch agent details"
-        ) from e
+        raise DatabaseError("Failed to fetch agent details") from e
 
 
 async def get_store_creators(
@@ -336,9 +482,7 @@ async def get_store_creators(
         # Sanitize and validate search query by escaping special characters
         sanitized_query = search_query.strip()
         if not sanitized_query or len(sanitized_query) > 100:  # Reasonable length limit
-            raise backend.server.v2.store.exceptions.DatabaseError(
-                "Invalid search query"
-            )
+            raise DatabaseError("Invalid search query")
 
         # Escape special SQL characters
         sanitized_query = (
@@ -364,11 +508,9 @@ async def get_store_creators(
     try:
         # Validate pagination parameters
         if not isinstance(page, int) or page < 1:
-            raise backend.server.v2.store.exceptions.DatabaseError(
-                "Invalid page number"
-            )
+            raise DatabaseError("Invalid page number")
         if not isinstance(page_size, int) or page_size < 1 or page_size > 100:
-            raise backend.server.v2.store.exceptions.DatabaseError("Invalid page size")
+            raise DatabaseError("Invalid page size")
 
         # Get total count for pagination using sanitized where clause
         total = await prisma.models.Creator.prisma().count(
@@ -423,9 +565,7 @@ async def get_store_creators(
         )
     except Exception as e:
         logger.error(f"Error getting store creators: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch store creators"
-        ) from e
+        raise DatabaseError("Failed to fetch store creators") from e
 
 
 async def get_store_creator_details(
@@ -460,9 +600,7 @@ async def get_store_creator_details(
         raise
     except Exception as e:
         logger.error(f"Error getting store creator details: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch creator details"
-        ) from e
+        raise DatabaseError("Failed to fetch creator details") from e
 
 
 async def get_store_submissions(
@@ -725,7 +863,21 @@ async def create_store_submission(
             store_listing_version_id=store_listing_version_id,
             changes_summary=changes_summary,
         )
-
+    except prisma.errors.UniqueViolationError as exc:
+        # Attempt to check if the error was due to the slug field being unique
+        error_str = str(exc)
+        if "slug" in error_str.lower():
+            logger.debug(
+                f"Slug '{slug}' is already in use by another agent (agent_id: {agent_id}) for user {user_id}"
+            )
+            raise backend.server.v2.store.exceptions.SlugAlreadyInUseError(
+                f"The URL slug '{slug}' is already in use by another one of your agents. Please choose a different slug."
+            ) from exc
+        else:
+            # Reraise as a generic database error for other unique violations
+            raise DatabaseError(
+                f"Unique constraint violated (not slug): {error_str}"
+            ) from exc
     except (
         backend.server.v2.store.exceptions.AgentNotFoundError,
         backend.server.v2.store.exceptions.ListingExistsError,
@@ -733,9 +885,7 @@ async def create_store_submission(
         raise
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error creating store submission: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to create store submission"
-        ) from e
+        raise DatabaseError("Failed to create store submission") from e
 
 
 async def edit_store_submission(
@@ -856,9 +1006,7 @@ async def edit_store_submission(
             )
 
             if not updated_version:
-                raise backend.server.v2.store.exceptions.DatabaseError(
-                    "Failed to update store listing version"
-                )
+                raise DatabaseError("Failed to update store listing version")
             return backend.server.v2.store.model.StoreSubmission(
                 agent_id=current_version.agentGraphId,
                 agent_version=current_version.agentGraphVersion,
@@ -894,9 +1042,7 @@ async def edit_store_submission(
         raise
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error editing store submission: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to edit store submission"
-        ) from e
+        raise DatabaseError("Failed to edit store submission") from e
 
 
 async def create_store_version(
@@ -1011,9 +1157,7 @@ async def create_store_version(
         )
 
     except prisma.errors.PrismaError as e:
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to create new store version"
-        ) from e
+        raise DatabaseError("Failed to create new store version") from e
 
 
 async def create_store_review(
@@ -1053,9 +1197,7 @@ async def create_store_review(
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error creating store review: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to create store review"
-        ) from e
+        raise DatabaseError("Failed to create store review") from e
 
 
 async def get_user_profile(
@@ -1079,9 +1221,7 @@ async def get_user_profile(
         )
     except Exception as e:
         logger.error(f"Error getting user profile: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to get user profile"
-        ) from e
+        raise DatabaseError("Failed to get user profile") from e
 
 
 async def update_profile(
@@ -1118,7 +1258,7 @@ async def update_profile(
             logger.error(
                 f"Unauthorized update attempt for profile {existing_profile.id} by user {user_id}"
             )
-            raise backend.server.v2.store.exceptions.DatabaseError(
+            raise DatabaseError(
                 f"Unauthorized update attempt for profile {existing_profile.id} by user {user_id}"
             )
 
@@ -1143,9 +1283,7 @@ async def update_profile(
         )
         if updated_profile is None:
             logger.error(f"Failed to update profile for user {user_id}")
-            raise backend.server.v2.store.exceptions.DatabaseError(
-                "Failed to update profile"
-            )
+            raise DatabaseError("Failed to update profile")
 
         return backend.server.v2.store.model.CreatorDetails(
             name=updated_profile.name,
@@ -1160,9 +1298,7 @@ async def update_profile(
 
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error updating profile: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to update profile"
-        ) from e
+        raise DatabaseError("Failed to update profile") from e
 
 
 async def get_my_agents(
@@ -1230,9 +1366,7 @@ async def get_my_agents(
         )
     except Exception as e:
         logger.error(f"Error getting my agents: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to fetch my agents"
-        ) from e
+        raise DatabaseError("Failed to fetch my agents") from e
 
 
 async def get_agent(store_listing_version_id: str) -> GraphModel:
@@ -1494,7 +1628,7 @@ async def review_store_submission(
         )
 
         if not submission:
-            raise backend.server.v2.store.exceptions.DatabaseError(
+            raise DatabaseError(
                 f"Failed to update store listing version {store_listing_version_id}"
             )
 
@@ -1609,9 +1743,7 @@ async def review_store_submission(
 
     except Exception as e:
         logger.error(f"Could not create store submission review: {e}")
-        raise backend.server.v2.store.exceptions.DatabaseError(
-            "Failed to create store submission review"
-        ) from e
+        raise DatabaseError("Failed to create store submission review") from e
 
 
 async def get_admin_listings_with_versions(
@@ -1790,6 +1922,27 @@ async def get_admin_listings_with_versions(
         )
 
 
+async def check_submission_already_approved(
+    store_listing_version_id: str,
+) -> bool:
+    """Check the submission status of a store listing version."""
+    try:
+        store_listing_version = (
+            await prisma.models.StoreListingVersion.prisma().find_unique(
+                where={"id": store_listing_version_id}
+            )
+        )
+        if not store_listing_version:
+            return False
+        return (
+            store_listing_version.submissionStatus
+            == prisma.enums.SubmissionStatus.APPROVED
+        )
+    except Exception as e:
+        logger.error(f"Error checking submission status: {e}")
+        return False
+
+
 async def get_agent_as_admin(
     user_id: str | None,
     store_listing_version_id: str,

autogpt_platform/backend/backend/server/v2/store/db_test.py

@@ -20,7 +20,7 @@ async def setup_prisma():
     yield
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_get_store_agents(mocker):
     # Mock data
     mock_agents = [
@@ -42,6 +42,7 @@ async def test_get_store_agents(mocker):
             versions=["1.0"],
             updated_at=datetime.now(),
             is_available=False,
+            useForOnboarding=False,
         )
     ]
 
@@ -63,7 +64,7 @@ async def test_get_store_agents(mocker):
     mock_store_agent.return_value.count.assert_called_once()
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_get_store_agent_details(mocker):
     # Mock data
     mock_agent = prisma.models.StoreAgent(
@@ -84,6 +85,7 @@ async def test_get_store_agent_details(mocker):
         versions=["1.0"],
         updated_at=datetime.now(),
         is_available=False,
+        useForOnboarding=False,
     )
 
     # Mock active version agent (what we want to return for active version)
@@ -105,6 +107,7 @@ async def test_get_store_agent_details(mocker):
         versions=["1.0", "2.0"],
         updated_at=datetime.now(),
         is_available=True,
+        useForOnboarding=False,
     )
 
     # Create a mock StoreListing result
@@ -170,7 +173,7 @@ async def test_get_store_agent_details(mocker):
     mock_store_listing_db.return_value.find_first.assert_called_once()
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_get_store_creator_details(mocker):
     # Mock data
     mock_creator_data = prisma.models.Creator(
@@ -207,7 +210,7 @@ async def test_get_store_creator_details(mocker):
     )
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_create_store_submission(mocker):
     # Mock data
     mock_agent = prisma.models.AgentGraph(
@@ -248,6 +251,7 @@ async def test_create_store_submission(mocker):
                 isAvailable=True,
             )
         ],
+        useForOnboarding=False,
     )
 
     # Mock prisma calls
@@ -275,11 +279,10 @@ async def test_create_store_submission(mocker):
 
     # Verify mocks called correctly
     mock_agent_graph.return_value.find_first.assert_called_once()
-    mock_store_listing.return_value.find_first.assert_called_once()
     mock_store_listing.return_value.create.assert_called_once()
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_update_profile(mocker):
     # Mock data
     mock_profile = prisma.models.Profile(
@@ -324,7 +327,7 @@ async def test_update_profile(mocker):
     mock_profile_db.return_value.update.assert_called_once()
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_get_user_profile(mocker):
     # Mock data
     mock_profile = prisma.models.Profile(
@@ -356,3 +359,63 @@ async def test_get_user_profile(mocker):
     assert result.description == "Test description"
     assert result.links == ["link1", "link2"]
     assert result.avatar_url == "avatar.jpg"
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_get_store_agents_with_search_parameterized(mocker):
+    """Test that search query uses parameterized SQL - validates the fix works"""
+
+    # Call function with search query containing potential SQL injection
+    malicious_search = "test'; DROP TABLE StoreAgent; --"
+    result = await db.get_store_agents(search_query=malicious_search)
+
+    # Verify query executed safely
+    assert isinstance(result.agents, list)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_get_store_agents_with_search_and_filters_parameterized():
+    """Test parameterized SQL with multiple filters"""
+
+    # Call with multiple filters including potential injection attempts
+    result = await db.get_store_agents(
+        search_query="test",
+        creators=["creator1'; DROP TABLE Users; --", "creator2"],
+        category="AI'; DELETE FROM StoreAgent; --",
+        featured=True,
+        sorted_by="rating",
+        page=1,
+        page_size=20,
+    )
+
+    # Verify the query executed without error
+    assert isinstance(result.agents, list)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_get_store_agents_search_with_invalid_sort_by():
+    """Test that invalid sorted_by value doesn't cause SQL injection"""  # Try to inject SQL via sorted_by parameter
+    malicious_sort = "rating; DROP TABLE Users; --"
+    result = await db.get_store_agents(
+        search_query="test",
+        sorted_by=malicious_sort,
+    )
+
+    # Verify the query executed without error
+    # Invalid sort_by should fall back to default, not cause SQL injection
+    assert isinstance(result.agents, list)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_get_store_agents_search_category_array_injection():
+    """Test that category parameter is safely passed as a parameter"""
+    # Try SQL injection via category
+    malicious_category = "AI'; DROP TABLE StoreAgent; --"
+    result = await db.get_store_agents(
+        search_query="test",
+        category=malicious_category,
+    )
+
+    # Verify the query executed without error
+    # Category should be parameterized, preventing SQL injection
+    assert isinstance(result.agents, list)

autogpt_platform/backend/backend/server/v2/store/exceptions.py

@@ -1,4 +1,7 @@
-class MediaUploadError(Exception):
+from backend.util.exceptions import NotFoundError
+
+
+class MediaUploadError(ValueError):
     """Base exception for media upload errors"""
 
     pass
@@ -48,19 +51,19 @@ class VirusScanError(MediaUploadError):
     pass
 
 
-class StoreError(Exception):
+class StoreError(ValueError):
     """Base exception for store-related errors"""
 
     pass
 
 
-class AgentNotFoundError(StoreError):
+class AgentNotFoundError(NotFoundError):
     """Raised when an agent is not found"""
 
     pass
 
 
-class CreatorNotFoundError(StoreError):
+class CreatorNotFoundError(NotFoundError):
     """Raised when a creator is not found"""
 
     pass
@@ -72,25 +75,19 @@ class ListingExistsError(StoreError):
     pass
 
 
-class DatabaseError(StoreError):
-    """Raised when there is an error interacting with the database"""
-
-    pass
-
-
-class ProfileNotFoundError(StoreError):
+class ProfileNotFoundError(NotFoundError):
     """Raised when a profile is not found"""
 
     pass
 
 
-class ListingNotFoundError(StoreError):
+class ListingNotFoundError(NotFoundError):
     """Raised when a store listing is not found"""
 
     pass
 
 
-class SubmissionNotFoundError(StoreError):
+class SubmissionNotFoundError(NotFoundError):
     """Raised when a submission is not found"""
 
     pass
@@ -106,3 +103,9 @@ class UnauthorizedError(StoreError):
     """Raised when a user is not authorized to perform an action"""
 
     pass
+
+
+class SlugAlreadyInUseError(StoreError):
+    """Raised when a slug is already in use by another agent owned by the user"""
+
+    pass

autogpt_platform/backend/backend/server/v2/store/routes.py

@@ -6,9 +6,9 @@ import urllib.parse
 import autogpt_libs.auth
 import fastapi
 import fastapi.responses
-from autogpt_libs.utils.cache import cached
 
 import backend.data.graph
+import backend.server.v2.store.cache as store_cache
 import backend.server.v2.store.db
 import backend.server.v2.store.exceptions
 import backend.server.v2.store.image_gen
@@ -21,117 +21,6 @@ logger = logging.getLogger(__name__)
 router = fastapi.APIRouter()
 
 
-##############################################
-############### Caches #######################
-##############################################
-
-
-# Cache user profiles for 1 hour per user
-@cached(maxsize=1000, ttl_seconds=3600)
-async def _get_cached_user_profile(user_id: str):
-    """Cached helper to get user profile."""
-    return await backend.server.v2.store.db.get_user_profile(user_id)
-
-
-# Cache store agents list for 15 minutes
-# Different cache entries for different query combinations
-@cached(maxsize=5000, ttl_seconds=900)
-async def _get_cached_store_agents(
-    featured: bool,
-    creator: str | None,
-    sorted_by: str | None,
-    search_query: str | None,
-    category: str | None,
-    page: int,
-    page_size: int,
-):
-    """Cached helper to get store agents."""
-    return await backend.server.v2.store.db.get_store_agents(
-        featured=featured,
-        creators=[creator] if creator else None,
-        sorted_by=sorted_by,
-        search_query=search_query,
-        category=category,
-        page=page,
-        page_size=page_size,
-    )
-
-
-# Cache individual agent details for 15 minutes
-@cached(maxsize=200, ttl_seconds=900)
-async def _get_cached_agent_details(username: str, agent_name: str):
-    """Cached helper to get agent details."""
-    return await backend.server.v2.store.db.get_store_agent_details(
-        username=username, agent_name=agent_name
-    )
-
-
-# Cache agent graphs for 1 hour
-@cached(maxsize=200, ttl_seconds=3600)
-async def _get_cached_agent_graph(store_listing_version_id: str):
-    """Cached helper to get agent graph."""
-    return await backend.server.v2.store.db.get_available_graph(
-        store_listing_version_id
-    )
-
-
-# Cache agent by version for 1 hour
-@cached(maxsize=200, ttl_seconds=3600)
-async def _get_cached_store_agent_by_version(store_listing_version_id: str):
-    """Cached helper to get store agent by version ID."""
-    return await backend.server.v2.store.db.get_store_agent_by_version_id(
-        store_listing_version_id
-    )
-
-
-# Cache creators list for 1 hour
-@cached(maxsize=200, ttl_seconds=3600)
-async def _get_cached_store_creators(
-    featured: bool,
-    search_query: str | None,
-    sorted_by: str | None,
-    page: int,
-    page_size: int,
-):
-    """Cached helper to get store creators."""
-    return await backend.server.v2.store.db.get_store_creators(
-        featured=featured,
-        search_query=search_query,
-        sorted_by=sorted_by,
-        page=page,
-        page_size=page_size,
-    )
-
-
-# Cache individual creator details for 1 hour
-@cached(maxsize=100, ttl_seconds=3600)
-async def _get_cached_creator_details(username: str):
-    """Cached helper to get creator details."""
-    return await backend.server.v2.store.db.get_store_creator_details(
-        username=username.lower()
-    )
-
-
-# Cache user's own agents for 5 mins (shorter TTL as this changes more frequently)
-@cached(maxsize=500, ttl_seconds=300)
-async def _get_cached_my_agents(user_id: str, page: int, page_size: int):
-    """Cached helper to get user's agents."""
-    return await backend.server.v2.store.db.get_my_agents(
-        user_id, page=page, page_size=page_size
-    )
-
-
-# Cache user's submissions for 1 hour (shorter TTL as this changes frequently)
-@cached(maxsize=500, ttl_seconds=3600)
-async def _get_cached_submissions(user_id: str, page: int, page_size: int):
-    """Cached helper to get user's submissions."""
-    return await backend.server.v2.store.db.get_store_submissions(
-        user_id=user_id,
-        page=page,
-        page_size=page_size,
-    )
-
-
 ##############################################
 ############### Profile Endpoints ############
 ##############################################
@@ -151,23 +40,13 @@ async def get_profile(
     Get the profile details for the authenticated user.
     Cached for 1 hour per user.
     """
-    try:
-        profile = await _get_cached_user_profile(user_id)
-        if profile is None:
-            return fastapi.responses.JSONResponse(
-                status_code=404,
-                content={"detail": "Profile not found"},
-            )
-        return profile
-    except Exception as e:
-        logger.exception("Failed to fetch user profile for %s: %s", user_id, e)
+    profile = await backend.server.v2.store.db.get_user_profile(user_id)
+    if profile is None:
         return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "Failed to retrieve user profile",
-                "hint": "Check database connection.",
-            },
+            status_code=404,
+            content={"detail": "Profile not found"},
         )
+    return profile
 
 
 @router.post(
@@ -194,22 +73,10 @@ async def update_or_create_profile(
     Raises:
         HTTPException: If there is an error updating the profile
     """
-    try:
-        updated_profile = await backend.server.v2.store.db.update_profile(
-            user_id=user_id, profile=profile
-        )
-        # Clear the cache for this user after profile update
-        _get_cached_user_profile.cache_delete(user_id)
-        return updated_profile
-    except Exception as e:
-        logger.exception("Failed to update profile for user %s: %s", user_id, e)
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "Failed to update user profile",
-                "hint": "Validate request data.",
-            },
-        )
+    updated_profile = await backend.server.v2.store.db.update_profile(
+        user_id=user_id, profile=profile
+    )
+    return updated_profile
 
 
 ##############################################
@@ -234,7 +101,6 @@ async def get_agents(
 ):
     """
     Get a paginated list of agents from the store with optional filtering and sorting.
-    Results are cached for 15 minutes.
 
     Args:
         featured (bool, optional): Filter to only show featured agents. Defaults to False.
@@ -269,26 +135,16 @@ async def get_agents(
             status_code=422, detail="Page size must be greater than 0"
         )
 
-    try:
-        agents = await _get_cached_store_agents(
-            featured=featured,
-            creator=creator,
-            sorted_by=sorted_by,
-            search_query=search_query,
-            category=category,
-            page=page,
-            page_size=page_size,
-        )
-        return agents
-    except Exception as e:
-        logger.exception("Failed to retrieve store agents: %s", e)
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "Failed to retrieve store agents",
-                "hint": "Check database or search parameters.",
-            },
-        )
+    agents = await store_cache._get_cached_store_agents(
+        featured=featured,
+        creator=creator,
+        sorted_by=sorted_by,
+        search_query=search_query,
+        category=category,
+        page=page,
+        page_size=page_size,
+    )
+    return agents
 
 
 @router.get(
@@ -300,26 +156,16 @@ async def get_agents(
 async def get_agent(username: str, agent_name: str):
     """
     This is only used on the AgentDetails Page.
-    Results are cached for 15 minutes.
 
     It returns the store listing agents details.
     """
-    try:
-        username = urllib.parse.unquote(username).lower()
-        # URL decode the agent name since it comes from the URL path
-        agent_name = urllib.parse.unquote(agent_name).lower()
-        agent = await _get_cached_agent_details(
-            username=username, agent_name=agent_name
-        )
-        return agent
-    except Exception:
-        logger.exception("Exception occurred whilst getting store agent details")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "An error occurred while retrieving the store agent details"
-            },
-        )
+    username = urllib.parse.unquote(username).lower()
+    # URL decode the agent name since it comes from the URL path
+    agent_name = urllib.parse.unquote(agent_name).lower()
+    agent = await store_cache._get_cached_agent_details(
+        username=username, agent_name=agent_name
+    )
+    return agent
 
 
 @router.get(
@@ -331,17 +177,11 @@ async def get_agent(username: str, agent_name: str):
 async def get_graph_meta_by_store_listing_version_id(store_listing_version_id: str):
     """
     Get Agent Graph from Store Listing Version ID.
-    Results are cached for 1 hour.
     """
-    try:
-        graph = await _get_cached_agent_graph(store_listing_version_id)
-        return graph
-    except Exception:
-        logger.exception("Exception occurred whilst getting agent graph")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while retrieving the agent graph"},
-        )
+    graph = await backend.server.v2.store.db.get_available_graph(
+        store_listing_version_id
+    )
+    return graph
 
 
 @router.get(
@@ -354,17 +194,12 @@ async def get_graph_meta_by_store_listing_version_id(store_listing_version_id: s
 async def get_store_agent(store_listing_version_id: str):
     """
     Get Store Agent Details from Store Listing Version ID.
-    Results are cached for 1 hour.
     """
-    try:
-        agent = await _get_cached_store_agent_by_version(store_listing_version_id)
-        return agent
-    except Exception:
-        logger.exception("Exception occurred whilst getting store agent")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while retrieving the store agent"},
-        )
+    agent = await backend.server.v2.store.db.get_store_agent_by_version_id(
+        store_listing_version_id
+    )
+
+    return agent
 
 
 @router.post(
@@ -392,24 +227,17 @@ async def create_review(
     Returns:
         The created review
     """
-    try:
-        username = urllib.parse.unquote(username).lower()
-        agent_name = urllib.parse.unquote(agent_name).lower()
-        # Create the review
-        created_review = await backend.server.v2.store.db.create_store_review(
-            user_id=user_id,
-            store_listing_version_id=review.store_listing_version_id,
-            score=review.score,
-            comments=review.comments,
-        )
+    username = urllib.parse.unquote(username).lower()
+    agent_name = urllib.parse.unquote(agent_name).lower()
+    # Create the review
+    created_review = await backend.server.v2.store.db.create_store_review(
+        user_id=user_id,
+        store_listing_version_id=review.store_listing_version_id,
+        score=review.score,
+        comments=review.comments,
+    )
 
-        return created_review
-    except Exception:
-        logger.exception("Exception occurred whilst creating store review")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while creating the store review"},
-        )
+    return created_review
 
 
 ##############################################
@@ -435,8 +263,6 @@ async def get_creators(
     - Home Page Featured Creators
     - Search Results Page
 
-    Results are cached for 1 hour.
-
     ---
 
     To support this functionality we need:
@@ -454,21 +280,14 @@ async def get_creators(
             status_code=422, detail="Page size must be greater than 0"
         )
 
-    try:
-        creators = await _get_cached_store_creators(
-            featured=featured,
-            search_query=search_query,
-            sorted_by=sorted_by,
-            page=page,
-            page_size=page_size,
-        )
-        return creators
-    except Exception:
-        logger.exception("Exception occurred whilst getting store creators")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while retrieving the store creators"},
-        )
+    creators = await store_cache._get_cached_store_creators(
+        featured=featured,
+        search_query=search_query,
+        sorted_by=sorted_by,
+        page=page,
+        page_size=page_size,
+    )
+    return creators
 
 
 @router.get(
@@ -482,21 +301,11 @@ async def get_creator(
 ):
     """
     Get the details of a creator.
-    Results are cached for 1 hour.
     - Creator Details Page
     """
-    try:
-        username = urllib.parse.unquote(username).lower()
-        creator = await _get_cached_creator_details(username=username)
-        return creator
-    except Exception:
-        logger.exception("Exception occurred whilst getting creator details")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "An error occurred while retrieving the creator details"
-            },
-        )
+    username = urllib.parse.unquote(username).lower()
+    creator = await store_cache._get_cached_creator_details(username=username)
+    return creator
 
 
 ############################################
@@ -518,17 +327,11 @@ async def get_my_agents(
 ):
     """
     Get user's own agents.
-    Results are cached for 5 minutes per user.
     """
-    try:
-        agents = await _get_cached_my_agents(user_id, page=page, page_size=page_size)
-        return agents
-    except Exception:
-        logger.exception("Exception occurred whilst getting my agents")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while retrieving the my agents"},
-        )
+    agents = await backend.server.v2.store.db.get_my_agents(
+        user_id, page=page, page_size=page_size
+    )
+    return agents
 
 
 @router.delete(
@@ -552,26 +355,12 @@ async def delete_submission(
     Returns:
         bool: True if the submission was successfully deleted, False otherwise
     """
-    try:
-        result = await backend.server.v2.store.db.delete_store_submission(
-            user_id=user_id,
-            submission_id=submission_id,
-        )
+    result = await backend.server.v2.store.db.delete_store_submission(
+        user_id=user_id,
+        submission_id=submission_id,
+    )
 
-        # Clear submissions cache for this specific user after deletion
-        if result:
-            # Clear user's own agents cache - we don't know all page/size combinations
-            for page in range(1, 20):
-                # Clear user's submissions cache for common defaults
-                _get_cached_submissions.cache_delete(user_id, page=page, page_size=20)
-
-        return result
-    except Exception:
-        logger.exception("Exception occurred whilst deleting store submission")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while deleting the store submission"},
-        )
+    return result
 
 
 @router.get(
@@ -588,7 +377,6 @@ async def get_submissions(
 ):
     """
     Get a paginated list of store submissions for the authenticated user.
-    Results are cached for 1 hour per user.
 
     Args:
         user_id (str): ID of the authenticated user
@@ -610,19 +398,12 @@ async def get_submissions(
         raise fastapi.HTTPException(
             status_code=422, detail="Page size must be greater than 0"
         )
-    try:
-        listings = await _get_cached_submissions(
-            user_id, page=page, page_size=page_size
-        )
-        return listings
-    except Exception:
-        logger.exception("Exception occurred whilst getting store submissions")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={
-                "detail": "An error occurred while retrieving the store submissions"
-            },
-        )
+    listings = await backend.server.v2.store.db.get_store_submissions(
+        user_id=user_id,
+        page=page,
+        page_size=page_size,
+    )
+    return listings
 
 
 @router.post(
@@ -649,35 +430,23 @@ async def create_submission(
     Raises:
         HTTPException: If there is an error creating the submission
     """
-    try:
-        result = await backend.server.v2.store.db.create_store_submission(
-            user_id=user_id,
-            agent_id=submission_request.agent_id,
-            agent_version=submission_request.agent_version,
-            slug=submission_request.slug,
-            name=submission_request.name,
-            video_url=submission_request.video_url,
-            image_urls=submission_request.image_urls,
-            description=submission_request.description,
-            instructions=submission_request.instructions,
-            sub_heading=submission_request.sub_heading,
-            categories=submission_request.categories,
-            changes_summary=submission_request.changes_summary or "Initial Submission",
-            recommended_schedule_cron=submission_request.recommended_schedule_cron,
-        )
+    result = await backend.server.v2.store.db.create_store_submission(
+        user_id=user_id,
+        agent_id=submission_request.agent_id,
+        agent_version=submission_request.agent_version,
+        slug=submission_request.slug,
+        name=submission_request.name,
+        video_url=submission_request.video_url,
+        image_urls=submission_request.image_urls,
+        description=submission_request.description,
+        instructions=submission_request.instructions,
+        sub_heading=submission_request.sub_heading,
+        categories=submission_request.categories,
+        changes_summary=submission_request.changes_summary or "Initial Submission",
+        recommended_schedule_cron=submission_request.recommended_schedule_cron,
+    )
 
-        # Clear user's own agents cache - we don't know all page/size combinations
-        for page in range(1, 20):
-            # Clear user's submissions cache for common defaults
-            _get_cached_submissions.cache_delete(user_id, page=page, page_size=20)
-
-        return result
-    except Exception:
-        logger.exception("Exception occurred whilst creating store submission")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while creating the store submission"},
-        )
+    return result
 
 
 @router.put(
@@ -720,11 +489,6 @@ async def edit_submission(
         recommended_schedule_cron=submission_request.recommended_schedule_cron,
     )
 
-    # Clear user's own agents cache - we don't know all page/size combinations
-    for page in range(1, 20):
-        # Clear user's submissions cache for common defaults
-        _get_cached_submissions.cache_delete(user_id, page=page, page_size=20)
-
     return result
 
 
@@ -751,36 +515,10 @@ async def upload_submission_media(
     Raises:
         HTTPException: If there is an error uploading the media
     """
-    try:
-        media_url = await backend.server.v2.store.media.upload_media(
-            user_id=user_id, file=file
-        )
-        return media_url
-    except backend.server.v2.store.exceptions.VirusDetectedError as e:
-        logger.warning(f"Virus detected in uploaded file: {e.threat_name}")
-        return fastapi.responses.JSONResponse(
-            status_code=400,
-            content={
-                "detail": f"File rejected due to virus detection: {e.threat_name}",
-                "error_type": "virus_detected",
-                "threat_name": e.threat_name,
-            },
-        )
-    except backend.server.v2.store.exceptions.VirusScanError as e:
-        logger.error(f"Virus scanning failed: {str(e)}")
-        return fastapi.responses.JSONResponse(
-            status_code=503,
-            content={
-                "detail": "Virus scanning service unavailable. Please try again later.",
-                "error_type": "virus_scan_failed",
-            },
-        )
-    except Exception:
-        logger.exception("Exception occurred whilst uploading submission media")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while uploading the media file"},
-        )
+    media_url = await backend.server.v2.store.media.upload_media(
+        user_id=user_id, file=file
+    )
+    return media_url
 
 
 @router.post(
@@ -803,44 +541,35 @@ async def generate_image(
     Returns:
         JSONResponse: JSON containing the URL of the generated image
     """
-    try:
-        agent = await backend.data.graph.get_graph(agent_id, user_id=user_id)
+    agent = await backend.data.graph.get_graph(agent_id, user_id=user_id)
 
-        if not agent:
-            raise fastapi.HTTPException(
-                status_code=404, detail=f"Agent with ID {agent_id} not found"
-            )
-        # Use .jpeg here since we are generating JPEG images
-        filename = f"agent_{agent_id}.jpeg"
+    if not agent:
+        raise fastapi.HTTPException(
+            status_code=404, detail=f"Agent with ID {agent_id} not found"
+        )
+    # Use .jpeg here since we are generating JPEG images
+    filename = f"agent_{agent_id}.jpeg"
 
-        existing_url = await backend.server.v2.store.media.check_media_exists(
-            user_id, filename
-        )
-        if existing_url:
-            logger.info(f"Using existing image for agent {agent_id}")
-            return fastapi.responses.JSONResponse(content={"image_url": existing_url})
-        # Generate agent image as JPEG
-        image = await backend.server.v2.store.image_gen.generate_agent_image(
-            agent=agent
-        )
+    existing_url = await backend.server.v2.store.media.check_media_exists(
+        user_id, filename
+    )
+    if existing_url:
+        logger.info(f"Using existing image for agent {agent_id}")
+        return fastapi.responses.JSONResponse(content={"image_url": existing_url})
+    # Generate agent image as JPEG
+    image = await backend.server.v2.store.image_gen.generate_agent_image(agent=agent)
 
-        # Create UploadFile with the correct filename and content_type
-        image_file = fastapi.UploadFile(
-            file=image,
-            filename=filename,
-        )
+    # Create UploadFile with the correct filename and content_type
+    image_file = fastapi.UploadFile(
+        file=image,
+        filename=filename,
+    )
 
-        image_url = await backend.server.v2.store.media.upload_media(
-            user_id=user_id, file=image_file, use_file_name=True
-        )
+    image_url = await backend.server.v2.store.media.upload_media(
+        user_id=user_id, file=image_file, use_file_name=True
+    )
 
-        return fastapi.responses.JSONResponse(content={"image_url": image_url})
-    except Exception:
-        logger.exception("Exception occurred whilst generating submission image")
-        return fastapi.responses.JSONResponse(
-            status_code=500,
-            content={"detail": "An error occurred while generating the image"},
-        )
+    return fastapi.responses.JSONResponse(content={"image_url": image_url})
 
 
 @router.get(
@@ -917,15 +646,10 @@ async def get_cache_metrics():
         )
 
     # Add metrics for each cache
-    add_cache_metrics("user_profile", _get_cached_user_profile)
-    add_cache_metrics("store_agents", _get_cached_store_agents)
-    add_cache_metrics("agent_details", _get_cached_agent_details)
-    add_cache_metrics("agent_graph", _get_cached_agent_graph)
-    add_cache_metrics("agent_by_version", _get_cached_store_agent_by_version)
-    add_cache_metrics("store_creators", _get_cached_store_creators)
-    add_cache_metrics("creator_details", _get_cached_creator_details)
-    add_cache_metrics("my_agents", _get_cached_my_agents)
-    add_cache_metrics("submissions", _get_cached_submissions)
+    add_cache_metrics("store_agents", store_cache._get_cached_store_agents)
+    add_cache_metrics("agent_details", store_cache._get_cached_agent_details)
+    add_cache_metrics("store_creators", store_cache._get_cached_store_creators)
+    add_cache_metrics("creator_details", store_cache._get_cached_creator_details)
 
     # Add metadata/help text at the beginning
     prometheus_output = [

autogpt_platform/backend/backend/server/v2/store/test_cache_delete.py

@@ -4,18 +4,12 @@ Test suite for verifying cache_delete functionality in store routes.
 Tests that specific cache entries can be deleted while preserving others.
 """
 
-import datetime
 from unittest.mock import AsyncMock, patch
 
 import pytest
 
-from backend.server.v2.store import routes
-from backend.server.v2.store.model import (
-    ProfileDetails,
-    StoreAgent,
-    StoreAgentDetails,
-    StoreAgentsResponse,
-)
+from backend.server.v2.store import cache as store_cache
+from backend.server.v2.store.model import StoreAgent, StoreAgentsResponse
 from backend.util.models import Pagination
 
 
@@ -54,10 +48,10 @@ class TestCacheDeletion:
             return_value=mock_response,
         ) as mock_db:
             # Clear cache first
-            routes._get_cached_store_agents.cache_clear()
+            store_cache._get_cached_store_agents.cache_clear()
 
             # First call - should hit database
-            result1 = await routes._get_cached_store_agents(
+            result1 = await store_cache._get_cached_store_agents(
                 featured=False,
                 creator=None,
                 sorted_by=None,
@@ -70,7 +64,7 @@ class TestCacheDeletion:
             assert result1.agents[0].agent_name == "Test Agent"
 
             # Second call with same params - should use cache
-            await routes._get_cached_store_agents(
+            await store_cache._get_cached_store_agents(
                 featured=False,
                 creator=None,
                 sorted_by=None,
@@ -82,7 +76,7 @@ class TestCacheDeletion:
             assert mock_db.call_count == 1  # No additional DB call
 
             # Third call with different params - should hit database
-            await routes._get_cached_store_agents(
+            await store_cache._get_cached_store_agents(
                 featured=True,  # Different param
                 creator=None,
                 sorted_by=None,
@@ -94,7 +88,7 @@ class TestCacheDeletion:
             assert mock_db.call_count == 2  # New DB call
 
             # Delete specific cache entry
-            deleted = routes._get_cached_store_agents.cache_delete(
+            deleted = store_cache._get_cached_store_agents.cache_delete(
                 featured=False,
                 creator=None,
                 sorted_by=None,
@@ -106,7 +100,7 @@ class TestCacheDeletion:
             assert deleted is True  # Entry was deleted
 
             # Try to delete non-existent entry
-            deleted = routes._get_cached_store_agents.cache_delete(
+            deleted = store_cache._get_cached_store_agents.cache_delete(
                 featured=False,
                 creator="nonexistent",
                 sorted_by=None,
@@ -118,7 +112,7 @@ class TestCacheDeletion:
             assert deleted is False  # Entry didn't exist
 
             # Call with deleted params - should hit database again
-            await routes._get_cached_store_agents(
+            await store_cache._get_cached_store_agents(
                 featured=False,
                 creator=None,
                 sorted_by=None,
@@ -130,7 +124,7 @@ class TestCacheDeletion:
             assert mock_db.call_count == 3  # New DB call after deletion
 
             # Call with featured=True - should still be cached
-            await routes._get_cached_store_agents(
+            await store_cache._get_cached_store_agents(
                 featured=True,
                 creator=None,
                 sorted_by=None,
@@ -141,105 +135,11 @@ class TestCacheDeletion:
             )
             assert mock_db.call_count == 3  # No additional DB call
 
-    @pytest.mark.asyncio
-    async def test_agent_details_cache_delete(self):
-        """Test that specific agent details cache entries can be deleted."""
-        mock_response = StoreAgentDetails(
-            store_listing_version_id="version1",
-            slug="test-agent",
-            agent_name="Test Agent",
-            agent_video="https://example.com/video.mp4",
-            agent_image=["https://example.com/image.jpg"],
-            creator="testuser",
-            creator_avatar="https://example.com/avatar.jpg",
-            sub_heading="Test subheading",
-            description="Test description",
-            categories=["productivity"],
-            runs=100,
-            rating=4.5,
-            versions=[],
-            last_updated=datetime.datetime(2024, 1, 1),
-        )
-
-        with patch(
-            "backend.server.v2.store.db.get_store_agent_details",
-            new_callable=AsyncMock,
-            return_value=mock_response,
-        ) as mock_db:
-            # Clear cache first
-            routes._get_cached_agent_details.cache_clear()
-
-            # First call - should hit database
-            await routes._get_cached_agent_details(
-                username="testuser", agent_name="testagent"
-            )
-            assert mock_db.call_count == 1
-
-            # Second call - should use cache
-            await routes._get_cached_agent_details(
-                username="testuser", agent_name="testagent"
-            )
-            assert mock_db.call_count == 1  # No additional DB call
-
-            # Delete specific entry
-            deleted = routes._get_cached_agent_details.cache_delete(
-                username="testuser", agent_name="testagent"
-            )
-            assert deleted is True
-
-            # Call again - should hit database
-            await routes._get_cached_agent_details(
-                username="testuser", agent_name="testagent"
-            )
-            assert mock_db.call_count == 2  # New DB call after deletion
-
-    @pytest.mark.asyncio
-    async def test_user_profile_cache_delete(self):
-        """Test that user profile cache entries can be deleted."""
-        mock_response = ProfileDetails(
-            name="Test User",
-            username="testuser",
-            description="Test profile",
-            links=["https://example.com"],
-        )
-
-        with patch(
-            "backend.server.v2.store.db.get_user_profile",
-            new_callable=AsyncMock,
-            return_value=mock_response,
-        ) as mock_db:
-            # Clear cache first
-            routes._get_cached_user_profile.cache_clear()
-
-            # First call - should hit database
-            await routes._get_cached_user_profile("user123")
-            assert mock_db.call_count == 1
-
-            # Second call - should use cache
-            await routes._get_cached_user_profile("user123")
-            assert mock_db.call_count == 1
-
-            # Different user - should hit database
-            await routes._get_cached_user_profile("user456")
-            assert mock_db.call_count == 2
-
-            # Delete specific user's cache
-            deleted = routes._get_cached_user_profile.cache_delete("user123")
-            assert deleted is True
-
-            # user123 should hit database again
-            await routes._get_cached_user_profile("user123")
-            assert mock_db.call_count == 3
-
-            # user456 should still be cached
-            await routes._get_cached_user_profile("user456")
-            assert mock_db.call_count == 3  # No additional DB call
-
     @pytest.mark.asyncio
     async def test_cache_info_after_deletions(self):
         """Test that cache_info correctly reflects deletions."""
         # Clear all caches first
-        routes._get_cached_store_agents.cache_clear()
+        store_cache._get_cached_store_agents.cache_clear()
 
         mock_response = StoreAgentsResponse(
             agents=[],
@@ -258,7 +158,7 @@ class TestCacheDeletion:
         ):
             # Add multiple entries
             for i in range(5):
-                await routes._get_cached_store_agents(
+                await store_cache._get_cached_store_agents(
                     featured=False,
                     creator=f"creator{i}",
                     sorted_by=None,
@@ -269,12 +169,12 @@ class TestCacheDeletion:
                 )
 
             # Check cache size
-            info = routes._get_cached_store_agents.cache_info()
+            info = store_cache._get_cached_store_agents.cache_info()
             assert info["size"] == 5
 
             # Delete some entries
             for i in range(2):
-                deleted = routes._get_cached_store_agents.cache_delete(
+                deleted = store_cache._get_cached_store_agents.cache_delete(
                     featured=False,
                     creator=f"creator{i}",
                     sorted_by=None,
@@ -286,7 +186,7 @@ class TestCacheDeletion:
                 assert deleted is True
 
             # Check cache size after deletion
-            info = routes._get_cached_store_agents.cache_info()
+            info = store_cache._get_cached_store_agents.cache_info()
             assert info["size"] == 3
 
     @pytest.mark.asyncio
@@ -307,10 +207,10 @@ class TestCacheDeletion:
             new_callable=AsyncMock,
             return_value=mock_response,
         ) as mock_db:
-            routes._get_cached_store_agents.cache_clear()
+            store_cache._get_cached_store_agents.cache_clear()
 
             # Test with all parameters
-            await routes._get_cached_store_agents(
+            await store_cache._get_cached_store_agents(
                 featured=True,
                 creator="testuser",
                 sorted_by="rating",
@@ -322,7 +222,7 @@ class TestCacheDeletion:
             assert mock_db.call_count == 1
 
             # Delete with exact same parameters
-            deleted = routes._get_cached_store_agents.cache_delete(
+            deleted = store_cache._get_cached_store_agents.cache_delete(
                 featured=True,
                 creator="testuser",
                 sorted_by="rating",
@@ -334,7 +234,7 @@ class TestCacheDeletion:
             assert deleted is True
 
             # Try to delete with slightly different parameters
-            deleted = routes._get_cached_store_agents.cache_delete(
+            deleted = store_cache._get_cached_store_agents.cache_delete(
                 featured=True,
                 creator="testuser",
                 sorted_by="rating",

autogpt_platform/backend/backend/server/ws_api.py

@@ -329,7 +329,3 @@ class WebsocketServer(AppProcess):
             port=Config().websocket_server_port,
             log_config=None,
         )
-
-    def cleanup(self):
-        super().cleanup()
-        logger.info(f"[{self.service_name}] ⏳ Shutting down WebSocket Server...")

autogpt_platform/backend/backend/util/cache.py

@@ -0,0 +1,457 @@
+"""
+Caching utilities for the AutoGPT platform.
+
+Provides decorators for caching function results with support for:
+- In-memory caching with TTL
+- Shared Redis-backed caching across processes
+- Thread-local caching for request-scoped data
+- Thundering herd protection
+- LRU eviction with optional TTL refresh
+"""
+
+import asyncio
+import inspect
+import logging
+import pickle
+import threading
+import time
+from dataclasses import dataclass
+from functools import wraps
+from typing import Any, Callable, ParamSpec, Protocol, TypeVar, cast, runtime_checkable
+
+from redis import ConnectionPool, Redis
+
+from backend.util.retry import conn_retry
+from backend.util.settings import Settings
+
+P = ParamSpec("P")
+R = TypeVar("R")
+R_co = TypeVar("R_co", covariant=True)
+
+logger = logging.getLogger(__name__)
+settings = Settings()
+
+# RECOMMENDED REDIS CONFIGURATION FOR PRODUCTION:
+# Configure Redis with the following settings for optimal caching performance:
+#   maxmemory-policy allkeys-lru    # Evict least recently used keys when memory limit reached
+#   maxmemory 2gb                   # Set memory limit (adjust based on your needs)
+#   save ""                         # Disable persistence if using Redis purely for caching
+
+# Create a dedicated Redis connection pool for caching (binary mode for pickle)
+_cache_pool: ConnectionPool | None = None
+
+
+@conn_retry("Redis", "Acquiring cache connection pool")
+def _get_cache_pool() -> ConnectionPool:
+    """Get or create a connection pool for cache operations."""
+    global _cache_pool
+    if _cache_pool is None:
+        _cache_pool = ConnectionPool(
+            host=settings.config.redis_host,
+            port=settings.config.redis_port,
+            password=settings.config.redis_password or None,
+            decode_responses=False,  # Binary mode for pickle
+            max_connections=50,
+            socket_keepalive=True,
+            socket_connect_timeout=5,
+            retry_on_timeout=True,
+        )
+    return _cache_pool
+
+
+redis = Redis(connection_pool=_get_cache_pool())
+
+
+@dataclass
+class CachedValue:
+    """Wrapper for cached values with timestamp to avoid tuple ambiguity."""
+
+    result: Any
+    timestamp: float
+
+
+def _make_hashable_key(
+    args: tuple[Any, ...], kwargs: dict[str, Any]
+) -> tuple[Any, ...]:
+    """
+    Convert args and kwargs into a hashable cache key.
+
+    Handles unhashable types like dict, list, set by converting them to
+    their sorted string representations.
+    """
+
+    def make_hashable(obj: Any) -> Any:
+        """Recursively convert an object to a hashable representation."""
+        if isinstance(obj, dict):
+            # Sort dict items to ensure consistent ordering
+            return (
+                "__dict__",
+                tuple(sorted((k, make_hashable(v)) for k, v in obj.items())),
+            )
+        elif isinstance(obj, (list, tuple)):
+            return ("__list__", tuple(make_hashable(item) for item in obj))
+        elif isinstance(obj, set):
+            return ("__set__", tuple(sorted(make_hashable(item) for item in obj)))
+        elif hasattr(obj, "__dict__"):
+            # Handle objects with __dict__ attribute
+            return ("__obj__", obj.__class__.__name__, make_hashable(obj.__dict__))
+        else:
+            # For basic hashable types (str, int, bool, None, etc.)
+            try:
+                hash(obj)
+                return obj
+            except TypeError:
+                # Fallback: convert to string representation
+                return ("__str__", str(obj))
+
+    hashable_args = tuple(make_hashable(arg) for arg in args)
+    hashable_kwargs = tuple(sorted((k, make_hashable(v)) for k, v in kwargs.items()))
+    return (hashable_args, hashable_kwargs)
+
+
+def _make_redis_key(key: tuple[Any, ...], func_name: str) -> str:
+    """Convert a hashable key tuple to a Redis key string."""
+    # Ensure key is already hashable
+    hashable_key = key if isinstance(key, tuple) else (key,)
+    return f"cache:{func_name}:{hash(hashable_key)}"
+
+
+@runtime_checkable
+class CachedFunction(Protocol[P, R_co]):
+    """Protocol for cached functions with cache management methods."""
+
+    def cache_clear(self, pattern: str | None = None) -> None:
+        """Clear cached entries. If pattern provided, clear matching entries only."""
+        return None
+
+    def cache_info(self) -> dict[str, int | None]:
+        """Get cache statistics."""
+        return {}
+
+    def cache_delete(self, *args: P.args, **kwargs: P.kwargs) -> bool:
+        """Delete a specific cache entry by its arguments. Returns True if entry existed."""
+        return False
+
+    def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R_co:
+        """Call the cached function."""
+        return None  # type: ignore
+
+
+def cached(
+    *,
+    maxsize: int = 128,
+    ttl_seconds: int,
+    shared_cache: bool = False,
+    refresh_ttl_on_get: bool = False,
+) -> Callable[[Callable], CachedFunction]:
+    """
+    Thundering herd safe cache decorator for both sync and async functions.
+
+    Uses double-checked locking to prevent multiple threads/coroutines from
+    executing the expensive operation simultaneously during cache misses.
+
+    Args:
+        maxsize: Maximum number of cached entries (only for in-memory cache)
+        ttl_seconds: Time to live in seconds. Required - entries must expire.
+        shared_cache: If True, use Redis for cross-process caching
+        refresh_ttl_on_get: If True, refresh TTL when cache entry is accessed (LRU behavior)
+
+    Returns:
+        Decorated function with caching capabilities
+
+    Example:
+        @cached(ttl_seconds=300)  # 5 minute TTL
+        def expensive_sync_operation(param: str) -> dict:
+            return {"result": param}
+
+        @cached(ttl_seconds=600, shared_cache=True, refresh_ttl_on_get=True)
+        async def expensive_async_operation(param: str) -> dict:
+            return {"result": param}
+    """
+
+    def decorator(target_func):
+        cache_storage: dict[tuple, CachedValue] = {}
+        _event_loop_locks: dict[Any, asyncio.Lock] = {}
+
+        def _get_from_redis(redis_key: str) -> Any | None:
+            """Get value from Redis, optionally refreshing TTL."""
+            try:
+                if refresh_ttl_on_get:
+                    # Use GETEX to get value and refresh expiry atomically
+                    cached_bytes = redis.getex(redis_key, ex=ttl_seconds)
+                else:
+                    cached_bytes = redis.get(redis_key)
+
+                if cached_bytes and isinstance(cached_bytes, bytes):
+                    return pickle.loads(cached_bytes)
+            except Exception as e:
+                logger.error(
+                    f"Redis error during cache check for {target_func.__name__}: {e}"
+                )
+            return None
+
+        def _set_to_redis(redis_key: str, value: Any) -> None:
+            """Set value in Redis with TTL."""
+            try:
+                pickled_value = pickle.dumps(value, protocol=pickle.HIGHEST_PROTOCOL)
+                redis.setex(redis_key, ttl_seconds, pickled_value)
+            except Exception as e:
+                logger.error(
+                    f"Redis error storing cache for {target_func.__name__}: {e}"
+                )
+
+        def _get_from_memory(key: tuple) -> Any | None:
+            """Get value from in-memory cache, checking TTL."""
+            if key in cache_storage:
+                cached_data = cache_storage[key]
+                if time.time() - cached_data.timestamp < ttl_seconds:
+                    logger.debug(
+                        f"Cache hit for {target_func.__name__} args: {key[0]} kwargs: {key[1]}"
+                    )
+                    return cached_data.result
+            return None
+
+        def _set_to_memory(key: tuple, value: Any) -> None:
+            """Set value in in-memory cache with timestamp."""
+            cache_storage[key] = CachedValue(result=value, timestamp=time.time())
+
+            # Cleanup if needed
+            if len(cache_storage) > maxsize:
+                cutoff = maxsize // 2
+                oldest_keys = list(cache_storage.keys())[:-cutoff] if cutoff > 0 else []
+                for old_key in oldest_keys:
+                    cache_storage.pop(old_key, None)
+
+        if inspect.iscoroutinefunction(target_func):
+
+            def _get_cache_lock():
+                """Get or create an asyncio.Lock for the current event loop."""
+                try:
+                    loop = asyncio.get_running_loop()
+                except RuntimeError:
+                    loop = None
+
+                if loop not in _event_loop_locks:
+                    return _event_loop_locks.setdefault(loop, asyncio.Lock())
+                return _event_loop_locks[loop]
+
+            @wraps(target_func)
+            async def async_wrapper(*args: P.args, **kwargs: P.kwargs):
+                key = _make_hashable_key(args, kwargs)
+                redis_key = (
+                    _make_redis_key(key, target_func.__name__) if shared_cache else ""
+                )
+
+                # Fast path: check cache without lock
+                if shared_cache:
+                    result = _get_from_redis(redis_key)
+                    if result is not None:
+                        return result
+                else:
+                    result = _get_from_memory(key)
+                    if result is not None:
+                        return result
+
+                # Slow path: acquire lock for cache miss/expiry
+                async with _get_cache_lock():
+                    # Double-check: another coroutine might have populated cache
+                    if shared_cache:
+                        result = _get_from_redis(redis_key)
+                        if result is not None:
+                            return result
+                    else:
+                        result = _get_from_memory(key)
+                        if result is not None:
+                            return result
+
+                    # Cache miss - execute function
+                    logger.debug(f"Cache miss for {target_func.__name__}")
+                    result = await target_func(*args, **kwargs)
+
+                    # Store result
+                    if shared_cache:
+                        _set_to_redis(redis_key, result)
+                    else:
+                        _set_to_memory(key, result)
+
+                    return result
+
+            wrapper = async_wrapper
+
+        else:
+            # Sync function with threading.Lock
+            cache_lock = threading.Lock()
+
+            @wraps(target_func)
+            def sync_wrapper(*args: P.args, **kwargs: P.kwargs):
+                key = _make_hashable_key(args, kwargs)
+                redis_key = (
+                    _make_redis_key(key, target_func.__name__) if shared_cache else ""
+                )
+
+                # Fast path: check cache without lock
+                if shared_cache:
+                    result = _get_from_redis(redis_key)
+                    if result is not None:
+                        return result
+                else:
+                    result = _get_from_memory(key)
+                    if result is not None:
+                        return result
+
+                # Slow path: acquire lock for cache miss/expiry
+                with cache_lock:
+                    # Double-check: another thread might have populated cache
+                    if shared_cache:
+                        result = _get_from_redis(redis_key)
+                        if result is not None:
+                            return result
+                    else:
+                        result = _get_from_memory(key)
+                        if result is not None:
+                            return result
+
+                    # Cache miss - execute function
+                    logger.debug(f"Cache miss for {target_func.__name__}")
+                    result = target_func(*args, **kwargs)
+
+                    # Store result
+                    if shared_cache:
+                        _set_to_redis(redis_key, result)
+                    else:
+                        _set_to_memory(key, result)
+
+                    return result
+
+            wrapper = sync_wrapper
+
+        # Add cache management methods
+        def cache_clear(pattern: str | None = None) -> None:
+            """Clear cache entries. If pattern provided, clear matching entries."""
+            if shared_cache:
+                if pattern:
+                    # Clear entries matching pattern
+                    keys = list(
+                        redis.scan_iter(f"cache:{target_func.__name__}:{pattern}")
+                    )
+                else:
+                    # Clear all cache keys
+                    keys = list(redis.scan_iter(f"cache:{target_func.__name__}:*"))
+
+                if keys:
+                    pipeline = redis.pipeline()
+                    for key in keys:
+                        pipeline.delete(key)
+                    pipeline.execute()
+            else:
+                if pattern:
+                    # For in-memory cache, pattern matching not supported
+                    logger.warning(
+                        "Pattern-based clearing not supported for in-memory cache"
+                    )
+                else:
+                    cache_storage.clear()
+
+        def cache_info() -> dict[str, int | None]:
+            if shared_cache:
+                cache_keys = list(redis.scan_iter(f"cache:{target_func.__name__}:*"))
+                return {
+                    "size": len(cache_keys),
+                    "maxsize": None,  # Redis manages its own size
+                    "ttl_seconds": ttl_seconds,
+                }
+            else:
+                return {
+                    "size": len(cache_storage),
+                    "maxsize": maxsize,
+                    "ttl_seconds": ttl_seconds,
+                }
+
+        def cache_delete(*args, **kwargs) -> bool:
+            """Delete a specific cache entry. Returns True if entry existed."""
+            key = _make_hashable_key(args, kwargs)
+            if shared_cache:
+                redis_key = _make_redis_key(key, target_func.__name__)
+                if redis.exists(redis_key):
+                    redis.delete(redis_key)
+                    return True
+                return False
+            else:
+                if key in cache_storage:
+                    del cache_storage[key]
+                    return True
+                return False
+
+        setattr(wrapper, "cache_clear", cache_clear)
+        setattr(wrapper, "cache_info", cache_info)
+        setattr(wrapper, "cache_delete", cache_delete)
+
+        return cast(CachedFunction, wrapper)
+
+    return decorator
+
+
+def thread_cached(func):
+    """
+    Thread-local cache decorator for both sync and async functions.
+
+    Each thread gets its own cache, which is useful for request-scoped caching
+    in web applications where you want to cache within a single request but
+    not across requests.
+
+    Args:
+        func: The function to cache
+
+    Returns:
+        Decorated function with thread-local caching
+
+    Example:
+        @thread_cached
+        def expensive_operation(param: str) -> dict:
+            return {"result": param}
+
+        @thread_cached  # Works with async too
+        async def expensive_async_operation(param: str) -> dict:
+            return {"result": param}
+    """
+    thread_local = threading.local()
+
+    def _clear():
+        if hasattr(thread_local, "cache"):
+            del thread_local.cache
+
+    if inspect.iscoroutinefunction(func):
+
+        @wraps(func)
+        async def async_wrapper(*args, **kwargs):
+            cache = getattr(thread_local, "cache", None)
+            if cache is None:
+                cache = thread_local.cache = {}
+            key = _make_hashable_key(args, kwargs)
+            if key not in cache:
+                cache[key] = await func(*args, **kwargs)
+            return cache[key]
+
+        setattr(async_wrapper, "clear_cache", _clear)
+        return async_wrapper
+
+    else:
+
+        @wraps(func)
+        def sync_wrapper(*args, **kwargs):
+            cache = getattr(thread_local, "cache", None)
+            if cache is None:
+                cache = thread_local.cache = {}
+            key = _make_hashable_key(args, kwargs)
+            if key not in cache:
+                cache[key] = func(*args, **kwargs)
+            return cache[key]
+
+        setattr(sync_wrapper, "clear_cache", _clear)
+        return sync_wrapper
+
+
+def clear_thread_cache(func: Callable) -> None:
+    """Clear thread-local cache for a function."""
+    if clear := getattr(func, "clear_cache", None):
+        clear()

autogpt_platform/backend/backend/util/cache_test.py

@@ -16,7 +16,7 @@ from unittest.mock import Mock
 
 import pytest
 
-from autogpt_libs.utils.cache import cached, clear_thread_cache, thread_cached
+from backend.util.cache import cached, clear_thread_cache, thread_cached
 
 
 class TestThreadCached:
@@ -332,7 +332,7 @@ class TestCache:
         """Test basic sync caching functionality."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         def expensive_sync_function(x: int, y: int = 0) -> int:
             nonlocal call_count
             call_count += 1
@@ -358,7 +358,7 @@ class TestCache:
         """Test basic async caching functionality."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         async def expensive_async_function(x: int, y: int = 0) -> int:
             nonlocal call_count
             call_count += 1
@@ -385,7 +385,7 @@ class TestCache:
         call_count = 0
         results = []
 
-        @cached()
+        @cached(ttl_seconds=300)
         def slow_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -412,7 +412,7 @@ class TestCache:
         """Test that concurrent async calls don't cause thundering herd."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         async def slow_async_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -508,7 +508,7 @@ class TestCache:
         """Test cache clearing functionality."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         def clearable_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -537,7 +537,7 @@ class TestCache:
         """Test cache clearing functionality with async function."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         async def async_clearable_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -567,7 +567,7 @@ class TestCache:
         """Test that cached async functions return actual results, not coroutines."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         async def async_result_function(x: int) -> str:
             nonlocal call_count
             call_count += 1
@@ -593,7 +593,7 @@ class TestCache:
         """Test selective cache deletion functionality."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         def deletable_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -636,7 +636,7 @@ class TestCache:
         """Test selective cache deletion functionality with async function."""
         call_count = 0
 
-        @cached()
+        @cached(ttl_seconds=300)
         async def async_deletable_function(x: int) -> int:
             nonlocal call_count
             call_count += 1
@@ -674,3 +674,450 @@ class TestCache:
         # Try to delete non-existent entry
         was_deleted = async_deletable_function.cache_delete(99)
         assert was_deleted is False
+
+
+class TestSharedCache:
+    """Tests for shared_cache (Redis-backed) functionality."""
+
+    def test_sync_shared_cache_basic(self):
+        """Test basic shared cache functionality with sync function."""
+        call_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def shared_sync_function(x: int, y: int = 0) -> int:
+            nonlocal call_count
+            call_count += 1
+            return x + y
+
+        # Clear any existing cache
+        shared_sync_function.cache_clear()
+
+        # First call
+        result1 = shared_sync_function(10, 20)
+        assert result1 == 30
+        assert call_count == 1
+
+        # Second call - should use Redis cache
+        result2 = shared_sync_function(10, 20)
+        assert result2 == 30
+        assert call_count == 1
+
+        # Different args - should call function again
+        result3 = shared_sync_function(15, 25)
+        assert result3 == 40
+        assert call_count == 2
+
+        # Cleanup
+        shared_sync_function.cache_clear()
+
+    @pytest.mark.asyncio
+    async def test_async_shared_cache_basic(self):
+        """Test basic shared cache functionality with async function."""
+        call_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        async def shared_async_function(x: int, y: int = 0) -> int:
+            nonlocal call_count
+            call_count += 1
+            await asyncio.sleep(0.01)
+            return x + y
+
+        # Clear any existing cache
+        shared_async_function.cache_clear()
+
+        # First call
+        result1 = await shared_async_function(10, 20)
+        assert result1 == 30
+        assert call_count == 1
+
+        # Second call - should use Redis cache
+        result2 = await shared_async_function(10, 20)
+        assert result2 == 30
+        assert call_count == 1
+
+        # Different args - should call function again
+        result3 = await shared_async_function(15, 25)
+        assert result3 == 40
+        assert call_count == 2
+
+        # Cleanup
+        shared_async_function.cache_clear()
+
+    def test_shared_cache_ttl_refresh(self):
+        """Test TTL refresh functionality with shared cache."""
+        call_count = 0
+
+        @cached(ttl_seconds=2, shared_cache=True, refresh_ttl_on_get=True)
+        def ttl_refresh_function(x: int) -> int:
+            nonlocal call_count
+            call_count += 1
+            return x * 10
+
+        # Clear any existing cache
+        ttl_refresh_function.cache_clear()
+
+        # First call
+        result1 = ttl_refresh_function(3)
+        assert result1 == 30
+        assert call_count == 1
+
+        # Wait 1 second
+        time.sleep(1)
+
+        # Second call - should refresh TTL and use cache
+        result2 = ttl_refresh_function(3)
+        assert result2 == 30
+        assert call_count == 1
+
+        # Wait another 1.5 seconds (total 2.5s from first call, 1.5s from second)
+        time.sleep(1.5)
+
+        # Third call - TTL should have been refreshed, so still cached
+        result3 = ttl_refresh_function(3)
+        assert result3 == 30
+        assert call_count == 1
+
+        # Wait 2.1 seconds - now it should expire
+        time.sleep(2.1)
+
+        # Fourth call - should call function again
+        result4 = ttl_refresh_function(3)
+        assert result4 == 30
+        assert call_count == 2
+
+        # Cleanup
+        ttl_refresh_function.cache_clear()
+
+    def test_shared_cache_without_ttl_refresh(self):
+        """Test that TTL doesn't refresh when refresh_ttl_on_get=False."""
+        call_count = 0
+
+        @cached(ttl_seconds=2, shared_cache=True, refresh_ttl_on_get=False)
+        def no_ttl_refresh_function(x: int) -> int:
+            nonlocal call_count
+            call_count += 1
+            return x * 10
+
+        # Clear any existing cache
+        no_ttl_refresh_function.cache_clear()
+
+        # First call
+        result1 = no_ttl_refresh_function(4)
+        assert result1 == 40
+        assert call_count == 1
+
+        # Wait 1 second
+        time.sleep(1)
+
+        # Second call - should use cache but NOT refresh TTL
+        result2 = no_ttl_refresh_function(4)
+        assert result2 == 40
+        assert call_count == 1
+
+        # Wait another 1.1 seconds (total 2.1s from first call)
+        time.sleep(1.1)
+
+        # Third call - should have expired
+        result3 = no_ttl_refresh_function(4)
+        assert result3 == 40
+        assert call_count == 2
+
+        # Cleanup
+        no_ttl_refresh_function.cache_clear()
+
+    def test_shared_cache_complex_objects(self):
+        """Test caching complex objects with shared cache (pickle serialization)."""
+        call_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def complex_object_function(x: int) -> dict:
+            nonlocal call_count
+            call_count += 1
+            return {
+                "number": x,
+                "squared": x**2,
+                "nested": {"list": [1, 2, x], "tuple": (x, x * 2)},
+                "string": f"value_{x}",
+            }
+
+        # Clear any existing cache
+        complex_object_function.cache_clear()
+
+        # First call
+        result1 = complex_object_function(5)
+        assert result1["number"] == 5
+        assert result1["squared"] == 25
+        assert result1["nested"]["list"] == [1, 2, 5]
+        assert call_count == 1
+
+        # Second call - should use cache
+        result2 = complex_object_function(5)
+        assert result2 == result1
+        assert call_count == 1
+
+        # Cleanup
+        complex_object_function.cache_clear()
+
+    def test_shared_cache_info(self):
+        """Test cache_info for shared cache."""
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def info_shared_function(x: int) -> int:
+            return x * 2
+
+        # Clear any existing cache
+        info_shared_function.cache_clear()
+
+        # Check initial info
+        info = info_shared_function.cache_info()
+        assert info["size"] == 0
+        assert info["maxsize"] is None  # Redis manages size
+        assert info["ttl_seconds"] == 30
+
+        # Add some entries
+        info_shared_function(1)
+        info_shared_function(2)
+        info_shared_function(3)
+
+        info = info_shared_function.cache_info()
+        assert info["size"] == 3
+
+        # Cleanup
+        info_shared_function.cache_clear()
+
+    def test_shared_cache_delete(self):
+        """Test selective deletion with shared cache."""
+        call_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def delete_shared_function(x: int) -> int:
+            nonlocal call_count
+            call_count += 1
+            return x * 3
+
+        # Clear any existing cache
+        delete_shared_function.cache_clear()
+
+        # Add entries
+        delete_shared_function(1)
+        delete_shared_function(2)
+        delete_shared_function(3)
+        assert call_count == 3
+
+        # Verify cached
+        delete_shared_function(1)
+        delete_shared_function(2)
+        assert call_count == 3
+
+        # Delete specific entry
+        was_deleted = delete_shared_function.cache_delete(2)
+        assert was_deleted is True
+
+        # Entry for x=2 should be gone
+        delete_shared_function(2)
+        assert call_count == 4
+
+        # Others should still be cached
+        delete_shared_function(1)
+        delete_shared_function(3)
+        assert call_count == 4
+
+        # Try to delete non-existent
+        was_deleted = delete_shared_function.cache_delete(99)
+        assert was_deleted is False
+
+        # Cleanup
+        delete_shared_function.cache_clear()
+
+    @pytest.mark.asyncio
+    async def test_async_shared_cache_thundering_herd(self):
+        """Test that shared cache prevents thundering herd for async functions."""
+        call_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        async def shared_slow_function(x: int) -> int:
+            nonlocal call_count
+            call_count += 1
+            await asyncio.sleep(0.1)
+            return x * x
+
+        # Clear any existing cache
+        shared_slow_function.cache_clear()
+
+        # Launch multiple concurrent tasks
+        tasks = [shared_slow_function(8) for _ in range(10)]
+        results = await asyncio.gather(*tasks)
+
+        # All should return same result
+        assert all(r == 64 for r in results)
+        # Only one should have executed
+        assert call_count == 1
+
+        # Cleanup
+        shared_slow_function.cache_clear()
+
+    def test_shared_cache_clear_pattern(self):
+        """Test pattern-based cache clearing (Redis feature)."""
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def pattern_function(category: str, item: int) -> str:
+            return f"{category}_{item}"
+
+        # Clear any existing cache
+        pattern_function.cache_clear()
+
+        # Add various entries
+        pattern_function("fruit", 1)
+        pattern_function("fruit", 2)
+        pattern_function("vegetable", 1)
+        pattern_function("vegetable", 2)
+
+        info = pattern_function.cache_info()
+        assert info["size"] == 4
+
+        # Note: Pattern clearing with wildcards requires specific Redis scan
+        # implementation. The current code clears by pattern but needs
+        # adjustment for partial matching. For now, test full clear.
+        pattern_function.cache_clear()
+        info = pattern_function.cache_info()
+        assert info["size"] == 0
+
+    def test_shared_vs_local_cache_isolation(self):
+        """Test that shared and local caches are isolated."""
+        shared_count = 0
+        local_count = 0
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def shared_function(x: int) -> int:
+            nonlocal shared_count
+            shared_count += 1
+            return x * 2
+
+        @cached(ttl_seconds=30, shared_cache=False)
+        def local_function(x: int) -> int:
+            nonlocal local_count
+            local_count += 1
+            return x * 2
+
+        # Clear caches
+        shared_function.cache_clear()
+        local_function.cache_clear()
+
+        # Call both with same args
+        shared_result = shared_function(5)
+        local_result = local_function(5)
+
+        assert shared_result == local_result == 10
+        assert shared_count == 1
+        assert local_count == 1
+
+        # Call again - both should use their respective caches
+        shared_function(5)
+        local_function(5)
+        assert shared_count == 1
+        assert local_count == 1
+
+        # Clear only shared cache
+        shared_function.cache_clear()
+
+        # Shared should recompute, local should still use cache
+        shared_function(5)
+        local_function(5)
+        assert shared_count == 2
+        assert local_count == 1
+
+        # Cleanup
+        shared_function.cache_clear()
+        local_function.cache_clear()
+
+    @pytest.mark.asyncio
+    async def test_shared_cache_concurrent_different_keys(self):
+        """Test that concurrent calls with different keys work correctly."""
+        call_counts = {}
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        async def multi_key_function(key: str) -> str:
+            if key not in call_counts:
+                call_counts[key] = 0
+            call_counts[key] += 1
+            await asyncio.sleep(0.05)
+            return f"result_{key}"
+
+        # Clear cache
+        multi_key_function.cache_clear()
+
+        # Launch concurrent tasks with different keys
+        keys = ["a", "b", "c", "d", "e"]
+        tasks = []
+        for key in keys:
+            # Multiple calls per key
+            tasks.extend([multi_key_function(key) for _ in range(3)])
+
+        results = await asyncio.gather(*tasks)
+
+        # Verify results
+        for i, key in enumerate(keys):
+            expected = f"result_{key}"
+            # Each key appears 3 times in results
+            key_results = results[i * 3 : (i + 1) * 3]
+            assert all(r == expected for r in key_results)
+
+        # Each key should only be computed once
+        for key in keys:
+            assert call_counts[key] == 1
+
+        # Cleanup
+        multi_key_function.cache_clear()
+
+    def test_shared_cache_performance_comparison(self):
+        """Compare performance of shared vs local cache."""
+        import statistics
+
+        shared_times = []
+        local_times = []
+
+        @cached(ttl_seconds=30, shared_cache=True)
+        def shared_perf_function(x: int) -> int:
+            time.sleep(0.01)  # Simulate work
+            return x * 2
+
+        @cached(ttl_seconds=30, shared_cache=False)
+        def local_perf_function(x: int) -> int:
+            time.sleep(0.01)  # Simulate work
+            return x * 2
+
+        # Clear caches
+        shared_perf_function.cache_clear()
+        local_perf_function.cache_clear()
+
+        # Warm up both caches
+        for i in range(5):
+            shared_perf_function(i)
+            local_perf_function(i)
+
+        # Measure cache hit times
+        for i in range(5):
+            # Shared cache hit
+            start = time.time()
+            shared_perf_function(i)
+            shared_times.append(time.time() - start)
+
+            # Local cache hit
+            start = time.time()
+            local_perf_function(i)
+            local_times.append(time.time() - start)
+
+        # Local cache should be faster (no Redis round-trip)
+        avg_shared = statistics.mean(shared_times)
+        avg_local = statistics.mean(local_times)
+
+        print(f"Avg shared cache hit time: {avg_shared:.6f}s")
+        print(f"Avg local cache hit time: {avg_local:.6f}s")
+
+        # Local should be significantly faster for cache hits
+        # Redis adds network latency even for cache hits
+        assert avg_local < avg_shared
+
+        # Cleanup
+        shared_perf_function.cache_clear()
+        local_perf_function.cache_clear()

autogpt_platform/backend/backend/util/clients.py

@@ -4,8 +4,7 @@ Centralized service client helpers with thread caching.
 
 from typing import TYPE_CHECKING
 
-from autogpt_libs.utils.cache import cached, thread_cached
-
+from backend.util.cache import cached, thread_cached
 from backend.util.settings import Settings
 
 settings = Settings()
@@ -34,12 +33,14 @@ def get_database_manager_client() -> "DatabaseManagerClient":
 
 
 @thread_cached
-def get_database_manager_async_client() -> "DatabaseManagerAsyncClient":
+def get_database_manager_async_client(
+    should_retry: bool = True,
+) -> "DatabaseManagerAsyncClient":
     """Get a thread-cached DatabaseManagerAsyncClient with request retry enabled."""
     from backend.executor import DatabaseManagerAsyncClient
     from backend.util.service import get_service_client
 
-    return get_service_client(DatabaseManagerAsyncClient, request_retry=True)
+    return get_service_client(DatabaseManagerAsyncClient, request_retry=should_retry)
 
 
 @thread_cached
@@ -118,7 +119,7 @@ def get_integration_credentials_store() -> "IntegrationCredentialsStore":
 # ============ Supabase Clients ============ #
 
 
-@cached()
+@cached(ttl_seconds=3600)
 def get_supabase() -> "Client":
     """Get a process-cached synchronous Supabase client instance."""
     from supabase import create_client
@@ -128,7 +129,7 @@ def get_supabase() -> "Client":
     )
 
 
-@cached()
+@cached(ttl_seconds=3600)
 async def get_async_supabase() -> "AClient":
     """Get a process-cached asynchronous Supabase client instance."""
     from supabase import create_async_client

autogpt_platform/backend/backend/util/exceptions.py

@@ -86,3 +86,9 @@ class GraphValidationError(ValueError):
                 for node_id, errors in self.node_errors.items()
             ]
         )
+
+
+class DatabaseError(Exception):
+    """Raised when there is an error interacting with the database"""
+
+    pass

autogpt_platform/backend/backend/util/feature_flag.py

@@ -5,12 +5,12 @@ from functools import wraps
 from typing import Any, Awaitable, Callable, TypeVar
 
 import ldclient
-from autogpt_libs.utils.cache import cached
 from fastapi import HTTPException
 from ldclient import Context, LDClient
 from ldclient.config import Config
 from typing_extensions import ParamSpec
 
+from backend.util.cache import cached
 from backend.util.settings import Settings
 
 logger = logging.getLogger(__name__)
@@ -35,6 +35,12 @@ class Flag(str, Enum):
     AI_ACTIVITY_STATUS = "ai-agent-execution-summary"
     BETA_BLOCKS = "beta-blocks"
     AGENT_ACTIVITY = "agent-activity"
+    ENABLE_PLATFORM_PAYMENT = "enable-platform-payment"
+
+
+def is_configured() -> bool:
+    """Check if LaunchDarkly is configured with an SDK key."""
+    return bool(settings.secrets.launch_darkly_sdk_key)
 
 
 def get_client() -> LDClient:
@@ -57,9 +63,9 @@ def initialize_launchdarkly() -> None:
     config = Config(sdk_key)
     ldclient.set_config(config)
 
+    global _is_initialized
+    _is_initialized = True
     if ldclient.get().is_initialized():
-        global _is_initialized
-        _is_initialized = True
         logger.info("LaunchDarkly client initialized successfully")
     else:
         logger.error("LaunchDarkly client failed to initialize")
@@ -212,7 +218,8 @@ def feature_flag(
 
                 if not get_client().is_initialized():
                     logger.warning(
-                        f"LaunchDarkly not initialized, using default={default}"
+                        "LaunchDarkly not initialized, "
+                        f"using default {flag_key}={repr(default)}"
                     )
                     is_enabled = default
                 else:
@@ -226,8 +233,9 @@ def feature_flag(
                     else:
                         # Log warning and use default for non-boolean values
                         logger.warning(
-                            f"Feature flag {flag_key} returned non-boolean value: {flag_value} (type: {type(flag_value).__name__}). "
-                            f"Using default={default}"
+                            f"Feature flag {flag_key} returned non-boolean value: "
+                            f"{repr(flag_value)} (type: {type(flag_value).__name__}). "
+                            f"Using default value {repr(default)}"
                         )
                         is_enabled = default
 

autogpt_platform/backend/backend/util/json.py

@@ -1,26 +1,22 @@
-import json
+import logging
 import re
-from typing import Any, Type, TypeGuard, TypeVar, overload
+from typing import Any, Type, TypeVar, overload
 
 import jsonschema
 import orjson
-from fastapi.encoders import jsonable_encoder
+from fastapi.encoders import jsonable_encoder as to_dict
 from prisma import Json
-from pydantic import BaseModel
 
+from .truncate import truncate
 from .type import type_match
 
+logger = logging.getLogger(__name__)
+
 # Precompiled regex to remove PostgreSQL-incompatible control characters
 # Removes \u0000-\u0008, \u000B-\u000C, \u000E-\u001F, \u007F (keeps tab \u0009, newline \u000A, carriage return \u000D)
 POSTGRES_CONTROL_CHARS = re.compile(r"[\x00-\x08\x0B-\x0C\x0E-\x1F\x7F]")
 
 
-def to_dict(data) -> dict:
-    if isinstance(data, BaseModel):
-        data = data.model_dump()
-    return jsonable_encoder(data)
-
-
 def dumps(
     data: Any, *args: Any, indent: int | None = None, option: int = 0, **kwargs: Any
 ) -> str:
@@ -109,34 +105,57 @@ def validate_with_jsonschema(
         return str(e)
 
 
-def is_list_of_basemodels(value: object) -> TypeGuard[list[BaseModel]]:
-    return isinstance(value, list) and all(
-        isinstance(item, BaseModel) for item in value
-    )
+def _sanitize_string(value: str) -> str:
+    """Remove PostgreSQL-incompatible control characters from string."""
+    return POSTGRES_CONTROL_CHARS.sub("", value)
 
 
-def convert_pydantic_to_json(output_data: Any) -> Any:
-    if isinstance(output_data, BaseModel):
-        return output_data.model_dump()
-    if is_list_of_basemodels(output_data):
-        return [item.model_dump() for item in output_data]
-    return output_data
+def sanitize_json(data: Any) -> Any:
+    try:
+        # Use two-pass approach for consistent string sanitization:
+        # 1. First convert to basic JSON-serializable types (handles Pydantic models)
+        # 2. Then sanitize strings in the result
+        basic_result = to_dict(data)
+        return to_dict(basic_result, custom_encoder={str: _sanitize_string})
+    except Exception as e:
+        # Log the failure and fall back to string representation
+        logger.error(
+            "SafeJson fallback to string representation due to serialization error: %s (%s). "
+            "Data type: %s, Data preview: %s",
+            type(e).__name__,
+            truncate(str(e), 200),
+            type(data).__name__,
+            truncate(str(data), 100),
+        )
+
+        # Ultimate fallback: convert to string representation and sanitize
+        return _sanitize_string(str(data))
 
 
-def SafeJson(data: Any) -> Json:
+class SafeJson(Json):
     """
     Safely serialize data and return Prisma's Json type.
-    Sanitizes null bytes to prevent PostgreSQL 22P05 errors.
-    """
-    if isinstance(data, BaseModel):
-        json_string = data.model_dump_json(
-            warnings="error",
-            exclude_none=True,
-            fallback=lambda v: None,
-        )
-    else:
-        json_string = dumps(data, default=lambda v: None)
+    Sanitizes control characters to prevent PostgreSQL 22P05 errors.
 
-    # Remove PostgreSQL-incompatible control characters in single regex operation
-    sanitized_json = POSTGRES_CONTROL_CHARS.sub("", json_string)
-    return Json(json.loads(sanitized_json))
+    This function:
+    1. Converts Pydantic models to dicts (recursively using to_dict)
+    2. Recursively removes PostgreSQL-incompatible control characters from strings
+    3. Returns a Prisma Json object safe for database storage
+
+    Uses to_dict (jsonable_encoder) with a custom encoder to handle both Pydantic
+    conversion and control character sanitization in a two-pass approach.
+
+    Args:
+        data: Input data to sanitize and convert to Json
+
+    Returns:
+        Prisma Json object with control characters removed
+
+    Examples:
+        >>> SafeJson({"text": "Hello\\x00World"})  # null char removed
+        >>> SafeJson({"path": "C:\\\\temp"})  # backslashes preserved
+        >>> SafeJson({"data": "Text\\\\u0000here"})  # literal backslash-u preserved
+    """
+
+    def __init__(self, data: Any):
+        super().__init__(sanitize_json(data))

autogpt_platform/backend/backend/util/logging.py

@@ -8,10 +8,7 @@ settings = Settings()
 def configure_logging():
     import autogpt_libs.logging.config
 
-    if (
-        settings.config.behave_as == BehaveAs.LOCAL
-        or settings.config.app_env == AppEnvironment.LOCAL
-    ):
+    if not is_structured_logging_enabled():
         autogpt_libs.logging.config.configure_logging(force_cloud_logging=False)
     else:
         autogpt_libs.logging.config.configure_logging(force_cloud_logging=True)
@@ -20,6 +17,14 @@ def configure_logging():
     logging.getLogger("httpx").setLevel(logging.WARNING)
 
 
+def is_structured_logging_enabled() -> bool:
+    """Check if structured logging (cloud logging) is enabled."""
+    return not (
+        settings.config.behave_as == BehaveAs.LOCAL
+        or settings.config.app_env == AppEnvironment.LOCAL
+    )
+
+
 class TruncatedLogger:
     def __init__(
         self,

autogpt_platform/backend/backend/util/metrics.py

@@ -3,13 +3,17 @@ from enum import Enum
 
 import sentry_sdk
 from pydantic import SecretStr
+from sentry_sdk.integrations import DidNotEnable
 from sentry_sdk.integrations.anthropic import AnthropicIntegration
 from sentry_sdk.integrations.asyncio import AsyncioIntegration
+from sentry_sdk.integrations.launchdarkly import LaunchDarklyIntegration
 from sentry_sdk.integrations.logging import LoggingIntegration
 
+from backend.util import feature_flag
 from backend.util.settings import Settings
 
 settings = Settings()
+logger = logging.getLogger(__name__)
 
 
 class DiscordChannel(str, Enum):
@@ -19,6 +23,12 @@ class DiscordChannel(str, Enum):
 
 def sentry_init():
     sentry_dsn = settings.secrets.sentry_dsn
+    integrations = []
+    if feature_flag.is_configured():
+        try:
+            integrations.append(LaunchDarklyIntegration(feature_flag.get_client()))
+        except DidNotEnable as e:
+            logger.error(f"Error enabling LaunchDarklyIntegration for Sentry: {e}")
     sentry_sdk.init(
         dsn=sentry_dsn,
         traces_sample_rate=1.0,
@@ -31,7 +41,8 @@ def sentry_init():
             AnthropicIntegration(
                 include_prompts=False,
             ),
-        ],
+        ]
+        + integrations,
     )
 
 

autogpt_platform/backend/backend/util/process.py

@@ -8,18 +8,9 @@ from typing import Optional
 
 from backend.util.logging import configure_logging
 from backend.util.metrics import sentry_init
+from backend.util.settings import set_service_name
 
 logger = logging.getLogger(__name__)
-_SERVICE_NAME = "MainProcess"
-
-
-def get_service_name():
-    return _SERVICE_NAME
-
-
-def set_service_name(name: str):
-    global _SERVICE_NAME
-    _SERVICE_NAME = name
 
 
 class AppProcess(ABC):
@@ -28,7 +19,8 @@ class AppProcess(ABC):
     """
 
     process: Optional[Process] = None
-    cleaned_up = False
+    _shutting_down: bool = False
+    _cleaned_up: bool = False
 
     if "forkserver" in get_all_start_methods():
         set_start_method("forkserver", force=True)
@@ -52,7 +44,6 @@ class AppProcess(ABC):
     def service_name(self) -> str:
         return self.__class__.__name__
 
-    @abstractmethod
     def cleanup(self):
         """
         Implement this method on a subclass to do post-execution cleanup,
@@ -74,7 +65,8 @@ class AppProcess(ABC):
             self.run()
         except BaseException as e:
             logger.warning(
-                f"[{self.service_name}] Termination request: {type(e).__name__}; {e} executing cleanup."
+                f"[{self.service_name}] 🛑 Terminating because of {type(e).__name__}: {e}",  # noqa
+                exc_info=e if not isinstance(e, SystemExit) else None,
             )
             # Send error to Sentry before cleanup
             if not isinstance(e, (KeyboardInterrupt, SystemExit)):
@@ -85,8 +77,12 @@ class AppProcess(ABC):
                 except Exception:
                     pass  # Silently ignore if Sentry isn't available
         finally:
-            self.cleanup()
-            logger.info(f"[{self.service_name}] Terminated.")
+            if not self._cleaned_up:
+                self._cleaned_up = True
+                logger.info(f"[{self.service_name}] 🧹 Running cleanup")
+                self.cleanup()
+                logger.info(f"[{self.service_name}] ✅ Cleanup done")
+            logger.info(f"[{self.service_name}] 🛑 Terminated")
 
     @staticmethod
     def llprint(message: str):
@@ -97,8 +93,8 @@ class AppProcess(ABC):
         os.write(sys.stdout.fileno(), (message + "\n").encode())
 
     def _self_terminate(self, signum: int, frame):
-        if not self.cleaned_up:
-            self.cleaned_up = True
+        if not self._shutting_down:
+            self._shutting_down = True
             sys.exit(0)
         else:
             self.llprint(

autogpt_platform/backend/backend/util/request.py

@@ -13,7 +13,7 @@ import idna
 from aiohttp import FormData, abc
 from tenacity import retry, retry_if_result, wait_exponential_jitter
 
-from backend.util.json import json
+from backend.util.json import loads
 
 # Retry status codes for which we will automatically retry the request
 THROTTLE_RETRY_STATUS_CODES: set[int] = {429, 500, 502, 503, 504, 408}
@@ -175,10 +175,15 @@ async def validate_url(
                     f"for hostname {ascii_hostname} is not allowed."
                 )
 
+    # Reconstruct the netloc with IDNA-encoded hostname and preserve port
+    netloc = ascii_hostname
+    if parsed.port:
+        netloc = f"{ascii_hostname}:{parsed.port}"
+
     return (
         URL(
             parsed.scheme,
-            ascii_hostname,
+            netloc,
             quote(parsed.path, safe="/%:@"),
             parsed.params,
             parsed.query,
@@ -259,7 +264,7 @@ class Response:
         """
         Parse the body as JSON and return the resulting Python object.
         """
-        return json.loads(
+        return loads(
             self.content.decode(encoding or "utf-8", errors="replace"), **kwargs
         )
 

autogpt_platform/backend/backend/util/retry.py

@@ -13,41 +13,80 @@ from tenacity import (
     wait_exponential_jitter,
 )
 
-from backend.util.process import get_service_name
+from backend.util.settings import get_service_name
 
 logger = logging.getLogger(__name__)
 
 # Alert threshold for excessive retries
 EXCESSIVE_RETRY_THRESHOLD = 50
 
+# Rate limiting for alerts - track last alert time per function+error combination
+_alert_rate_limiter = {}
+_rate_limiter_lock = threading.Lock()
+ALERT_RATE_LIMIT_SECONDS = 300  # 5 minutes between same alerts
+
+
+def should_send_alert(func_name: str, exception: Exception, context: str = "") -> bool:
+    """Check if we should send an alert based on rate limiting."""
+    # Create a unique key for this function+error+context combination
+    error_signature = (
+        f"{context}:{func_name}:{type(exception).__name__}:{str(exception)[:100]}"
+    )
+    current_time = time.time()
+
+    with _rate_limiter_lock:
+        last_alert_time = _alert_rate_limiter.get(error_signature, 0)
+        if current_time - last_alert_time >= ALERT_RATE_LIMIT_SECONDS:
+            _alert_rate_limiter[error_signature] = current_time
+            return True
+        return False
+
+
+def send_rate_limited_discord_alert(
+    func_name: str, exception: Exception, context: str, alert_msg: str, channel=None
+) -> bool:
+    """
+    Send a Discord alert with rate limiting.
+
+    Returns True if alert was sent, False if rate limited.
+    """
+    if not should_send_alert(func_name, exception, context):
+        return False
+
+    try:
+        from backend.util.clients import get_notification_manager_client
+        from backend.util.metrics import DiscordChannel
+
+        notification_client = get_notification_manager_client()
+        notification_client.discord_system_alert(
+            alert_msg, channel or DiscordChannel.PLATFORM
+        )
+        return True
+
+    except Exception as alert_error:
+        logger.error(f"Failed to send Discord alert: {alert_error}")
+        return False
+
 
 def _send_critical_retry_alert(
     func_name: str, attempt_number: int, exception: Exception, context: str = ""
 ):
     """Send alert when a function is approaching the retry failure threshold."""
-    try:
-        # Import here to avoid circular imports
-        from backend.util.clients import get_notification_manager_client
 
-        notification_client = get_notification_manager_client()
-
-        prefix = f"{context}: " if context else ""
-        alert_msg = (
-            f"🚨 CRITICAL: Operation Approaching Failure Threshold: {prefix}'{func_name}'\n\n"
-            f"Current attempt: {attempt_number}/{EXCESSIVE_RETRY_THRESHOLD}\n"
-            f"Error: {type(exception).__name__}: {exception}\n\n"
-            f"This operation is about to fail permanently. Investigate immediately."
-        )
-
-        notification_client.discord_system_alert(alert_msg)
+    prefix = f"{context}: " if context else ""
+    if send_rate_limited_discord_alert(
+        func_name,
+        exception,
+        context,
+        f"🚨 CRITICAL: Operation Approaching Failure Threshold: {prefix}'{func_name}'\n\n"
+        f"Current attempt: {attempt_number}/{EXCESSIVE_RETRY_THRESHOLD}\n"
+        f"Error: {type(exception).__name__}: {exception}\n\n"
+        f"This operation is about to fail permanently. Investigate immediately.",
+    ):
         logger.critical(
             f"CRITICAL ALERT SENT: Operation {func_name} at attempt {attempt_number}"
         )
 
-    except Exception as alert_error:
-        logger.error(f"Failed to send critical retry alert: {alert_error}")
-        # Don't let alerting failures break the main flow
-
 
 def _create_retry_callback(context: str = ""):
     """Create a retry callback with optional context."""
@@ -66,7 +105,7 @@ def _create_retry_callback(context: str = ""):
                 f"{type(exception).__name__}: {exception}"
             )
         else:
-            # Retry attempt - send critical alert only once at threshold
+            # Retry attempt - send critical alert only once at threshold (rate limited)
             if attempt_number == EXCESSIVE_RETRY_THRESHOLD:
                 _send_critical_retry_alert(
                     func_name, attempt_number, exception, context
@@ -131,7 +170,7 @@ def _log_prefix(resource_name: str, conn_id: str):
 def conn_retry(
     resource_name: str,
     action_name: str,
-    max_retry: int = 5,
+    max_retry: int = 100,
     max_wait: float = 30,
 ):
     conn_id = str(uuid4())
@@ -139,10 +178,29 @@ def conn_retry(
     def on_retry(retry_state):
         prefix = _log_prefix(resource_name, conn_id)
         exception = retry_state.outcome.exception()
+        attempt_number = retry_state.attempt_number
+        func_name = getattr(retry_state.fn, "__name__", "unknown")
 
         if retry_state.outcome.failed and retry_state.next_action is None:
             logger.error(f"{prefix} {action_name} failed after retries: {exception}")
         else:
+            if attempt_number == EXCESSIVE_RETRY_THRESHOLD:
+                if send_rate_limited_discord_alert(
+                    func_name,
+                    exception,
+                    f"{resource_name}_infrastructure",
+                    f"🚨 **Critical Infrastructure Connection Issue**\n"
+                    f"Resource: {resource_name}\n"
+                    f"Action: {action_name}\n"
+                    f"Function: {func_name}\n"
+                    f"Current attempt: {attempt_number}/{max_retry + 1}\n"
+                    f"Error: {type(exception).__name__}: {str(exception)[:200]}{'...' if len(str(exception)) > 200 else ''}\n\n"
+                    f"Infrastructure component is approaching failure threshold. Investigate immediately.",
+                ):
+                    logger.critical(
+                        f"INFRASTRUCTURE ALERT SENT: {resource_name} at {attempt_number} attempts"
+                    )
+
             logger.warning(
                 f"{prefix} {action_name} failed: {exception}. Retrying now..."
             )
@@ -218,8 +276,8 @@ def continuous_retry(*, retry_delay: float = 1.0):
 
         @wraps(func)
         async def async_wrapper(*args, **kwargs):
+            counter = 0
             while True:
-                counter = 0
                 try:
                     return await func(*args, **kwargs)
                 except Exception as exc:

autogpt_platform/backend/backend/util/retry_test.py

@@ -1,8 +1,19 @@
 import asyncio
+import threading
+import time
+from unittest.mock import Mock, patch
 
 import pytest
 
-from backend.util.retry import conn_retry
+from backend.util.retry import (
+    ALERT_RATE_LIMIT_SECONDS,
+    _alert_rate_limiter,
+    _rate_limiter_lock,
+    _send_critical_retry_alert,
+    conn_retry,
+    create_retry_decorator,
+    should_send_alert,
+)
 
 
 def test_conn_retry_sync_function():
@@ -47,3 +58,194 @@ async def test_conn_retry_async_function():
     with pytest.raises(ValueError) as e:
         await test_function()
         assert str(e.value) == "Test error"
+
+
+class TestRetryRateLimiting:
+    """Test the rate limiting functionality for critical retry alerts."""
+
+    def setup_method(self):
+        """Reset rate limiter state before each test."""
+        with _rate_limiter_lock:
+            _alert_rate_limiter.clear()
+
+    def test_should_send_alert_allows_first_occurrence(self):
+        """Test that the first occurrence of an error allows alert."""
+        exc = ValueError("test error")
+        assert should_send_alert("test_func", exc, "test_context") is True
+
+    def test_should_send_alert_rate_limits_duplicate(self):
+        """Test that duplicate errors are rate limited."""
+        exc = ValueError("test error")
+
+        # First call should be allowed
+        assert should_send_alert("test_func", exc, "test_context") is True
+
+        # Second call should be rate limited
+        assert should_send_alert("test_func", exc, "test_context") is False
+
+    def test_should_send_alert_allows_different_errors(self):
+        """Test that different errors are allowed even if same function."""
+        exc1 = ValueError("error 1")
+        exc2 = ValueError("error 2")
+
+        # First error should be allowed
+        assert should_send_alert("test_func", exc1, "test_context") is True
+
+        # Different error should also be allowed
+        assert should_send_alert("test_func", exc2, "test_context") is True
+
+    def test_should_send_alert_allows_different_contexts(self):
+        """Test that same error in different contexts is allowed."""
+        exc = ValueError("test error")
+
+        # First context should be allowed
+        assert should_send_alert("test_func", exc, "context1") is True
+
+        # Different context should also be allowed
+        assert should_send_alert("test_func", exc, "context2") is True
+
+    def test_should_send_alert_allows_different_functions(self):
+        """Test that same error in different functions is allowed."""
+        exc = ValueError("test error")
+
+        # First function should be allowed
+        assert should_send_alert("func1", exc, "test_context") is True
+
+        # Different function should also be allowed
+        assert should_send_alert("func2", exc, "test_context") is True
+
+    def test_should_send_alert_respects_time_window(self):
+        """Test that alerts are allowed again after the rate limit window."""
+        exc = ValueError("test error")
+
+        # First call should be allowed
+        assert should_send_alert("test_func", exc, "test_context") is True
+
+        # Immediately after should be rate limited
+        assert should_send_alert("test_func", exc, "test_context") is False
+
+        # Mock time to simulate passage of rate limit window
+        current_time = time.time()
+        with patch("backend.util.retry.time.time") as mock_time:
+            # Simulate time passing beyond rate limit window
+            mock_time.return_value = current_time + ALERT_RATE_LIMIT_SECONDS + 1
+            assert should_send_alert("test_func", exc, "test_context") is True
+
+    def test_should_send_alert_thread_safety(self):
+        """Test that rate limiting is thread-safe."""
+        exc = ValueError("test error")
+        results = []
+
+        def check_alert():
+            result = should_send_alert("test_func", exc, "test_context")
+            results.append(result)
+
+        # Create multiple threads trying to send the same alert
+        threads = [threading.Thread(target=check_alert) for _ in range(10)]
+
+        # Start all threads
+        for thread in threads:
+            thread.start()
+
+        # Wait for all threads to complete
+        for thread in threads:
+            thread.join()
+
+        # Only one thread should have been allowed to send the alert
+        assert sum(results) == 1
+        assert len([r for r in results if r is True]) == 1
+        assert len([r for r in results if r is False]) == 9
+
+    @patch("backend.util.clients.get_notification_manager_client")
+    def test_send_critical_retry_alert_rate_limiting(self, mock_get_client):
+        """Test that _send_critical_retry_alert respects rate limiting."""
+        mock_client = Mock()
+        mock_get_client.return_value = mock_client
+
+        exc = ValueError("spend_credits API error")
+
+        # First alert should be sent
+        _send_critical_retry_alert("spend_credits", 50, exc, "Service communication")
+        assert mock_client.discord_system_alert.call_count == 1
+
+        # Second identical alert should be rate limited (not sent)
+        _send_critical_retry_alert("spend_credits", 50, exc, "Service communication")
+        assert mock_client.discord_system_alert.call_count == 1  # Still 1, not 2
+
+        # Different error should be allowed
+        exc2 = ValueError("different API error")
+        _send_critical_retry_alert("spend_credits", 50, exc2, "Service communication")
+        assert mock_client.discord_system_alert.call_count == 2
+
+    @patch("backend.util.clients.get_notification_manager_client")
+    def test_send_critical_retry_alert_handles_notification_failure(
+        self, mock_get_client
+    ):
+        """Test that notification failures don't break the rate limiter."""
+        mock_client = Mock()
+        mock_client.discord_system_alert.side_effect = Exception("Notification failed")
+        mock_get_client.return_value = mock_client
+
+        exc = ValueError("test error")
+
+        # Should not raise exception even if notification fails
+        _send_critical_retry_alert("test_func", 50, exc, "test_context")
+
+        # Rate limiter should still work for subsequent calls
+        assert should_send_alert("test_func", exc, "test_context") is False
+
+    def test_error_signature_generation(self):
+        """Test that error signatures are generated correctly for rate limiting."""
+        # Test with long exception message (should be truncated to 100 chars)
+        long_message = "x" * 200
+        exc = ValueError(long_message)
+
+        # Should not raise exception and should work normally
+        assert should_send_alert("test_func", exc, "test_context") is True
+        assert should_send_alert("test_func", exc, "test_context") is False
+
+    def test_real_world_scenario_spend_credits_spam(self):
+        """Test the real-world scenario that was causing spam."""
+        # Simulate the exact error that was causing issues
+        exc = Exception(
+            "HTTP 500: Server error '500 Internal Server Error' for url 'http://autogpt-database-manager.prod-agpt.svc.cluster.local:8005/spend_credits'"
+        )
+
+        # First 50 attempts reach threshold - should send alert
+        with patch(
+            "backend.util.clients.get_notification_manager_client"
+        ) as mock_get_client:
+            mock_client = Mock()
+            mock_get_client.return_value = mock_client
+
+            _send_critical_retry_alert(
+                "_call_method_sync", 50, exc, "Service communication"
+            )
+            assert mock_client.discord_system_alert.call_count == 1
+
+            # Next 950 failures should not send alerts (rate limited)
+            for _ in range(950):
+                _send_critical_retry_alert(
+                    "_call_method_sync", 50, exc, "Service communication"
+                )
+
+            # Still only 1 alert sent total
+            assert mock_client.discord_system_alert.call_count == 1
+
+    @patch("backend.util.clients.get_notification_manager_client")
+    def test_retry_decorator_with_excessive_failures(self, mock_get_client):
+        """Test retry decorator behavior when it hits the alert threshold."""
+        mock_client = Mock()
+        mock_get_client.return_value = mock_client
+
+        @create_retry_decorator(
+            max_attempts=60, max_wait=0.1
+        )  # More than EXCESSIVE_RETRY_THRESHOLD, but fast
+        def always_failing_function():
+            raise ValueError("persistent failure")
+
+        with pytest.raises(ValueError):
+            always_failing_function()
+
+        # Should have sent exactly one alert at the threshold
+        assert mock_client.discord_system_alert.call_count == 1

autogpt_platform/backend/backend/util/service.py

@@ -4,9 +4,12 @@ import concurrent.futures
 import inspect
 import logging
 import os
+import signal
+import sys
 import threading
 import time
 from abc import ABC, abstractmethod
+from contextlib import asynccontextmanager
 from functools import update_wrapper
 from typing import (
     Any,
@@ -28,11 +31,12 @@ from fastapi import FastAPI, Request, responses
 from pydantic import BaseModel, TypeAdapter, create_model
 
 import backend.util.exceptions as exceptions
+from backend.monitoring.instrumentation import instrument_fastapi
 from backend.util.json import to_dict
 from backend.util.metrics import sentry_init
-from backend.util.process import AppProcess, get_service_name
+from backend.util.process import AppProcess
 from backend.util.retry import conn_retry, create_retry_decorator
-from backend.util.settings import Config
+from backend.util.settings import Config, get_service_name
 
 logger = logging.getLogger(__name__)
 T = TypeVar("T")
@@ -110,14 +114,44 @@ class BaseAppService(AppProcess, ABC):
         return target_host
 
     def run_service(self) -> None:
-        while True:
-            time.sleep(10)
+        # HACK: run the main event loop outside the main thread to disable Uvicorn's
+        # internal signal handlers, since there is no config option for this :(
+        shared_asyncio_thread = threading.Thread(
+            target=self._run_shared_event_loop,
+            daemon=True,
+            name=f"{self.service_name}-shared-event-loop",
+        )
+        shared_asyncio_thread.start()
+        shared_asyncio_thread.join()
+
+    def _run_shared_event_loop(self) -> None:
+        try:
+            self.shared_event_loop.run_forever()
+        finally:
+            logger.info(f"[{self.service_name}] 🛑 Shared event loop stopped")
+            self.shared_event_loop.close()  # ensure held resources are released
 
     def run_and_wait(self, coro: Coroutine[Any, Any, T]) -> T:
         return asyncio.run_coroutine_threadsafe(coro, self.shared_event_loop).result()
 
     def run(self):
-        self.shared_event_loop = asyncio.get_event_loop()
+        self.shared_event_loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(self.shared_event_loop)
+
+    def cleanup(self):
+        """
+        **💡 Overriding `AppService.lifespan` may be a more convenient option.**
+
+        Implement this method on a subclass to do post-execution cleanup,
+        e.g. disconnecting from a database or terminating child processes.
+
+        **Note:** if you override this method in a subclass, it must call
+        `super().cleanup()` *at the end*!
+        """
+        # Stop the shared event loop to allow resource clean-up
+        self.shared_event_loop.call_soon_threadsafe(self.shared_event_loop.stop)
+
+        super().cleanup()
 
 
 class RemoteCallError(BaseModel):
@@ -178,6 +212,7 @@ EXCEPTION_MAPPING = {
 
 class AppService(BaseAppService, ABC):
     fastapi_app: FastAPI
+    http_server: uvicorn.Server | None = None
     log_level: str = "info"
 
     def set_log_level(self, log_level: str):
@@ -189,11 +224,10 @@ class AppService(BaseAppService, ABC):
     def _handle_internal_http_error(status_code: int = 500, log_error: bool = True):
         def handler(request: Request, exc: Exception):
             if log_error:
-                if status_code == 500:
-                    log = logger.exception
-                else:
-                    log = logger.error
-                log(f"{request.method} {request.url.path} failed: {exc}")
+                logger.error(
+                    f"{request.method} {request.url.path} failed: {exc}",
+                    exc_info=exc if status_code == 500 else None,
+                )
             return responses.JSONResponse(
                 status_code=status_code,
                 content=RemoteCallError(
@@ -255,13 +289,13 @@ class AppService(BaseAppService, ABC):
 
             return sync_endpoint
 
-    @conn_retry("FastAPI server", "Starting FastAPI server")
+    @conn_retry("FastAPI server", "Running FastAPI server")
     def __start_fastapi(self):
         logger.info(
             f"[{self.service_name}] Starting RPC server at http://{api_host}:{self.get_port()}"
         )
 
-        server = uvicorn.Server(
+        self.http_server = uvicorn.Server(
             uvicorn.Config(
                 self.fastapi_app,
                 host=api_host,
@@ -270,18 +304,94 @@ class AppService(BaseAppService, ABC):
                 log_level=self.log_level,
             )
         )
-        self.shared_event_loop.run_until_complete(server.serve())
+        self.run_and_wait(self.http_server.serve())
+
+        # Perform clean-up when the server exits
+        if not self._cleaned_up:
+            self._cleaned_up = True
+            logger.info(f"[{self.service_name}] 🧹 Running cleanup")
+            self.cleanup()
+            logger.info(f"[{self.service_name}] ✅ Cleanup done")
+
+    def _self_terminate(self, signum: int, frame):
+        """Pass SIGTERM to Uvicorn so it can shut down gracefully"""
+        signame = signal.Signals(signum).name
+        if not self._shutting_down:
+            self._shutting_down = True
+            if self.http_server:
+                logger.info(
+                    f"[{self.service_name}] 🛑 Received {signame} ({signum}) - "
+                    "Entering RPC server graceful shutdown"
+                )
+                self.http_server.handle_exit(signum, frame)  # stop accepting requests
+
+                # NOTE: Actually stopping the process is triggered by:
+                # 1. The call to self.cleanup() at the end of __start_fastapi() 👆🏼
+                # 2. BaseAppService.cleanup() stopping the shared event loop
+            else:
+                logger.warning(
+                    f"[{self.service_name}] {signame} received before HTTP server init."
+                    " Terminating..."
+                )
+                sys.exit(0)
+
+        else:
+            # Expedite shutdown on second SIGTERM
+            logger.info(
+                f"[{self.service_name}] 🛑🛑 Received {signame} ({signum}), "
+                "but shutdown is already underway. Terminating..."
+            )
+            sys.exit(0)
+
+    @asynccontextmanager
+    async def lifespan(self, app: FastAPI):
+        """
+        The FastAPI/Uvicorn server's lifespan manager, used for setup and shutdown.
+
+        You can extend and use this in a subclass like:
+        ```
+        @asynccontextmanager
+        async def lifespan(self, app: FastAPI):
+            async with super().lifespan(app):
+                await db.connect()
+                yield
+                await db.disconnect()
+        ```
+        """
+        # Startup - this runs before Uvicorn starts accepting connections
+
+        yield
+
+        # Shutdown - this runs when FastAPI/Uvicorn shuts down
+        logger.info(f"[{self.service_name}] ✅ FastAPI has finished")
 
     async def health_check(self) -> str:
-        """
-        A method to check the health of the process.
-        """
+        """A method to check the health of the process."""
         return "OK"
 
     def run(self):
         sentry_init()
         super().run()
-        self.fastapi_app = FastAPI()
+
+        self.fastapi_app = FastAPI(lifespan=self.lifespan)
+
+        # Add Prometheus instrumentation to all services
+        try:
+            instrument_fastapi(
+                self.fastapi_app,
+                service_name=self.service_name,
+                expose_endpoint=True,
+                endpoint="/metrics",
+                include_in_schema=False,
+            )
+        except ImportError:
+            logger.warning(
+                f"Prometheus instrumentation not available for {self.service_name}"
+            )
+        except Exception as e:
+            logger.error(
+                f"Failed to instrument {self.service_name} with Prometheus: {e}"
+            )
 
         # Register the exposed API routes.
         for attr_name, attr in vars(type(self)).items():
@@ -306,7 +416,11 @@ class AppService(BaseAppService, ABC):
         )
 
         # Start the FastAPI server in a separate thread.
-        api_thread = threading.Thread(target=self.__start_fastapi, daemon=True)
+        api_thread = threading.Thread(
+            target=self.__start_fastapi,
+            daemon=True,
+            name=f"{self.service_name}-http-server",
+        )
         api_thread.start()
 
         # Run the main service loop (blocking).

autogpt_platform/backend/backend/util/service_test.py

@@ -1,3 +1,5 @@
+import asyncio
+import contextlib
 import time
 from functools import cached_property
 from unittest.mock import Mock
@@ -18,20 +20,11 @@ from backend.util.service import (
 TEST_SERVICE_PORT = 8765
 
 
-def wait_for_service_ready(service_client_type, timeout_seconds=30):
-    """Helper method to wait for a service to be ready using health check with retry."""
-    client = get_service_client(service_client_type, request_retry=True)
-    client.health_check()  # This will retry until service is ready
-
-
 class ServiceTest(AppService):
     def __init__(self):
         super().__init__()
         self.fail_count = 0
 
-    def cleanup(self):
-        pass
-
     @classmethod
     def get_port(cls) -> int:
         return TEST_SERVICE_PORT
@@ -41,10 +34,17 @@ class ServiceTest(AppService):
         result = super().__enter__()
 
         # Wait for the service to be ready
-        wait_for_service_ready(ServiceTestClient)
+        self.wait_until_ready()
 
         return result
 
+    def wait_until_ready(self, timeout_seconds: int = 5):
+        """Helper method to wait for a service to be ready using health check with retry."""
+        client = get_service_client(
+            ServiceTestClient, call_timeout=timeout_seconds, request_retry=True
+        )
+        client.health_check()  # This will retry until service is ready\
+
     @expose
     def add(self, a: int, b: int) -> int:
         return a + b
@@ -490,3 +490,167 @@ class TestHTTPErrorRetryBehavior:
                 )
 
             assert exc_info.value.status_code == status_code
+
+
+class TestGracefulShutdownService(AppService):
+    """Test service with slow endpoints for testing graceful shutdown"""
+
+    @classmethod
+    def get_port(cls) -> int:
+        return 18999  # Use a specific test port
+
+    def __init__(self):
+        super().__init__()
+        self.request_log = []
+        self.cleanup_called = False
+        self.cleanup_completed = False
+
+    @expose
+    async def slow_endpoint(self, duration: int = 5) -> dict:
+        """Endpoint that takes time to complete"""
+        start_time = time.time()
+        self.request_log.append(f"slow_endpoint started at {start_time}")
+
+        await asyncio.sleep(duration)
+
+        end_time = time.time()
+        result = {
+            "message": "completed",
+            "duration": end_time - start_time,
+            "start_time": start_time,
+            "end_time": end_time,
+        }
+        self.request_log.append(f"slow_endpoint completed at {end_time}")
+        return result
+
+    @expose
+    def fast_endpoint(self) -> dict:
+        """Fast endpoint for testing rejection during shutdown"""
+        timestamp = time.time()
+        self.request_log.append(f"fast_endpoint called at {timestamp}")
+        return {"message": "fast", "timestamp": timestamp}
+
+    def cleanup(self):
+        """Override cleanup to track when it's called"""
+        self.cleanup_called = True
+        self.request_log.append(f"cleanup started at {time.time()}")
+
+        # Call parent cleanup
+        super().cleanup()
+
+        self.cleanup_completed = True
+        self.request_log.append(f"cleanup completed at {time.time()}")
+
+
+@pytest.fixture(scope="function")
+async def test_service():
+    """Run the test service in a separate process"""
+
+    service = TestGracefulShutdownService()
+    service.start(background=True)
+
+    base_url = f"http://localhost:{service.get_port()}"
+
+    await wait_until_service_ready(base_url)
+    yield service, base_url
+
+    service.stop()
+
+
+async def wait_until_service_ready(base_url: str, timeout: float = 10):
+    start_time = time.time()
+    while time.time() - start_time <= timeout:
+        async with httpx.AsyncClient(timeout=5) as client:
+            with contextlib.suppress(httpx.ConnectError):
+                response = await client.get(f"{base_url}/health_check", timeout=5)
+
+                if response.status_code == 200 and response.json() == "OK":
+                    return
+
+        await asyncio.sleep(0.5)
+
+    raise RuntimeError(f"Service at {base_url} not available after {timeout} seconds")
+
+
+async def send_slow_request(base_url: str) -> dict:
+    """Send a slow request and return the result"""
+    async with httpx.AsyncClient(timeout=30) as client:
+        response = await client.post(f"{base_url}/slow_endpoint", json={"duration": 5})
+        assert response.status_code == 200
+        return response.json()
+
+
+@pytest.mark.asyncio
+async def test_graceful_shutdown(test_service):
+    """Test that AppService handles graceful shutdown correctly"""
+    service, test_service_url = test_service
+
+    # Start a slow request that should complete even after shutdown
+    slow_task = asyncio.create_task(send_slow_request(test_service_url))
+
+    # Give the slow request time to start
+    await asyncio.sleep(1)
+
+    # Send SIGTERM to the service process
+    shutdown_start_time = time.time()
+    service.process.terminate()  # This sends SIGTERM
+
+    # Wait a moment for shutdown to start
+    await asyncio.sleep(0.5)
+
+    # Try to send a new request - should be rejected or connection refused
+    try:
+        async with httpx.AsyncClient(timeout=5) as client:
+            response = await client.post(f"{test_service_url}/fast_endpoint", json={})
+            # Should get 503 Service Unavailable during shutdown
+            assert response.status_code == 503
+            assert "shutting down" in response.json()["detail"].lower()
+    except httpx.ConnectError:
+        # Connection refused is also acceptable - server stopped accepting
+        pass
+
+    # The slow request should still complete successfully
+    slow_result = await slow_task
+    assert slow_result["message"] == "completed"
+    assert 4.9 < slow_result["duration"] < 5.5  # Should have taken ~5 seconds
+
+    # Wait for the service to fully shut down
+    service.process.join(timeout=15)
+    shutdown_end_time = time.time()
+
+    # Verify the service actually terminated
+    assert not service.process.is_alive()
+
+    # Verify shutdown took reasonable time (slow request - 1s + cleanup)
+    shutdown_duration = shutdown_end_time - shutdown_start_time
+    assert 4 <= shutdown_duration <= 6  # ~5s request - 1s + buffer
+
+    print(f"Shutdown took {shutdown_duration:.2f} seconds")
+    print(f"Slow request completed in: {slow_result['duration']:.2f} seconds")
+
+
+@pytest.mark.asyncio
+async def test_health_check_during_shutdown(test_service):
+    """Test that health checks behave correctly during shutdown"""
+    service, test_service_url = test_service
+
+    # Health check should pass initially
+    async with httpx.AsyncClient(timeout=5) as client:
+        response = await client.get(f"{test_service_url}/health_check")
+        assert response.status_code == 200
+
+    # Send SIGTERM
+    service.process.terminate()
+
+    # Wait for shutdown to begin
+    await asyncio.sleep(1)
+
+    # Health check should now fail or connection should be refused
+    try:
+        async with httpx.AsyncClient(timeout=5) as client:
+            response = await client.get(f"{test_service_url}/health_check")
+            # Could either get 503, 500 (unhealthy), or connection error
+            assert response.status_code in [500, 503]
+    except (httpx.ConnectError, httpx.ConnectTimeout):
+        # Connection refused/timeout is also acceptable
+        pass

autogpt_platform/backend/backend/util/settings.py

@@ -15,6 +15,17 @@ from backend.util.data import get_data_path
 
 T = TypeVar("T", bound=BaseSettings)
 
+_SERVICE_NAME = "MainProcess"
+
+
+def get_service_name():
+    return _SERVICE_NAME
+
+
+def set_service_name(name: str):
+    global _SERVICE_NAME
+    _SERVICE_NAME = name
+
 
 class AppEnvironment(str, Enum):
     LOCAL = "local"
@@ -148,6 +159,12 @@ class Config(UpdateTrackingModel["Config"], BaseSettings):
         default=60 * 60,
         description="Time in seconds for how far back to check for the late executions.",
     )
+    max_concurrent_graph_executions_per_user: int = Field(
+        default=25,
+        ge=1,
+        le=1000,
+        description="Maximum number of concurrent graph executions allowed per user per graph.",
+    )
 
     block_error_rate_threshold: float = Field(
         default=0.5,
@@ -248,6 +265,7 @@ class Config(UpdateTrackingModel["Config"], BaseSettings):
         default="localhost",
         description="The host for the RabbitMQ server",
     )
+
     rabbitmq_port: int = Field(
         default=5672,
         description="The port for the RabbitMQ server",
@@ -258,6 +276,21 @@ class Config(UpdateTrackingModel["Config"], BaseSettings):
         description="The vhost for the RabbitMQ server",
     )
 
+    redis_host: str = Field(
+        default="localhost",
+        description="The host for the Redis server",
+    )
+
+    redis_port: int = Field(
+        default=6379,
+        description="The port for the Redis server",
+    )
+
+    redis_password: str = Field(
+        default="",
+        description="The password for the Redis server (empty string if no password)",
+    )
+
     postmark_sender_email: str = Field(
         default="invalid@invalid.com",
         description="The email address to use for sending emails",

autogpt_platform/backend/backend/util/test_json.py

@@ -1,5 +1,5 @@
 import datetime
-from typing import Any, Optional
+from typing import Any, Optional, cast
 
 from prisma import Json
 from pydantic import BaseModel
@@ -231,6 +231,14 @@ class TestSafeJson:
         result = SafeJson(problematic_data)
         assert isinstance(result, Json)
 
+        # Verify that dangerous control characters are actually removed
+        result_data = result.data
+        assert "\x00" not in str(result_data)  # null byte removed
+        assert "\x07" not in str(result_data)  # bell removed
+        assert "\x0C" not in str(result_data)  # form feed removed
+        assert "\x1B" not in str(result_data)  # escape removed
+        assert "\x7F" not in str(result_data)  # delete removed
+
         # Test that safe whitespace characters are preserved
         safe_data = {
             "with_tab": "text with \t tab",
@@ -241,3 +249,508 @@ class TestSafeJson:
 
         safe_result = SafeJson(safe_data)
         assert isinstance(safe_result, Json)
+
+        # Verify safe characters are preserved
+        safe_result_data = cast(dict[str, Any], safe_result.data)
+        assert isinstance(safe_result_data, dict)
+        with_tab = safe_result_data.get("with_tab", "")
+        with_newline = safe_result_data.get("with_newline", "")
+        with_carriage_return = safe_result_data.get("with_carriage_return", "")
+        assert "\t" in str(with_tab)  # tab preserved
+        assert "\n" in str(with_newline)  # newline preserved
+        assert "\r" in str(with_carriage_return)  # carriage return preserved
+
+    def test_web_scraping_content_sanitization(self):
+        """Test sanitization of typical web scraping content with null characters."""
+        # Simulate web content that might contain null bytes from SearchTheWebBlock
+        web_content = "Article title\x00Hidden null\x01Start of heading\x08Backspace\x0CForm feed content\x1FUnit separator\x7FDelete char"
+
+        result = SafeJson(web_content)
+        assert isinstance(result, Json)
+
+        # Verify all problematic characters are removed
+        sanitized_content = str(result.data)
+        assert "\x00" not in sanitized_content
+        assert "\x01" not in sanitized_content
+        assert "\x08" not in sanitized_content
+        assert "\x0C" not in sanitized_content
+        assert "\x1F" not in sanitized_content
+        assert "\x7F" not in sanitized_content
+
+        # Verify the content is still readable
+        assert "Article title" in sanitized_content
+        assert "Hidden null" in sanitized_content
+        assert "content" in sanitized_content
+
+    def test_legitimate_code_preservation(self):
+        """Test that legitimate code with backslashes and escapes is preserved."""
+        # File paths with backslashes should be preserved
+        file_paths = {
+            "windows_path": "C:\\Users\\test\\file.txt",
+            "network_path": "\\\\server\\share\\folder",
+            "escaped_backslashes": "String with \\\\ double backslashes",
+        }
+
+        result = SafeJson(file_paths)
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify file paths are preserved correctly (JSON converts \\\\ back to \\)
+        windows_path = result_data.get("windows_path", "")
+        network_path = result_data.get("network_path", "")
+        escaped_backslashes = result_data.get("escaped_backslashes", "")
+        assert "C:\\Users\\test\\file.txt" in str(windows_path)
+        assert "\\server\\share" in str(network_path)
+        assert "\\" in str(escaped_backslashes)
+
+    def test_legitimate_json_escapes_preservation(self):
+        """Test that legitimate JSON escape sequences are preserved."""
+        # These should all be preserved as they're valid and useful
+        legitimate_escapes = {
+            "quotes": 'He said "Hello world!"',
+            "newlines": "Line 1\\nLine 2\\nLine 3",
+            "tabs": "Column1\\tColumn2\\tColumn3",
+            "unicode_chars": "Unicode: \u0048\u0065\u006c\u006c\u006f",  # "Hello"
+            "mixed_content": "Path: C:\\\\temp\\\\file.txt\\nSize: 1024 bytes",
+        }
+
+        result = SafeJson(legitimate_escapes)
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify all legitimate content is preserved
+        quotes = result_data.get("quotes", "")
+        newlines = result_data.get("newlines", "")
+        tabs = result_data.get("tabs", "")
+        unicode_chars = result_data.get("unicode_chars", "")
+        mixed_content = result_data.get("mixed_content", "")
+
+        assert '"' in str(quotes)
+        assert "Line 1" in str(newlines) and "Line 2" in str(newlines)
+        assert "Column1" in str(tabs) and "Column2" in str(tabs)
+        assert "Hello" in str(unicode_chars)  # Unicode should be decoded
+        assert "C:" in str(mixed_content) and "temp" in str(mixed_content)
+
+    def test_regex_patterns_dont_over_match(self):
+        """Test that our regex patterns don't accidentally match legitimate sequences."""
+        # Edge cases that could be problematic for regex
+        edge_cases = {
+            "file_with_b": "C:\\\\mybfile.txt",  # Contains 'bf' but not escape sequence
+            "file_with_f": "C:\\\\folder\\\\file.txt",  # Contains 'f' after backslashes
+            "json_like_string": '{"text": "\\\\bolder text"}',  # Looks like JSON escape but isn't
+            "unicode_like": "Code: \\\\u0040 (not a real escape)",  # Looks like Unicode escape
+        }
+
+        result = SafeJson(edge_cases)
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify edge cases are handled correctly - no content should be lost
+        file_with_b = result_data.get("file_with_b", "")
+        file_with_f = result_data.get("file_with_f", "")
+        json_like_string = result_data.get("json_like_string", "")
+        unicode_like = result_data.get("unicode_like", "")
+
+        assert "mybfile.txt" in str(file_with_b)
+        assert "folder" in str(file_with_f) and "file.txt" in str(file_with_f)
+        assert "bolder text" in str(json_like_string)
+        assert "\\u0040" in str(unicode_like)
+
+    def test_programming_code_preservation(self):
+        """Test that programming code with various escapes is preserved."""
+        # Common programming patterns that should be preserved
+        code_samples = {
+            "python_string": 'print("Hello\\\\nworld")',
+            "regex_pattern": "\\\\b[A-Za-z]+\\\\b",  # Word boundary regex
+            "json_string": '{"name": "test", "path": "C:\\\\\\\\folder"}',
+            "sql_escape": "WHERE name LIKE '%\\\\%%'",
+            "javascript": 'var path = "C:\\\\\\\\Users\\\\\\\\file.js";',
+        }
+
+        result = SafeJson(code_samples)
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify programming code is preserved
+        python_string = result_data.get("python_string", "")
+        regex_pattern = result_data.get("regex_pattern", "")
+        json_string = result_data.get("json_string", "")
+        sql_escape = result_data.get("sql_escape", "")
+        javascript = result_data.get("javascript", "")
+
+        assert "print(" in str(python_string)
+        assert "Hello" in str(python_string)
+        assert "[A-Za-z]+" in str(regex_pattern)
+        assert "name" in str(json_string)
+        assert "LIKE" in str(sql_escape)
+        assert "var path" in str(javascript)
+
+    def test_only_problematic_sequences_removed(self):
+        """Test that ONLY PostgreSQL-problematic sequences are removed, nothing else."""
+        # Mix of problematic and safe content (using actual control characters)
+        mixed_content = {
+            "safe_and_unsafe": "Good text\twith tab\x00NULL BYTE\nand newline\x08BACKSPACE",
+            "file_path_with_null": "C:\\temp\\file\x00.txt",
+            "json_with_controls": '{"text": "data\x01\x0C\x1F"}',
+        }
+
+        result = SafeJson(mixed_content)
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify only problematic characters are removed
+        safe_and_unsafe = result_data.get("safe_and_unsafe", "")
+        file_path_with_null = result_data.get("file_path_with_null", "")
+
+        assert "Good text" in str(safe_and_unsafe)
+        assert "\t" in str(safe_and_unsafe)  # Tab preserved
+        assert "\n" in str(safe_and_unsafe)  # Newline preserved
+        assert "\x00" not in str(safe_and_unsafe)  # Null removed
+        assert "\x08" not in str(safe_and_unsafe)  # Backspace removed
+
+        assert "C:\\temp\\file" in str(file_path_with_null)
+        assert ".txt" in str(file_path_with_null)
+        assert "\x00" not in str(file_path_with_null)  # Null removed from path
+
+    def test_invalid_escape_error_prevention(self):
+        """Test that SafeJson prevents 'Invalid \\escape' errors that occurred in upsert_execution_output."""
+        # This reproduces the exact scenario that was causing the error:
+        # POST /upsert_execution_output failed: Invalid \escape: line 1 column 36404 (char 36403)
+
+        # Create data with various problematic escape sequences that could cause JSON parsing errors
+        problematic_output_data = {
+            "web_content": "Article text\x00with null\x01and control\x08chars\x0C\x1F\x7F",
+            "file_path": "C:\\Users\\test\\file\x00.txt",
+            "json_like_string": '{"text": "data\x00\x08\x1F"}',
+            "escaped_sequences": "Text with \\u0000 and \\u0008 sequences",
+            "mixed_content": "Normal text\tproperly\nformatted\rwith\x00invalid\x08chars\x1Fmixed",
+            "large_text": "A" * 35000
+            + "\x00\x08\x1F"
+            + "B" * 5000,  # Large text like in the error
+        }
+
+        # This should not raise any JSON parsing errors
+        result = SafeJson(problematic_output_data)
+        assert isinstance(result, Json)
+
+        # Verify the result is a valid Json object that can be safely stored in PostgreSQL
+        result_data = cast(dict[str, Any], result.data)
+        assert isinstance(result_data, dict)
+
+        # Verify problematic characters are removed but safe content preserved
+        web_content = result_data.get("web_content", "")
+        file_path = result_data.get("file_path", "")
+        large_text = result_data.get("large_text", "")
+
+        # Check that control characters are removed
+        assert "\x00" not in str(web_content)
+        assert "\x01" not in str(web_content)
+        assert "\x08" not in str(web_content)
+        assert "\x0C" not in str(web_content)
+        assert "\x1F" not in str(web_content)
+        assert "\x7F" not in str(web_content)
+
+        # Check that legitimate content is preserved
+        assert "Article text" in str(web_content)
+        assert "with null" in str(web_content)
+        assert "and control" in str(web_content)
+        assert "chars" in str(web_content)
+
+        # Check file path handling
+        assert "C:\\Users\\test\\file" in str(file_path)
+        assert ".txt" in str(file_path)
+        assert "\x00" not in str(file_path)
+
+        # Check large text handling (the scenario from the error at char 36403)
+        assert len(str(large_text)) > 35000  # Content preserved
+        assert "A" * 1000 in str(large_text)  # A's preserved
+        assert "B" * 1000 in str(large_text)  # B's preserved
+        assert "\x00" not in str(large_text)  # Control chars removed
+        assert "\x08" not in str(large_text)
+        assert "\x1F" not in str(large_text)
+
+        # Most importantly: ensure the result can be JSON-serialized without errors
+        # This would have failed with the old approach
+        import json
+
+        json_string = json.dumps(result.data)  # Should not raise "Invalid \escape"
+        assert len(json_string) > 0
+
+        # And can be parsed back
+        parsed_back = json.loads(json_string)
+        assert isinstance(parsed_back, dict)
+
+    def test_dict_containing_pydantic_models(self):
+        """Test that dicts containing Pydantic models are properly serialized."""
+        # This reproduces the bug from PR #11187 where credential_inputs failed
+        model1 = SamplePydanticModel(name="Alice", age=30)
+        model2 = SamplePydanticModel(name="Bob", age=25)
+
+        data = {
+            "user1": model1,
+            "user2": model2,
+            "regular_data": "test",
+        }
+
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        # Verify it can be JSON serialized (this was the bug)
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "Alice" in json_string
+        assert "Bob" in json_string
+
+    def test_nested_pydantic_in_dict(self):
+        """Test deeply nested Pydantic models in dicts."""
+        inner_model = SamplePydanticModel(name="Inner", age=20)
+        middle_model = SamplePydanticModel(
+            name="Middle", age=30, metadata={"inner": inner_model}
+        )
+
+        data = {
+            "level1": {
+                "level2": {
+                    "model": middle_model,
+                    "other": "data",
+                }
+            }
+        }
+
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "Middle" in json_string
+        assert "Inner" in json_string
+
+    def test_list_containing_pydantic_models_in_dict(self):
+        """Test list of Pydantic models inside a dict."""
+        models = [SamplePydanticModel(name=f"User{i}", age=20 + i) for i in range(5)]
+
+        data = {
+            "users": models,
+            "count": len(models),
+        }
+
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "User0" in json_string
+        assert "User4" in json_string
+
+    def test_credentials_meta_input_scenario(self):
+        """Test the exact scenario from create_graph_execution that was failing."""
+
+        # Simulate CredentialsMetaInput structure
+        class MockCredentialsMetaInput(BaseModel):
+            id: str
+            title: Optional[str] = None
+            provider: str
+            type: str
+
+        cred_input = MockCredentialsMetaInput(
+            id="test-123", title="Test Credentials", provider="github", type="oauth2"
+        )
+
+        # This is how credential_inputs is structured in create_graph_execution
+        credential_inputs = {"github_creds": cred_input}
+
+        # This should work without TypeError
+        result = SafeJson(credential_inputs)
+        assert isinstance(result, Json)
+
+        # Verify it can be JSON serialized
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "test-123" in json_string
+        assert "github" in json_string
+        assert "oauth2" in json_string
+
+    def test_mixed_pydantic_and_primitives(self):
+        """Test complex mix of Pydantic models and primitive types."""
+        model = SamplePydanticModel(name="Test", age=25)
+
+        data = {
+            "models": [model, {"plain": "dict"}, "string", 123],
+            "nested": {
+                "model": model,
+                "list": [1, 2, model, 4],
+                "plain": "text",
+            },
+            "plain_list": [1, 2, 3],
+        }
+
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "Test" in json_string
+        assert "plain" in json_string
+
+    def test_pydantic_model_with_control_chars_in_dict(self):
+        """Test Pydantic model with control chars when nested in dict."""
+        model = SamplePydanticModel(
+            name="Test\x00User",  # Has null byte
+            age=30,
+            metadata={"info": "data\x08with\x0Ccontrols"},
+        )
+
+        data = {"credential": model}
+
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        # Verify control characters are removed
+        import json
+
+        json_string = json.dumps(result.data)
+        assert "\x00" not in json_string
+        assert "\x08" not in json_string
+        assert "\x0C" not in json_string
+        assert "TestUser" in json_string  # Name preserved minus null byte
+
+    def test_deeply_nested_pydantic_models_control_char_sanitization(self):
+        """Test that control characters are sanitized in deeply nested Pydantic models."""
+
+        # Create nested Pydantic models with control characters at different levels
+        class InnerModel(BaseModel):
+            deep_string: str
+            value: int = 42
+            metadata: dict = {}
+
+        class MiddleModel(BaseModel):
+            middle_string: str
+            inner: InnerModel
+            data: str
+
+        class OuterModel(BaseModel):
+            outer_string: str
+            middle: MiddleModel
+
+        # Create test data with control characters at every nesting level
+        inner = InnerModel(
+            deep_string="Deepest\x00Level\x08Control\x0CChars",  # Multiple control chars at deepest level
+            metadata={
+                "nested_key": "Nested\x1FValue\x7FDelete"
+            },  # Control chars in nested dict
+        )
+
+        middle = MiddleModel(
+            middle_string="Middle\x01StartOfHeading\x1FUnitSeparator",
+            inner=inner,
+            data="Some\x0BVerticalTab\x0EShiftOut",
+        )
+
+        outer = OuterModel(outer_string="Outer\x00Null\x07Bell", middle=middle)
+
+        # Wrap in a dict with additional control characters
+        data = {
+            "top_level": "Top\x00Level\x08Backspace",
+            "nested_model": outer,
+            "list_with_strings": [
+                "List\x00Item1",
+                "List\x0CItem2\x1F",
+                {"dict_in_list": "Dict\x08Value"},
+            ],
+        }
+
+        # Process with SafeJson
+        result = SafeJson(data)
+        assert isinstance(result, Json)
+
+        # Verify all control characters are removed at every level
+        import json
+
+        json_string = json.dumps(result.data)
+
+        # Check that NO control characters remain anywhere
+        control_chars = [
+            "\x00",
+            "\x01",
+            "\x02",
+            "\x03",
+            "\x04",
+            "\x05",
+            "\x06",
+            "\x07",
+            "\x08",
+            "\x0B",
+            "\x0C",
+            "\x0E",
+            "\x0F",
+            "\x10",
+            "\x11",
+            "\x12",
+            "\x13",
+            "\x14",
+            "\x15",
+            "\x16",
+            "\x17",
+            "\x18",
+            "\x19",
+            "\x1A",
+            "\x1B",
+            "\x1C",
+            "\x1D",
+            "\x1E",
+            "\x1F",
+            "\x7F",
+        ]
+
+        for char in control_chars:
+            assert (
+                char not in json_string
+            ), f"Control character {repr(char)} found in result"
+
+        # Verify specific sanitized content is present (control chars removed but text preserved)
+        result_data = cast(dict[str, Any], result.data)
+
+        # Top level
+        assert "TopLevelBackspace" in json_string
+
+        # Outer model level
+        assert "OuterNullBell" in json_string
+
+        # Middle model level
+        assert "MiddleStartOfHeadingUnitSeparator" in json_string
+        assert "SomeVerticalTabShiftOut" in json_string
+
+        # Inner model level (deepest nesting)
+        assert "DeepestLevelControlChars" in json_string
+
+        # Nested dict in model
+        assert "NestedValueDelete" in json_string
+
+        # List items
+        assert "ListItem1" in json_string
+        assert "ListItem2" in json_string
+        assert "DictValue" in json_string
+
+        # Verify structure is preserved (not just converted to string)
+        assert isinstance(result_data, dict)
+        assert isinstance(result_data["nested_model"], dict)
+        assert isinstance(result_data["nested_model"]["middle"], dict)
+        assert isinstance(result_data["nested_model"]["middle"]["inner"], dict)
+        assert isinstance(result_data["list_with_strings"], list)
+
+        # Verify specific deep values are accessible and sanitized
+        nested_model = cast(dict[str, Any], result_data["nested_model"])
+        middle = cast(dict[str, Any], nested_model["middle"])
+        inner = cast(dict[str, Any], middle["inner"])
+
+        deep_string = inner["deep_string"]
+        assert deep_string == "DeepestLevelControlChars"
+
+        metadata = cast(dict[str, Any], inner["metadata"])
+        nested_metadata = metadata["nested_key"]
+        assert nested_metadata == "NestedValueDelete"

autogpt_platform/backend/backend/util/text.py

@@ -3,6 +3,7 @@ import logging
 import bleach
 from bleach.css_sanitizer import CSSSanitizer
 from jinja2 import BaseLoader
+from jinja2.exceptions import TemplateError
 from jinja2.sandbox import SandboxedEnvironment
 from markupsafe import Markup
 
@@ -101,8 +102,11 @@ class TextFormatter:
 
     def format_string(self, template_str: str, values=None, **kwargs) -> str:
         """Regular template rendering with escaping"""
-        template = self.env.from_string(template_str)
-        return template.render(values or {}, **kwargs)
+        try:
+            template = self.env.from_string(template_str)
+            return template.render(values or {}, **kwargs)
+        except TemplateError as e:
+            raise ValueError(e) from e
 
     def format_email(
         self,

autogpt_platform/backend/migrations/20251003085016_update_store_agent_view_latest_version_only/migration.sql

@@ -0,0 +1,62 @@
+BEGIN;
+
+-- Drop and recreate the StoreAgent view with isAvailable field
+DROP VIEW IF EXISTS "StoreAgent";
+
+CREATE OR REPLACE VIEW "StoreAgent" AS
+WITH latest_versions AS (
+    SELECT 
+        "storeListingId",
+        MAX(version) AS max_version
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+),
+agent_versions AS (
+    SELECT 
+        "storeListingId",
+        array_agg(DISTINCT version::text ORDER BY version::text) AS versions
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+)
+SELECT
+    sl.id AS listing_id,
+    slv.id AS "storeListingVersionId",
+    slv."createdAt" AS updated_at,
+    sl.slug,
+    COALESCE(slv.name, '') AS agent_name,
+    slv."videoUrl" AS agent_video,
+    COALESCE(slv."imageUrls", ARRAY[]::text[]) AS agent_image,
+    slv."isFeatured" AS featured,
+    p.username AS creator_username,  -- Allow NULL for malformed sub-agents
+    p."avatarUrl" AS creator_avatar,  -- Allow NULL for malformed sub-agents
+    slv."subHeading" AS sub_heading,
+    slv.description,
+    slv.categories,
+    COALESCE(ar.run_count, 0::bigint) AS runs,
+    COALESCE(rs.avg_rating, 0.0)::double precision AS rating,
+    COALESCE(av.versions, ARRAY[slv.version::text]) AS versions,
+    slv."isAvailable" AS is_available  -- Add isAvailable field to filter sub-agents
+FROM "StoreListing" sl
+JOIN latest_versions lv 
+    ON sl.id = lv."storeListingId"
+JOIN "StoreListingVersion" slv 
+    ON slv."storeListingId" = lv."storeListingId"
+   AND slv.version = lv.max_version
+   AND slv."submissionStatus" = 'APPROVED'
+JOIN "AgentGraph" a 
+    ON slv."agentGraphId" = a.id 
+    AND slv."agentGraphVersion" = a.version
+LEFT JOIN "Profile" p 
+    ON sl."owningUserId" = p."userId"
+LEFT JOIN "mv_review_stats" rs 
+    ON sl.id = rs."storeListingId"
+LEFT JOIN "mv_agent_run_counts" ar 
+    ON a.id = ar."agentGraphId"
+LEFT JOIN agent_versions av 
+    ON sl.id = av."storeListingId"
+WHERE sl."isDeleted" = false
+  AND sl."hasApprovedVersion" = true;
+
+COMMIT;

autogpt_platform/backend/migrations/20251007084233_log_search_terms/migration.sql

@@ -0,0 +1,11 @@
+-- CreateTable
+CREATE TABLE "SearchTerms" (
+    "id" BIGSERIAL NOT NULL,
+    "createdDate" TIMESTAMP(3) NOT NULL,
+    "searchTerm" TEXT NOT NULL,
+
+    CONSTRAINT "SearchTerms_pkey" PRIMARY KEY ("id")
+);
+
+-- CreateIndex
+CREATE INDEX "SearchTerms_createdDate_idx" ON "SearchTerms"("createdDate");

autogpt_platform/backend/migrations/20251011120000_add_user_balance_column/migration.sql

@@ -0,0 +1,16 @@
+-- Create UserBalance table for atomic credit operations
+-- This replaces the need for User.balance column and provides better separation of concerns
+-- UserBalance records are automatically created by the application when users interact with the credit system
+
+-- CreateTable (only if it doesn't exist)
+CREATE TABLE IF NOT EXISTS "UserBalance" (
+    "userId" TEXT NOT NULL,
+    "balance" INTEGER NOT NULL DEFAULT 0,
+    "updatedAt" TIMESTAMP(3) NOT NULL,
+
+    CONSTRAINT "UserBalance_pkey" PRIMARY KEY ("userId"),
+    CONSTRAINT "UserBalance_userId_fkey" FOREIGN KEY ("userId") REFERENCES "User"("id") ON DELETE CASCADE ON UPDATE CASCADE
+);
+
+-- CreateIndex (only if it doesn't exist)
+CREATE INDEX IF NOT EXISTS "UserBalance_userId_idx" ON "UserBalance"("userId");

autogpt_platform/backend/migrations/20251013071709_add_use_for_onboarding/migration.sql

@@ -0,0 +1,65 @@
+BEGIN;
+-- AlterTable
+ALTER TABLE "StoreListing" ADD COLUMN     "useForOnboarding" BOOLEAN NOT NULL DEFAULT false;
+
+-- Drop and recreate the StoreAgent view with useForOnboarding field
+DROP VIEW IF EXISTS "StoreAgent";
+
+CREATE OR REPLACE VIEW "StoreAgent" AS
+WITH latest_versions AS (
+    SELECT 
+        "storeListingId",
+        MAX(version) AS max_version
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+),
+agent_versions AS (
+    SELECT 
+        "storeListingId",
+        array_agg(DISTINCT version::text ORDER BY version::text) AS versions
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+)
+SELECT
+    sl.id AS listing_id,
+    slv.id AS "storeListingVersionId",
+    slv."createdAt" AS updated_at,
+    sl.slug,
+    COALESCE(slv.name, '') AS agent_name,
+    slv."videoUrl" AS agent_video,
+    COALESCE(slv."imageUrls", ARRAY[]::text[]) AS agent_image,
+    slv."isFeatured" AS featured,
+    p.username AS creator_username,  -- Allow NULL for malformed sub-agents
+    p."avatarUrl" AS creator_avatar,  -- Allow NULL for malformed sub-agents
+    slv."subHeading" AS sub_heading,
+    slv.description,
+    slv.categories,
+    COALESCE(ar.run_count, 0::bigint) AS runs,
+    COALESCE(rs.avg_rating, 0.0)::double precision AS rating,
+    COALESCE(av.versions, ARRAY[slv.version::text]) AS versions,
+    slv."isAvailable" AS is_available,
+    COALESCE(sl."useForOnboarding", false) AS "useForOnboarding"
+FROM "StoreListing" sl
+JOIN latest_versions lv 
+    ON sl.id = lv."storeListingId"
+JOIN "StoreListingVersion" slv 
+    ON slv."storeListingId" = lv."storeListingId"
+   AND slv.version = lv.max_version
+   AND slv."submissionStatus" = 'APPROVED'
+JOIN "AgentGraph" a 
+    ON slv."agentGraphId" = a.id 
+    AND slv."agentGraphVersion" = a.version
+LEFT JOIN "Profile" p 
+    ON sl."owningUserId" = p."userId"
+LEFT JOIN "mv_review_stats" rs 
+    ON sl.id = rs."storeListingId"
+LEFT JOIN "mv_agent_run_counts" ar 
+    ON a.id = ar."agentGraphId"
+LEFT JOIN agent_versions av 
+    ON sl.id = av."storeListingId"
+WHERE sl."isDeleted" = false
+  AND sl."hasApprovedVersion" = true;
+
+COMMIT;

autogpt_platform/backend/migrations/20251016093049_add_full_text_search/migration.sql

@@ -0,0 +1,100 @@
+-- AlterTable
+ALTER TABLE "StoreListingVersion" ADD COLUMN "search" tsvector DEFAULT ''::tsvector;
+
+-- Add trigger to update the search column with the tsvector of the agent
+-- Function to be invoked by trigger
+
+-- Drop the trigger first
+DROP TRIGGER IF EXISTS "update_tsvector" ON "StoreListingVersion";
+
+-- Drop the function completely
+DROP FUNCTION IF EXISTS update_tsvector_column();
+
+-- Now recreate it fresh
+CREATE OR REPLACE FUNCTION update_tsvector_column() RETURNS TRIGGER AS $$
+BEGIN
+  NEW.search := to_tsvector('english', 
+    COALESCE(NEW.name, '') || ' ' ||
+    COALESCE(NEW.description, '') || ' ' ||
+    COALESCE(NEW."subHeading", '')
+  );
+  RETURN NEW;
+END;
+$$ LANGUAGE plpgsql SECURITY DEFINER SET search_path = platform, pg_temp;
+
+-- Recreate the trigger
+CREATE TRIGGER "update_tsvector"
+BEFORE INSERT OR UPDATE ON "StoreListingVersion"
+FOR EACH ROW
+EXECUTE FUNCTION update_tsvector_column();
+
+UPDATE "StoreListingVersion"
+SET search = to_tsvector('english', 
+    COALESCE(name, '') || ' ' ||
+    COALESCE(description, '') || ' ' ||
+    COALESCE("subHeading", '')
+)
+WHERE search IS NULL;
+
+-- Drop and recreate the StoreAgent view with isAvailable field
+DROP VIEW IF EXISTS "StoreAgent";
+
+CREATE OR REPLACE VIEW "StoreAgent" AS
+WITH latest_versions AS (
+    SELECT 
+        "storeListingId",
+        MAX(version) AS max_version
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+),
+agent_versions AS (
+    SELECT 
+        "storeListingId",
+        array_agg(DISTINCT version::text ORDER BY version::text) AS versions
+    FROM "StoreListingVersion"
+    WHERE "submissionStatus" = 'APPROVED'
+    GROUP BY "storeListingId"
+)
+SELECT
+    sl.id AS listing_id,
+    slv.id AS "storeListingVersionId",
+    slv."createdAt" AS updated_at,
+    sl.slug,
+    COALESCE(slv.name, '') AS agent_name,
+    slv."videoUrl" AS agent_video,
+    COALESCE(slv."imageUrls", ARRAY[]::text[]) AS agent_image,
+    slv."isFeatured" AS featured,
+    p.username AS creator_username,  -- Allow NULL for malformed sub-agents
+    p."avatarUrl" AS creator_avatar,  -- Allow NULL for malformed sub-agents
+    slv."subHeading" AS sub_heading,
+    slv.description,
+    slv.categories,
+    slv.search,
+    COALESCE(ar.run_count, 0::bigint) AS runs,
+    COALESCE(rs.avg_rating, 0.0)::double precision AS rating,
+    COALESCE(av.versions, ARRAY[slv.version::text]) AS versions,
+    COALESCE(sl."useForOnboarding", false) AS "useForOnboarding",
+    slv."isAvailable" AS is_available  -- Add isAvailable field to filter sub-agents
+FROM "StoreListing" sl
+JOIN latest_versions lv 
+    ON sl.id = lv."storeListingId"
+JOIN "StoreListingVersion" slv 
+    ON slv."storeListingId" = lv."storeListingId"
+   AND slv.version = lv.max_version
+   AND slv."submissionStatus" = 'APPROVED'
+JOIN "AgentGraph" a 
+    ON slv."agentGraphId" = a.id 
+    AND slv."agentGraphVersion" = a.version
+LEFT JOIN "Profile" p 
+    ON sl."owningUserId" = p."userId"
+LEFT JOIN "mv_review_stats" rs 
+    ON sl.id = rs."storeListingId"
+LEFT JOIN "mv_agent_run_counts" ar 
+    ON a.id = ar."agentGraphId"
+LEFT JOIN agent_versions av 
+    ON sl.id = av."storeListingId"
+WHERE sl."isDeleted" = false
+  AND sl."hasApprovedVersion" = true;
+
+COMMIT;

autogpt_platform/backend/migrations/20251027162201_migrate_claude_3_5_to_4_5_models/migration.sql

@@ -0,0 +1,21 @@
+-- Migrate Claude 3.5 models to Claude 4.5 models
+-- This updates all AgentNode blocks that use deprecated Claude 3.5 models to the new 4.5 models
+-- See: https://docs.anthropic.com/en/docs/about-claude/models/legacy-model-guide
+
+-- Update Claude 3.5 Sonnet to Claude 4.5 Sonnet
+UPDATE "AgentNode"
+SET    "constantInput" = JSONB_SET(
+         "constantInput"::jsonb,
+         '{model}',
+         '"claude-sonnet-4-5-20250929"'::jsonb
+       )
+WHERE  "constantInput"::jsonb->>'model' = 'claude-3-5-sonnet-latest';
+
+-- Update Claude 3.5 Haiku to Claude 4.5 Haiku
+UPDATE "AgentNode"
+SET    "constantInput" = JSONB_SET(
+         "constantInput"::jsonb,
+         '{model}',
+         '"claude-haiku-4-5-20251001"'::jsonb
+       )
+WHERE  "constantInput"::jsonb->>'model' = 'claude-3-5-haiku-latest';

autogpt_platform/backend/schema.prisma

@@ -5,10 +5,11 @@ datasource db {
 }
 
 generator client {
-  provider             = "prisma-client-py"
-  recursive_type_depth = -1
-  interface            = "asyncio"
-  previewFeatures      = ["views"]
+  provider               = "prisma-client-py"
+  recursive_type_depth   = -1
+  interface              = "asyncio"
+  previewFeatures        = ["views", "fullTextSearch"]
+  partial_type_generator = "backend/data/partial_types.py"
 }
 
 // User model to mirror Auth provider users
@@ -45,6 +46,7 @@ model User {
   AnalyticsDetails     AnalyticsDetails[]
   AnalyticsMetrics     AnalyticsMetrics[]
   CreditTransactions   CreditTransaction[]
+  UserBalance          UserBalance?
 
   AgentPresets  AgentPreset[]
   LibraryAgents LibraryAgent[]
@@ -118,9 +120,9 @@ model AgentGraph {
   createdAt DateTime  @default(now())
   updatedAt DateTime? @updatedAt
 
-  name        String?
-  description String?
-  instructions String?
+  name                    String?
+  description             String?
+  instructions            String?
   recommendedScheduleCron String?
 
   isActive Boolean @default(true)
@@ -382,9 +384,9 @@ model AgentGraphExecution {
   stats Json?
 
   // Sharing fields
-  isShared    Boolean   @default(false)
-  shareToken  String?   @unique
-  sharedAt    DateTime?
+  isShared   Boolean   @default(false)
+  shareToken String?   @unique
+  sharedAt   DateTime?
 
   @@index([agentGraphId, agentGraphVersion])
   @@index([userId, isDeleted, createdAt])
@@ -545,7 +547,7 @@ model CreditTransaction {
   createdAt      DateTime @default(now())
 
   userId String
-  User   User?   @relation(fields: [userId], references: [id], onDelete: NoAction)
+  User   User?  @relation(fields: [userId], references: [id], onDelete: NoAction)
 
   amount Int
   type   CreditTransactionType
@@ -587,6 +589,16 @@ model CreditRefundRequest {
 ////////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////
 
+model SearchTerms {
+  // User ID not being logged as this is anonymous analytics data
+  // Not using uuid as we want to minimise table size
+  id          BigInt   @id @default(autoincrement())
+  createdDate DateTime
+  searchTerm  String
+
+  @@index([createdDate])
+}
+
 model Profile {
   id        String   @id @default(uuid())
   createdAt DateTime @default(now())
@@ -653,10 +665,12 @@ view StoreAgent {
   sub_heading      String
   description      String
   categories       String[]
+  search           Unsupported("tsvector")? @default(dbgenerated("''::tsvector"))
   runs             Int
   rating           Float
   versions         String[]
   is_available     Boolean  @default(true)
+  useForOnboarding Boolean  @default(false)
 
   // Materialized views used (refreshed every 15 minutes via pg_cron):
   // - mv_agent_run_counts - Pre-aggregated agent execution counts by agentGraphId
@@ -734,6 +748,9 @@ model StoreListing {
   // URL-friendly identifier for this agent (moved from StoreListingVersion)
   slug String
 
+  // Allow this agent to be used during onboarding
+  useForOnboarding Boolean @default(false)
+
   // The currently active version that should be shown to users
   activeVersionId String?              @unique
   ActiveVersion   StoreListingVersion? @relation("ActiveVersion", fields: [activeVersionId], references: [id])
@@ -769,13 +786,13 @@ model StoreListingVersion {
   AgentGraph        AgentGraph @relation(fields: [agentGraphId, agentGraphVersion], references: [id, version])
 
   // Content fields
-  name        String
-  subHeading  String
-  videoUrl    String?
-  imageUrls   String[]
-  description String
+  name         String
+  subHeading   String
+  videoUrl     String?
+  imageUrls    String[]
+  description  String
   instructions String?
-  categories  String[]
+  categories   String[]
 
   isFeatured Boolean @default(false)
 
@@ -783,6 +800,8 @@ model StoreListingVersion {
   // Old versions can be made unavailable by the author if desired
   isAvailable Boolean @default(true)
 
+  search Unsupported("tsvector")? @default(dbgenerated("''::tsvector"))
+
   // Version workflow state
   submissionStatus SubmissionStatus @default(DRAFT)
   submittedAt      DateTime?
@@ -873,6 +892,16 @@ model APIKey {
   @@index([userId, status])
 }
 
+model UserBalance {
+  userId    String   @id
+  balance   Int      @default(0)
+  updatedAt DateTime @updatedAt
+
+  user User @relation(fields: [userId], references: [id], onDelete: Cascade)
+
+  @@index([userId])
+}
+
 enum APIKeyStatus {
   ACTIVE
   REVOKED

autogpt_platform/backend/snapshots/auth_user

@@ -1,5 +1,5 @@
 {
   "email": "test@example.com",
-  "id": "test-user-id",
+  "id": "3e53486c-cf57-477e-ba2a-cb02dc828e1a",
   "name": "Test User"
 }

autogpt_platform/backend/snapshots/grph_single

@@ -28,6 +28,6 @@
   "recommended_schedule_cron": null,
   "sub_graphs": [],
   "trigger_setup_info": null,
-  "user_id": "test-user-id",
+  "user_id": "3e53486c-cf57-477e-ba2a-cb02dc828e1a",
   "version": 1
 }

autogpt_platform/backend/snapshots/grphs_all

@@ -26,7 +26,7 @@
     "recommended_schedule_cron": null,
     "sub_graphs": [],
     "trigger_setup_info": null,
-    "user_id": "test-user-id",
+    "user_id": "3e53486c-cf57-477e-ba2a-cb02dc828e1a",
     "version": 1
   }
 ]

autogpt_platform/backend/test/blocks/test_youtube.py

@@ -0,0 +1,140 @@
+from unittest.mock import Mock, patch
+
+import pytest
+from youtube_transcript_api._errors import NoTranscriptFound
+from youtube_transcript_api._transcripts import FetchedTranscript, Transcript
+
+from backend.blocks.youtube import TranscribeYoutubeVideoBlock
+
+
+class TestTranscribeYoutubeVideoBlock:
+    """Test cases for TranscribeYoutubeVideoBlock language fallback functionality."""
+
+    def setup_method(self):
+        """Set up test fixtures."""
+        self.youtube_block = TranscribeYoutubeVideoBlock()
+
+    def test_extract_video_id_standard_url(self):
+        """Test extracting video ID from standard YouTube URL."""
+        url = "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
+        video_id = self.youtube_block.extract_video_id(url)
+        assert video_id == "dQw4w9WgXcQ"
+
+    def test_extract_video_id_short_url(self):
+        """Test extracting video ID from shortened youtu.be URL."""
+        url = "https://youtu.be/dQw4w9WgXcQ"
+        video_id = self.youtube_block.extract_video_id(url)
+        assert video_id == "dQw4w9WgXcQ"
+
+    def test_extract_video_id_embed_url(self):
+        """Test extracting video ID from embed URL."""
+        url = "https://www.youtube.com/embed/dQw4w9WgXcQ"
+        video_id = self.youtube_block.extract_video_id(url)
+        assert video_id == "dQw4w9WgXcQ"
+
+    @patch("backend.blocks.youtube.YouTubeTranscriptApi")
+    def test_get_transcript_english_available(self, mock_api_class):
+        """Test getting transcript when English is available."""
+        # Setup mock
+        mock_api = Mock()
+        mock_api_class.return_value = mock_api
+        mock_transcript = Mock(spec=FetchedTranscript)
+        mock_api.fetch.return_value = mock_transcript
+
+        # Execute
+        result = TranscribeYoutubeVideoBlock.get_transcript("test_video_id")
+
+        # Assert
+        assert result == mock_transcript
+        mock_api.fetch.assert_called_once_with(video_id="test_video_id")
+        mock_api.list.assert_not_called()
+
+    @patch("backend.blocks.youtube.YouTubeTranscriptApi")
+    def test_get_transcript_fallback_to_first_available(self, mock_api_class):
+        """Test fallback to first available language when English is not available."""
+        # Setup mock
+        mock_api = Mock()
+        mock_api_class.return_value = mock_api
+
+        # Create mock transcript list with Hungarian transcript
+        mock_transcript_list = Mock()
+        mock_transcript_hu = Mock(spec=Transcript)
+        mock_fetched_transcript = Mock(spec=FetchedTranscript)
+        mock_transcript_hu.fetch.return_value = mock_fetched_transcript
+
+        # Set up the transcript list to have manually created transcripts empty
+        # and generated transcripts with Hungarian
+        mock_transcript_list._manually_created_transcripts = {}
+        mock_transcript_list._generated_transcripts = {"hu": mock_transcript_hu}
+
+        # Mock API to raise NoTranscriptFound for English, then return list
+        mock_api.fetch.side_effect = NoTranscriptFound(
+            "test_video_id", ("en",), mock_transcript_list
+        )
+        mock_api.list.return_value = mock_transcript_list
+
+        # Execute
+        result = TranscribeYoutubeVideoBlock.get_transcript("test_video_id")
+
+        # Assert
+        assert result == mock_fetched_transcript
+        mock_api.fetch.assert_called_once_with(video_id="test_video_id")
+        mock_api.list.assert_called_once_with("test_video_id")
+        mock_transcript_hu.fetch.assert_called_once()
+
+    @patch("backend.blocks.youtube.YouTubeTranscriptApi")
+    def test_get_transcript_prefers_manually_created(self, mock_api_class):
+        """Test that manually created transcripts are preferred over generated ones."""
+        # Setup mock
+        mock_api = Mock()
+        mock_api_class.return_value = mock_api
+
+        # Create mock transcript list with both manual and generated transcripts
+        mock_transcript_list = Mock()
+        mock_transcript_manual = Mock(spec=Transcript)
+        mock_transcript_generated = Mock(spec=Transcript)
+        mock_fetched_manual = Mock(spec=FetchedTranscript)
+        mock_transcript_manual.fetch.return_value = mock_fetched_manual
+
+        # Set up the transcript list
+        mock_transcript_list._manually_created_transcripts = {
+            "es": mock_transcript_manual
+        }
+        mock_transcript_list._generated_transcripts = {"hu": mock_transcript_generated}
+
+        # Mock API to raise NoTranscriptFound for English
+        mock_api.fetch.side_effect = NoTranscriptFound(
+            "test_video_id", ("en",), mock_transcript_list
+        )
+        mock_api.list.return_value = mock_transcript_list
+
+        # Execute
+        result = TranscribeYoutubeVideoBlock.get_transcript("test_video_id")
+
+        # Assert - should use manually created transcript first
+        assert result == mock_fetched_manual
+        mock_transcript_manual.fetch.assert_called_once()
+        mock_transcript_generated.fetch.assert_not_called()
+
+    @patch("backend.blocks.youtube.YouTubeTranscriptApi")
+    def test_get_transcript_no_transcripts_available(self, mock_api_class):
+        """Test that exception is re-raised when no transcripts are available at all."""
+        # Setup mock
+        mock_api = Mock()
+        mock_api_class.return_value = mock_api
+
+        # Create mock transcript list with no transcripts
+        mock_transcript_list = Mock()
+        mock_transcript_list._manually_created_transcripts = {}
+        mock_transcript_list._generated_transcripts = {}
+
+        # Mock API to raise NoTranscriptFound
+        original_exception = NoTranscriptFound(
+            "test_video_id", ("en",), mock_transcript_list
+        )
+        mock_api.fetch.side_effect = original_exception
+        mock_api.list.return_value = mock_transcript_list
+
+        # Execute and assert exception is raised
+        with pytest.raises(NoTranscriptFound):
+            TranscribeYoutubeVideoBlock.get_transcript("test_video_id")

autogpt_platform/backend/test/e2e_test_data.py

@@ -749,10 +749,11 @@ class TestDataCreator:
         """Add credits to users."""
         print("Adding credits to users...")
 
-        credit_model = get_user_credit_model()
-
         for user in self.users:
             try:
+                # Get user-specific credit model
+                credit_model = await get_user_credit_model(user["id"])
+
                 # Skip credits for disabled credit model to avoid errors
                 if (
                     hasattr(credit_model, "__class__")