## Changes 🏗️
- Refactor homepage redirect logic to always point to `/`
- the `/` route handles whether to redirect to `/copilot` or `/library`
based on flag
- Simplify `useGetFlag` checks
- Add `<FeatureFlagRedirect />` and `<FeatureFlagPage />` wrapper
components
- helpers to do 1 thing or the other, depending on chat enabled/disabled
- avoids boilerplate code, checking flagss and redirects mistakes
(especially around race conditions with LD init )
## 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] Log in / out of AutoGPT with flag disabled/enabled
- [x] Sign up to AutoGPT with flag disabled/enabled
- [x] Redirects to homepage always work `/`
- [x] Can't access Copilot with disabled flag
[OPEN-2743: Ability to Update Sub-Agents in Graph (Without
Re-Adding)](https://linear.app/autogpt/issue/OPEN-2743/ability-to-update-sub-agents-in-graph-without-re-adding)
Updating sub-graphs is a cumbersome experience at the moment, this
should help. :)
Demo in Builder v2:
https://github.com/user-attachments/assets/df564f32-4d1d-432c-bb91-fe9065068360https://github.com/user-attachments/assets/f169471a-1f22-46e9-a958-ddb72d3f65af
### Changes 🏗️
- Add sub-graph update banner with I/O incompatibility notification and
resolution mode
- Red visual indicators for broken inputs/outputs and edges
- Update bars and tooltips show compatibility details
- Sub-agent update UI with compatibility checks, incompatibility dialog,
and guided resolution workflow
- Resolution mode banner guiding users to remove incompatible
connections
- Visual controls to stage/apply updates and auto-apply when broken
connections are fixed
Technical:
- Builder v1: Add `CustomNode` > `IncompatibilityDialog` +
`SubAgentUpdateBar` sub-components
- Builder v2: Add `SubAgentUpdateFeature` + `ResolutionModeBar` +
`IncompatibleUpdateDialog` + `useSubAgentUpdateState` sub-components
- Add `useSubAgentUpdate` hook
- Related fixes in Builder v1:
- Fix static edges not rendering as such
- Fix edge styling not applying
- Related fixes in Builder v2:
- Fix excess spacing for nested node input fields
Other:
- "Retry" button in error view now reloads the page instead of
navigating to `/marketplace`
### 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:
- CI for existing frontend UX flows
- [x] Updating to a new sub-agent version with compatibility issues: UX
flow works
- [x] Updating to a new sub-agent version with *no* compatibility
issues: works
- [x] Designer approves of the look
---------
Co-authored-by: abhi1992002 <abhimanyu1992002@gmail.com>
Co-authored-by: Abhimanyu Yadav <122007096+Abhi1992002@users.noreply.github.com>
Make onboarding task completion backend-authoritative which prevents
cheating (previously users could mark all tasks as completed instantly
and get rewards) and makes task completion more reliable. Completion of
tasks is moved backend with exception of introductory onboarding tasks
and visit-page type tasks.
### Changes 🏗️
- Move incrementing run counter backend and make webhook-triggered and
scheduled task execution count as well
- Use user timezone for calculating run streak
- Frontend task completion is moved from update onboarding state to
separate endpoint and guarded so only frontend tasks can be completed
- Graph creation, execution and add marketplace agent to library accept
`source`, so appropriate tasks can be completed
- Replace `client.ts` api calls with orval generated and remove no
longer used functions from `client.ts`
- Add `resolveResponse` helper function that unwraps orval generated
call result to 2xx response
Small changes&bug fixes:
- Make Redis notification bus serialize all payload fields
- Fix confetti when group is finished
- Collapse finished group when opening Wallet
- Play confetti only for tasks that are listed in the Wallet UI
### 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] Onboarding can be finished
- [x] All tasks can be finished and work properly
- [x] Confetti works properly
## Summary
This PR implements a comprehensive Human In The Loop (HITL) block that
allows agents to pause execution and wait for human
approval/modification of data before continuing.
https://github.com/user-attachments/assets/c027d731-17d3-494c-85ca-97c3bf33329c
## Key Features
- Added WAITING_FOR_REVIEW status to AgentExecutionStatus enum
- Created PendingHumanReview database table for storing review requests
- Implemented HumanInTheLoopBlock that extracts input data and creates
review entries
- Added API endpoints at /api/executions/review for fetching and
reviewing pending data
- Updated execution manager to properly handle waiting status and resume
after approval
## Frontend Components
- PendingReviewCard for individual review handling
- PendingReviewsList for multiple reviews
- FloatingReviewsPanel for graph builder integration
- Integrated review UI into 3 locations: legacy library, new library,
and graph builder
## Technical Implementation
- Added proper type safety throughout with SafeJson handling
- Optimized database queries using count functions instead of full data
fetching
- Fixed imports to be top-level instead of local
- All formatters and linters pass
## Test plan
- [ ] Test Human In The Loop block creation in graph builder
- [ ] Test block execution pauses and creates pending review
- [ ] Test review UI appears in all 3 locations
- [ ] Test data modification and approval workflow
- [ ] Test rejection workflow
- [ ] Test execution resumes after approval
🤖 Generated with [Claude Code](https://claude.ai/code)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Added Human-In-The-Loop review workflows to pause executions for human
validation.
* Users can approve or reject pending tasks, optionally editing
submitted data and adding a message.
* New "Waiting for Review" execution status with UI indicators across
run lists, badges, and activity views.
* Review management UI: pending review cards, list view, and a floating
reviews panel for quick access.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude <noreply@anthropic.com>
- Resolves#11345
### Changes 🏗️
- Move tool use routing logic from frontend to backend: routing info was
being baked into graph links by the frontend, inconsistently, causing
issues
- Rework tool use routing to use target node ID instead of target block
name
- Add a bit of magic to `NodeOutputs` component to show tool node title
instead of ID
DX:
- Removed `build` from `.prettierignore` -> re-enable formatting for
builder components
### 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] Use SDM block in a graph; verify it works
- [x] Use SDM block with agent executor block as tool; verify it works
- Tests for `parse_execution_output` pass (checked by CI)
## Summary
This PR fixes an issue where multiple keyboard save event listeners were
being registered when the same save hook was used in multiple
components, causing the graph to be saved multiple times (3x) when using
Ctrl/Cmd+S.
## Changes
- **Created a centralized `useSaveGraph` hook** in
`/hooks/useSaveGraph.ts` that encapsulates all graph saving logic
- **Refactored `useNewSaveControl`** to use the new centralized hook
instead of duplicating save logic
- **Updated `useRunGraph` and `useScheduleGraph`** to use the
centralized `useSaveGraph` hook directly
- **Simplified the save control component** by removing redundant logic
and using cleaner naming conventions
## Problem
The previous implementation had the save logic duplicated in
`useNewSaveControl`, and when this hook was used in multiple places
(NewSaveControl component, RunGraph, ScheduleGraph), each instance would
register its own keyboard event listener for Ctrl/Cmd+S. This caused:
- Multiple save requests being sent simultaneously
- "Unique constraint failed on the fields: ('id', 'version')" errors
from the backend
- Poor performance due to unnecessary re-renders
## Solution
By centralizing the save logic in a dedicated `useSaveGraph` hook:
- Save logic is now in one place, making it easier to maintain
- Components can use the save functionality without registering
duplicate event listeners
- The keyboard shortcut listener is only registered once in the
`useNewSaveControl` hook
- Other components (RunGraph, ScheduleGraph) can call `saveGraph`
directly without side effects
## Testing
- [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] Verified Ctrl/Cmd+S saves the graph only once
- [x] Tested save functionality from Save Control popup
- [x] Confirmed Run Graph and Schedule Graph still save before execution
- [x] Verified no duplicate save requests in network tab
- [x] Checked that save toast notifications appear correctly
Beads are reset when saving but not on run which can result in beads
from previous runs accumulating on the opened graph.
### Changes 🏗️
- Move bead reset code to function and call it before run
### 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] Beads reset on every run
- Resolves#10980
- 2nd attempt after #11075 broke some things
Fixes unnecessary graph re-saving when no changes were made after
initial save. More specifically, this PR fixes two causes of this issue:
- Frontend node IDs were being compared to backend IDs, which won't
match if the graph has been modified and saved since loading.
- `fillDefaults` was being applied to all nodes (including existing
ones) on element creation, and empty values were being stripped
*post-save* with `removeEmptyStringsAndNulls`. This invisible
auto-modification of node input data meant that in some common cases the
graph would never be in sync with the backend.
### Changes 🏗️
- Fix node ID handling
- Use `node.data.backend_id ?? node.id` instead of `node.id` in
`prepareSaveableGraph`
- Also map link source/sink IDs to their corresponding backend IDs
- Add note about `node.data.backend_id` to `_saveAgent`
- Use `node.data.backend_id || node.id` as display ID in `CustomNode`
- Prevent auto-modification of node input data on existing nodes
- Prune empty values (`undefined`, `null`, `""`) from node input data
*pre-save* instead of post-save
- Related: improve typing and functionality of
`fillObjectDefaultsFromSchema` (moved and renamed from `fillDefaults`)
### 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] Node display ID updates on save
- [x] Clicking save a second time (without making more changes) doesn't
cause re-save
- [x] Updating nodes with dynamic input links (e.g. Create Dictionary
Block) doesn't make the links disappear
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Prevented unintended auto-modification of existing nodes during
editing
* Improved consistency of node and connection identifiers in saved
graphs
* **Improvements**
* Enhanced node title display logic for clearer node identification
* Optimized data cleanup utilities for more robust input processing in
the builder
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
## Changes 🏗️
We weren't fetching all library agents, just the first 15... to compute
the agent map on the Agent Activity dropdown. We suspect that is causing
some agent executions coming as `Unknown agent`.
In this changes, I'm fetching all the library agents upfront ( _without
blocking page load_ ) and caching them on the browser, so we have all
the details to render the agent runs. This is re-used in the library as
well for fast initial load on the agents list page.
## 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] First request populates cache; subsequent identical requests hit
cache
- [x] Editing an agent invalidates relevant cache keys and serves fresh
data
- [x] Different query params generate distinct cache entries
- [x] Cache layer gracefully falls back to live data on errors
- [x] 404 behavior for unknown agents unchanged
### For configuration changes:
None
- Resolves#10980
Fixes unnecessary graph re-saving when no changes were made after
initial save. The issue occurred because frontend node IDs weren't
synced with backend IDs after save operations.
### Changes 🏗️
- Update actual node.id to match backend node ID after save
- Update edge references with new node IDs
- Properly sync visual editor state with backend
### Test Plan 📋
- [x] TypeScript compilation passes
- [x] Pre-commit hooks pass
- [x] Manual test: Save graph, verify no re-save needed on subsequent
save/run
Wallet update removed `BUILDER_OPEN` and `BUILDER_RUN_AGENT`.
### Changes 🏗️
- Restore completion codepaths for `BUILDER_OPEN` and
`BUILDER_RUN_AGENT` for analytical purposes
### 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] Tasks are completed silently
### Changes 🏗️
- Rename wallet and update design
- Update tasks and add Hidden Tasks section
- Update onboarding backend code and related db migration
- Add progress bar for some tasks
### 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] All tasks can be finished
- [x] Finished tasks add correct amount of credits
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Nicholas Tindle <nicholas.tindle@agpt.co>
Co-authored-by: Reinier van der Leer <pwuts@agpt.co>
- Resolves#10926
- Fixes a bug introduced in #10779
### Changes 🏗️
- Fix `.metadata.position` in graph save payload
- Make node reconciliation after graph save more robust
### 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] Moved nodes don't disappear on graph save
## Changes 🏗️
Following up my initial PR to tidy up the `components` folder
https://github.com/Significant-Gravitas/AutoGPT/pull/10940.
This is mostly moving files around and renaming some + documenting them
on the design system as needed. Should be pretty safe as long as types
on the CI pass.
## 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] Run the app locally
- [x] Click around, looks ok
- [x] Test and types pass on the CI
### For configuration changes:
None
We want users to set up triggers through the Library rather than the
Builder.
- Resolves#10413https://github.com/user-attachments/assets/515ed80d-6569-4e26-862f-2a663115218c
### Changes 🏗️
- Update node UI to push users to Library for trigger set-up and
management
- Add note redirecting to Library for trigger set-up
- Remove webhook status indicator and webhook URL section
- Add `libraryAgent: LibraryAgent` to `BuilderContext` for access inside
`CustomNode`
- Move library agent loader from `FlowEditor` to `useAgentGraph`
- Implement `migrate_legacy_triggered_graphs` migrator function
- Remove `on_node_activate` hook (which previously handled webhook
setup)
- Propagate `created_at` from DB to `GraphModel` and
`LibraryAgentPreset` models
### 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] Existing node triggers are converted to triggered presets (visible
in the Library)
- [x] Converted triggered presets work
- [x] Trigger node inputs are disabled and handles are hidden
- [x] Trigger node message links to the correct Library Agent when saved
- Resolves#10926
### Changes 🏗️
- Fix save no-op if graph has no changes
### 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] Saving a graph after only moving nodes doesn't make those nodes
disappear
## Summary
<img width="1000" alt="Screenshot 2025-09-02 at 9 46 49 PM"
src="https://github.com/user-attachments/assets/d78100c7-7974-4d37-a788-757764d8b6b7"
/>
<img width="1000" alt="Screenshot 2025-09-02 at 9 20 24 PM"
src="https://github.com/user-attachments/assets/cd092963-8e26-4198-b65a-4416b2307a50"
/>
<img width="1000" alt="Screenshot 2025-09-02 at 9 22 30 PM"
src="https://github.com/user-attachments/assets/e16b3bdb-c48c-4dec-9281-b2a35b3e21d0"
/>
<img width="1000" alt="Screenshot 2025-09-02 at 9 20 38 PM"
src="https://github.com/user-attachments/assets/11d74a39-f4b4-4fce-8d30-0e6a925f3a9b"
/>
• Added recommended schedule cron expression as an optional input
throughout the platform
• Implemented complete data flow from builder → store submission → agent
library → run page
• Fixed UI layout issues including button text overflow and ensured
proper component reusability
## Changes
### Backend
- Added `recommended_schedule_cron` field to `AgentGraph` schema and
database migration
- Updated API models (`LibraryAgent`, `MyAgent`,
`StoreSubmissionRequest`) to include the new field
- Enhanced store submission approval flow to persist recommended
schedule to database
### Frontend
- Added recommended schedule input to builder page (SaveControl
component) with overflow-safe styling
- Updated store submission modal (PublishAgentModal) with schedule
configuration
- Enhanced agent run page with schedule tip display and pre-filled
schedule dialog
- Refactored `CronSchedulerDialog` with discriminated union types for
better reusability
- Fixed layout issues including button text truncation and popover width
constraints
- Implemented robust cron expression parsing with 100% reversibility
between UI and cron format
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
- Resolves#9307
### Changes 🏗️
- feat(library): Create presets from runs
- Prevent creating preset from run with unknown credentials
- Fix running presets with credentials
- Add `credential_inputs` parameter to `execute_preset` endpoint
API:
- Return `GraphExecutionMeta` from `*/execute` endpoints
### 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:
- Go to `/library/agents/[id]` for an agent that *does not* require
credentials
- Click the menu on any run and select "Pin as a preset"; fill out the
dialog and submit
- [x] -> UI works
- [x] -> Operation succeeds and dialog closes
- [x] -> New preset is shown at the top of the runs list
- Go to `/library/agents/[id]` for an agent that *does* require
credentials
- Click the menu on any run and select "Pin as a preset"; fill out the
dialog and submit
- [x] -> UI works
- [x] -> Error toast appears with descriptive message
- Initiate a new run; once finished, click "Create preset from run";
fill out the dialog and submit
- [x] -> UI works
- [x] -> Operation succeeds and dialog closes
- [x] -> New preset is shown at the top of the runs list
## Summary
- Adds ability to edit custom node titles by clicking a pencil icon that
appears on hover
- Custom titles are saved in node metadata and persist across saves
- Original node type is shown in tooltip when hovering over custom
titles
https://github.com/user-attachments/assets/a0a41ac9-1ffb-44c8-9e1c-f4c42e032b49
## Changes
- **CustomNode.tsx**:
- Added inline title editing with pencil icon on hover
- Implemented state management for title editing mode
- Added tooltip to show original node type for custom titles
- Prevents custom names from being copied when duplicating nodes
- **useAgentGraph.tsx**:
- Updated graph save/load logic to preserve metadata including custom
titles
- Ensures metadata persistence through all node operations
## Technical Details
- Uses existing `metadata` JSON field in AgentNode model (no database
changes needed)
- Stores custom title in `metadata.customized_name`
- Backward compatible - nodes without custom titles display normally
## Test Plan
- [x] Hover over node title shows pencil icon
- [x] Click pencil icon to edit title
- [x] Press Enter or blur to save, Escape to cancel
- [x] Custom title persists after saving graph
- [x] Tooltip shows original node type when hovering over custom title
- [x] Copying node doesn't copy custom name
- [x] Backward compatible with existing graphs
Backend for the Blocks Menu Redesign.
### Changes 🏗️
- Add optional `agent_name` to the `AgentExecutorBlock` - displayed as
the block name in the Builder
- Include `output_schema` in the `LibraryAgent` model
- Make `v2.store.db.py:get_store_agents` accept multiple creators filter
- Add `api/builder` router with endpoints (and accompanying logic in
`v2/builder/db` and models in `v2/builder/models`)
- `/suggestions`: elements for the suggestions tab
- `/categories`: categories with a number of blocks per each
- `/blocks`: blocks based on category, type or provider
- `/providers`: integration providers with their block counts
- `/serach`: search blocks (including integrations), marketplace agents
and user library agents
- `/counts`: element counts for each category in the Blocks Menu.
### 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] Modified function `get_store_agents` works in existing code paths
- [x] Agent executor block works
- [x] New endpoints work
- [x] Existing Builder menu is unaffected
---------
Co-authored-by: Abhimanyu Yadav <abhimanyu1992002@gmail.com>
Co-authored-by: Abhimanyu Yadav <122007096+Abhi1992002@users.noreply.github.com>
<!-- Clearly explain the need for these changes: -->
### Need for these changes 💥
This PR resolves Linear issue `SECRT-1290`, addressing a critical bug
where the scheduler API fails with a "Wrong number of fields" error when
empty or invalid cron expressions are submitted from the frontend. This
was causing production errors and a poor user experience. It was an off
by one error
### Changes 🏗️
Fix off by one error + add additional logging / error messaging when
someone makes an invalid cron
https://github.com/user-attachments/assets/775881a9-707b-4c4f-b23a-bd7118a358ee
### 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] Attempt to schedule an agent with an empty cron expression from
the UI and confirm a frontend toast error.
- [x] Attempt to schedule an agent with an incomplete yearly cron (no
months selected) from the UI and confirm a frontend toast error and UI
warning.
- [x] Attempt to schedule an agent with an incomplete monthly cron (no
days selected) from the UI and confirm a frontend toast error and UI
warning.
- [x] Attempt to schedule an agent with an incomplete weekly cron (no
days selected) from the UI and confirm a frontend toast error and UI
warning.
- [x] Verify that valid cron expressions can still be scheduled
successfully.
- [x] Run backend unit tests for scheduler cron validation.
- [x] Run frontend unit tests for cron expression utility.
---
Linear Issue: [SECRT-1290](https://linear.app/autogpt/issue/SECRT-1290)
<a
href="https://cursor.com/background-agent?bcId=bc-8bc10502-9498-4dbd-afa2-93e15990fa8c">
<picture>
<source media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/open-in-cursor-dark.svg">
<source media="(prefers-color-scheme: light)"
srcset="https://cursor.com/open-in-cursor-light.svg">
<img alt="Open in Cursor" src="https://cursor.com/open-in-cursor.svg">
</picture>
</a>
<a
href="https://cursor.com/agents?id=bc-8bc10502-9498-4dbd-afa2-93e15990fa8c">
<picture>
<source media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/open-in-web-dark.svg">
<source media="(prefers-color-scheme: light)"
srcset="https://cursor.com/open-in-web-light.svg">
<img alt="Open in Web" src="https://cursor.com/open-in-web.svg">
</picture>
</a>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Swifty <craigswift13@gmail.com>
- Resolves#10444
Sometimes, the order of nodes and/or links isn't consistent between
frontend and backend, which currently can result in unnecessary
re-saving of the graph when the user tries to run it.
Also, `sub_graphs` was not included in the frontend `Graph` type, which
can cause unchecked code issues when the object is propragated using
spread operators.
### Changes 🏗️
- fix(frontend/builder): Make `graphsEquivalent` insensitive to link and
node order
- dx(frontend): Fix typing of `Graph.sub_graphs` (and its variants)
### 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:
- Import an agent and open it in the builder
- Run it without making any changes to the graph itself
- [x] -> graph shouldn't re-save
- Resolves#10553
### Changes 🏗️
- Remove frontend graph validation in `useAgentGraph:saveAndRun(..)`
- Remove now unused `ajv` dependency
- Implement graph validation error propagation (backend->frontend)
- Add `GraphValidationError` type in frontend and backend
- Add `GraphModel.validate_graph_get_errors(..)` method
- Fix error handling & propagation in frontend API request logic
### 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] Saving & running a graph with missing required inputs gives a
node-specific error
- [x] Saving & running a graph with missing node credential inputs
succeeds with passed-in credentials
Make agent graph execution durable by making it retriable. When it fails
to retry, we should make the error visible to the UI.
<img width="900" height="495" alt="image"
src="https://github.com/user-attachments/assets/70e3e117-31e7-4704-8bdf-1802c6afc70b"
/>
<img width="900" height="407" alt="image"
src="https://github.com/user-attachments/assets/78ca6c28-6cc2-4aff-bfa9-9f94b7f89f77"
/>
### Changes 🏗️
* Make _on_graph_execution retriable
* Increase retry count for failing db-manager RPC
* Add test coverage for RPC failure retry
### 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:
<!-- Put your test plan here: -->
- [x] Allow graph execution retry
- Resolves#10458
### Changes 🏗️
Improve logic in `useAgentGraph`:
- Correctly handle unset `flowVersion` in checks in hooks
- Prevent unnecessary WebSocket re-connects
- Remove redundant WebSocket connection management logic
- Untangle hooks for initial load and set-up
- Simplify block filtering logic
### 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:
- Edit an agent in the builder
- [x] WebSocket doesn't re-connect unnecessarily
- [x] Graph doesn't reset on WebSocket re-connect
- [x] Graph doesn't reset on LaunchDarkly re-connect
## Changes 🏗️https://github.com/user-attachments/assets/dd635fa1-d8ea-4e5b-b719-2c7df8e57832
Using [LaunchDarkly](https://launchdarkly.com/), introduce the concept
of "beta" blocks, which are blocks that will be disabled in production
unless enabled via a feature flag. This allows us to safely hide and
test certain blocks in production safely.
## 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] Checkout and run FE locally
- [x] With the `beta-blocks` flag `disabled` in LD
- [x] Go to the builder and see **you can't** add the blocks specified
on the flag
- [x] With the `beta-blocks` flag `enabled` in LD
- [x] Go to the builder and see **you can** add the blocks specified on
the flag
### For configuration changes:
- [x] `.env.example` is updated or already compatible with my changes
🚧 We need to add the `NEXT_PUBLIC_LAUNCHDARKLY_CLIENT_ID` to the dev and
prod environments.
- Resolves#10313
- Resolves#10333
Before:
https://github.com/user-attachments/assets/a105b2b0-a90b-4bc6-89da-bef3f5a5fa1f
- No credentials input
- Stuttery experience when panning or zooming the viewport
After:
https://github.com/user-attachments/assets/f58d7864-055f-4e1c-a221-57154467c3aa
- Pretty much the same UX as in the Library, with fully-fledged
credentials input support
- Much smoother when moving around the canvas
### Changes 🏗️
Frontend:
- Add credentials input support to Run UX in Builder
- Pass run inputs instead of storing them on the input nodes
- Re-implement `RunnerInputUI` using `AgentRunDetailsView`; rename to
`RunnerInputDialog`
- Make `AgentRunDraftView` more flexible
- Remove `RunnerInputList`, `RunnerInputBlock`
- Make moving around in the Builder *smooooth* by reducing unnecessary
re-renders
- Clean up and partially re-write bead management logic
- Replace `request*` fire-and-forget methods in `useAgentGraph` with
direct action async callbacks
- Clean up run input UI components
- Simplify `RunnerUIWrapper`
- Add `isEmpty` utility function in `@/lib/utils` (expanding on
`_.isEmpty`)
- Fix default value handling in `TypeBasedInput` (**Note:** after all
the changes I've made I'm not sure this is still necessary)
- Improve & clean up Builder test implementations
Backend + API:
- Fix front-end `Node`, `GraphMeta`, and `Block` types
- Small refactor of `Graph` to match naming of some `LibraryAgent`
attributes
- Fix typing of `list_graphs`,
`get_graph_meta_by_store_listing_version_id` endpoints
- Add `GraphMeta` model and `GraphModel.meta()` shortcut
- Move `POST /library/agents/{library_agent_id}/setup-trigger` to `POST
/library/presets/setup-trigger`
### 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:
- Test the new functionality in the Builder:
- [x] Running an agent with (credentials) inputs from the builder
- [x] Beads behave correctly
- [x] Running an agent without any inputs from the builder
- [x] Scheduling an agent from the builder
- [x] Adding and searching blocks in the block menu
- [x] Test that all existing `AgentRunDraftView` functionality in the
Library still works the same
- [x] Run an agent
- [x] Schedule an agent
- [x] View past runs
- [x] Run an agent with inputs, then edit the agent's inputs and view
the agent in the Library (should be fine)
## Summary
- Enhanced graph execution cancellation and cleanup mechanisms
- Improved error handling and logging for graph execution lifecycle
- Added timeout handling for graph termination with proper status
updates
- Exposed a new API for stopping graph based on only graph_id or user_id
- Refactored logging metadata structure for better error tracking
## Key Changes
### Backend
- **Graph Execution Management**: Enhanced `stop_graph_execution` with
timeout-based waiting and proper status transitions
- **Execution Cleanup**: Added proper cancellation waiting with timeout
handling in executor manager
- **Logging Improvements**: Centralized `LogMetadata` class and improved
error logging consistency
- **API Enhancements**: Added bulk graph execution stopping
functionality
- **Error Handling**: Better exception handling and status management
for failed/cancelled executions
### Frontend
- **Status Safety**: Added null safety checks for status chips to
prevent runtime errors
- **Execution Control**: Simplified stop execution request handling
## Test Plan
- [x] Verify graph execution can be properly stopped and reaches
terminal state
- [x] Test timeout scenarios for stuck executions
- [x] Validate proper cleanup of running node executions when graph is
cancelled
- [x] Check frontend status chips handle undefined statuses gracefully
- [x] Test bulk execution stopping functionality
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
Complete the implementation of the Agent Run Scheduling UX in the
Library.
Demo:
https://github.com/user-attachments/assets/701adc63-452c-4d37-aeea-51788b2774f2
### Changes 🏗️
Frontend:
- Add "Schedule" button + dialog + logic to `AgentRunDraftView`
- Update corresponding logic on `AgentRunsPage`
- Add schedule name field to `CronSchedulerDialog`
- Amend Builder components `useAgentGraph`, `FlowEditor`,
`RunnerUIWrapper` to also handle schedule name input
- Split `CronScheduler` into `CronScheduler`+`CronSchedulerDialog`
- Make `AgentScheduleDetailsView` more fully functional
- Add schedule description to info box
- Add "Delete schedule" button
- Update schedule create/select/delete logic in `AgentRunsPage`
- Improve schedule UX in `AgentRunsSelectorList`
- Switch tabs automatically when a run or schedule is selected
- Remove now-redundant schedule filters
- Refactor `@/lib/monitor/cronExpressionManager` into
`@/lib/cron-expression-utils`
Backend + API:
- Add name and credentials to graph execution schedule job params
- Update schedule API
- `POST /schedules` -> `POST /graphs/{graph_id}/schedules`
- Add `GET /graphs/{graph_id}/schedules`
- Add not found error handling to `DELETE /schedules/{schedule_id}`
- Minor refactoring
Backend:
- Fix "`GraphModel`->`NodeModel` is not fully defined" error in
scheduler
- Add support for all exceptions defined in `backend.util.exceptions` to
RPC logic in `backend.util.service`
- Fix inconsistent log prefixing in `backend.executor.scheduler`
### 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:
- Create a simple agent with inputs and blocks that require credentials;
go to this agent in the Library
- Fill out the inputs and click "Schedule"; make it run every minute
(for testing purposes)
- [x] -> newly created schedule appears in the list
- [x] -> scheduled runs are successful
- Click "Delete schedule"
- [x] -> schedule no longer in list
- [x] -> on deleting the last schedule, view switches back to the Runs
list
- [x] -> no new runs occur from the deleted schedule
## Changes 🏗️
### ESLint Config
1. **Disabled `react-hooks/exhaustive-deps`:**
- to prevent unnecessary dependency proliferation and rely on code
review instead
2. **Added
[`next/typescript`](https://nextjs.org/docs/app/api-reference/config/eslint#with-typescript):**
- to the ESLint config to make sure we also have TS linting rules
3. **Added custom rule for `@typescript-eslint/no-unused-vars`:**
- to allow underscore-prefixed variables (convention for intentionally
unused), in some cases helpful
From now on, whenever we have unused variables or imports, the `lint` CI
will fail 🔴 , thanks to `next/typescript` that adds
`typescript-eslint/no-unused-vars` 💆🏽
### Minor Fixes
- Replaced empty interfaces with type aliases to resolve
`@typescript-eslint/no-empty-object-type` warnings
- Fixed unsafe non-null assertions with proper null checks
- Removed `@ts-ignore` comments in favour of proper type casting ( _when
possible_ 🙏🏽 )
### Google Analytics Component
- Changed Next.js Script strategy from `beforeInteractive` to
`afterInteractive` to resolve Next.js warnings
- this make sure loading analytics does not block page render 🙏🏽 (
_better page load time_ )
### Are these changes safe?
As long as the Typescript compiler does not complain ( check the
`type-check` job ) we should be save. Most changes are removing unused
code, if that code would be used somewhere else the compiler should
catch it and tell us 🫶
I also typed some code when possible, or bypassed the linter when I
thought it was fair for now. I disabled a couple ESLint rules. Most
importantly the `no-explicity-any` one as we have loads of stuff untyped
yet ( _this should be improved once API types are generated for us_ ).
### DX
Added some settings on `.vscode` folder 📁 so that files will be
formatted on save and also ESLint will fix errors on save when able 💯
### 📈 **Result:**
- ✅ All linting errors resolved
- ✅ Improved TypeScript strict mode compliance
- ✅ Better developer experience with cleaner code
## 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] Lint CI job passes
- [x] There is not type errors ( _TS will catch issue related to these
changes_ )
There are a few UI bugs on the builder that this PR addresses.
<img width="554" alt="image"
src="https://github.com/user-attachments/assets/1be70197-de7e-40fe-ab11-405c145e763d"
/>
### Changes 🏗️
Fix these UI issues:
* (screenshot attached above) Key-value input width was unintentionally
maxed out due to a stale CSS rule.
* When multiple executions within the same node are running, we pick the
latest status, making one running and one completed execution displayed
as completed.
* No balance errors were executed, only displayed while at least one
node execution was triggered, while this can be done directly when the
execution request is triggered.
* Run & Stop button glitch: it's still showing as stopped when the graph
is still running, this is due to way the UI code tracks execution in the
node-level, instead of graph level.
### 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:
<!-- Put your test plan here: -->
- [x] Manual tests on the described behaviours.
**Goal**: Allow parallel runs within a single node. Currently, we
prevent this to avoid unexpected ordering of the execution.
### Changes 🏗️
#### Executor changes
We decoupled the node execution output processing, which is responsible
for deciding the next executions from the node executor code.
Currently, `execute_node` does two big things:
* Runs the block’s execute(...) (which yields outputs).
* immediately enqueues the next nodes based on those outputs.
This PR makes:
* execute_node(node_exec) -> stream of (output_name, data). That purely
runs the block and yields each output as soon as it’s available.
* Move _enqueue_next_nodes into the graph executor. So the next
execution is handled serially by the graph executor to avoid concurrency
issues.
#### UI changes
The change on the executor also fixes the behavior of the execution
update to the UI We will report the execution output to the UI as soon
as it is available, not when the node execution is fully completed.
This, however, broke the bread calculation logic that assumes each
execution update will never overlap. So the change in this PR makes the
bead calculation take the overlap / duplicated execution update into
account, and simplify the overall calculation logic.
### 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:
<!-- Put your test plan here: -->
- [x] Execute this agent and observe its concurrency ordering
<img width="1424" alt="image"
src="https://github.com/user-attachments/assets/0fe8259f-9091-4ecc-b824-ce8e8819c2d2"
/>
### Changes 🏗️
This PR adds `Run 10 agents` step to wallet tasks that can be done by
running any agents 10 times either from Library or Builder (onboarding
agent run also counts).
- Merge `Finish onboarding` and `See results` steps into one in the
wallet
- User is redirected directly to onboarding agent runs in Library after
congrats screen
- Add `RUN_AGENTS` step and `agentRuns` integer to schema and related
migration
- Running agent from Library and Builder increments `agentRuns`
- Open NPS survey popup when 10 agents are run
- Fix resuming onboarding on login when unfinished
- Remove no longer needed `get-results.mp4` tutorial video
### 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] Onboarding can be completed and proper reward is awarded
- [x] `Run 10 agents` can be completed and reward is awarded
- [x] When unning different agents and the same agent
- [x] Running from library and builder counts
- [x] Onboarding is resumed to last finished step on login
- Resolves#9929
### Changes 🏗️
- Implement `BackendAPI.onWebSocketConnect(..)`
- Improve reliability and reactivity of `/library/agents/[id]`:
- Refresh page data and (re)subscribe and on WebSocket (re)connect
- Break up multi-action hooks into smaller parts to reduce unnecessary
re-renders and requests
- Reduce duplicate requests
- Use `onWebSocketConnect` in `useAgentGraph` as well
- Tidy up `autogpt-server-api/client.ts` a bit
### 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:
- Go to `/library/agents/[id]`
- Run the agent
- [x] -> UI should update normally with execution updates
- Suspend your computer, or restart the backend (or WS server) to break
the connection
- DO NOT REFRESH THE TAB ITSELF
- [x] -> On reconnect, page data should be refreshed (check in network
tab of dev tools)
- Run the agent again
- [x] -> UI should update normally with execution updates
Smart Decision Block was not able to work with sub agent with custom
name input & the bead were not properly propagated in the execution UI.
The scope of this PR is fixing it.
### Changes 🏗️
* Introduce an easy to parse format of tool edge:
`{tool}_^_{func}_~_{arg}`. Graph using SmartDecisionBlock needs to be
re-saved before execution to work.
* Reduce cluttering on a smart decision block logic.
* Fix beads not being shown for a smart decision block tool calling.
### 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] Execute an SDM with some special character input as a tool
<img width="672" alt="image"
src="https://github.com/user-attachments/assets/873556b3-c16a-4dd1-ad84-bc86c636c406"
/>
### Changes 🏗️
- Update onboarding to give user rewards for completing steps
- Remove `canvas-confetti` lib and add `party-js` instead; the former
didn't allow to play confetti from a component
- Add onboarding videos in `frontend/public/onboarding/`
- Remove Balance (`CreditsCard.tsx`) and add openable `Wallet.tsx` (and
accompanying `WalletTaskGroup.tsx`) instead that displays grouped
onboarding tasks with descriptions and short instructional videos
- Further relevant updates to `useOnboarding`, `types.ts`
- Implement onboarding rewards
- Add `onboarding_reward` function in `credit.py` that is used to reward
user for finished onboarding tasks safely - transaction key is
deterministic, so the same user won't be rewarded twice for the same
step.
- Add `reward_user` in `onboarding.py`
- Update `UserOnboarding` model and add a migration
<img width="464" alt="Screenshot 2025-04-05 at 6 06 29 PM"
src="https://github.com/user-attachments/assets/fca8d09e-0139-466b-b679-d24117ad01f0"
/>
### 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] Onboarding works
- [x] Tasks can be completed
- [x] Rewards are added correctly for all completed tasks
- Follow-up to #9657
<img width="280" alt="image"
src="https://github.com/user-attachments/assets/2f3cd683-db63-485f-8914-5654c34f1a4c"
/>
<img width="520" alt="image"
src="https://github.com/user-attachments/assets/de7e7cb9-61d4-4071-aea8-393ff5200c54"
/>
### Changes 🏗️
* Implement the input UI for Agent Input subtypes.
* Refactor node-input-component, extra out data type decision logic,
share it with runner/library input.
* Add `format` field for short-text, long-text, and mediafile type.
* Unify UI data type enum.
Out of scope:
- Styling for these inputs.
### 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] Use all the available agent input subtypes in an agent and run it
---------
Co-authored-by: Reinier van der Leer <pwuts@agpt.co>
- Resolves#8782
### Changes 🏗️
- feat(frontend/library): Use WS subscription to get real-time execution
updates
- feat(backend/ws_api): Send `GraphExecutionUpdate` on all new agent I/O
- Include agent I/O in `GraphExecutionUpdate` (by subclassing
`GraphExecution`)
- Add `IO_BLOCK_IDs` to `.blocks.io`
- feat(backend/ws_api): Add `subscribe_graph_executions` method to
WebSocket API
- feat(backend): Withhold `GraphExecution.node_executions` from requests
by non-graph-owners
- Split `GraphExecutionWithNodes` off of `GraphExecution`
- Use `GraphExecution` as much as possible, as it's a much cheaper query
than `GraphExecutionWithNodes`
- refactor(frontend): Make `GraphExecution.node_executions` optional
- fix(frontend): Parse dates in responses of `/executions` and
`/graphs/{graph_id}/executions`
- refactor(frontend/library): Move sorting logic for agent runs list
from `AgentRunsPage` to `AgentRunsSelectorList`
- refactor(backend/ws_api): Clean up message handler implementations
- refactor(backend/tests): Use `.data.execution.get_graph_execution(..)`
directly instead of `AgentServer.test_get_graph_run_results(..)`
Out-of-scope changes:
- refactor(backend): Remove unnecessary query include from
`.data.graph.get_graph_metadata(..)`
Demo:
https://github.com/user-attachments/assets/8ea6225d-7334-49cb-a522-83f153d840da
### 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:
- Go to `/library/agents/[id]` for an agent with inputs and outputs
- Draft and run a new run
- [x] -> should appear in the list of runs at the top
- [x] -> should be selected as soon as the request finishes
- [x] -> new I/O should appear as it is generated
- [x] -> status should be updated in real-time (both in list and in
adjacent details view)
- Click "Run again"
- [x] -> should appear in the list of runs at the top
- [x] -> should be selected as soon as the request finishes
- [x] -> new I/O should appear as it is generated
- [x] -> status should be updated in real-time (both in list and in
adjacent details view)
- Click "Open in builder" under "Agent actions"; run the agent from the
builder
- [x] -> should work the same as before
- [x] -> node I/O should appear in real-time
- [x] -> node execution statuses should update in real-time
- Prep work for #8782
- Prep work for #8779
### Changes 🏗️
- refactor(platform): Differentiate graph/node execution events
- fix(platform): Subscribe to execution updates by `graph_exec_id`
instead of `graph_id`+`graph_version`
- refactor(backend): Move all execution related models and functions
from `.data.graph` to `.data.execution`
- refactor(backend): Reorganize & refactor `.data.execution`
- fix(libs): Remove `load_dotenv` in `.auth.config` to fix test config
issues
- dx: Bump version of `black` in pre-commit config to v24.10.0 to match
poetry.lock
- Other minor refactoring in both frontend and backend
### 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:
- Run an agent in the builder
- [x] -> works normally, node I/O is updated in real time
- Run an agent in the library
- [x] -> works normally
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
The current execution update is unreliable, once you lose WebSocket
connection, you will receive no updates.
### Changes 🏗️
Fix web socket re-connection logic.
### 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] Run the app and execute an agent, then restart the API server, and
re-execute the app without refreshing the page.
### Changes 🏗️
This is a follow-up of
https://github.com/Significant-Gravitas/AutoGPT/pull/9610
* Addressing the PR comments described in the mentioned PR
* Removed debug logging
* Fix image state loading logic on agent upload process
### 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] Create an library agent and try to create a store listing.
<img width="1409" alt="image"
src="https://github.com/user-attachments/assets/dc86dc97-a33f-4336-ab90-19a53c6f7e0f"
/>
### Changes 🏗️
There are a few agent-loop issues that this PR is addressing:
* There is a lack of support for agent-loop in Anthropic.
* Duplicated system & user prompt as the main objective prompt in the
agent loop.
* A long rendered text of conversation history by
SmartDecisionMakerBlock agent-loop in the UI.
* A lack of execution input being rendered in the execution list making
it harder to debug.
https://github.com/user-attachments/assets/be430000-bde0-40c6-8f2e-c97ce45b5ed1
### 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:
<!-- Put your test plan here: -->
- [x] Create from scratch and execute an agent with at least 3 blocks
using SmartDecisionMaker Block.
- Resolves#9542
### Changes 🏗️
- feat(platform): Add "Delete run" button on `/library/agents/[id]` w/
confirm dialog
- Add `AgentGraphExecution.isDeleted` column for soft delete
- Add `DELETE /api/executions/{graph_exec_id}` endpoint
- feat(frontend): Add "Stop run" button on `/library/agents/[id]`
Technical improvements:
- refactor(frontend): Generalize `AgentDeleteConfirmDialog` ->
`DeleteConfirmDialog`
- refactor(frontend): Brand `GraphExecution.execution_id` and
`Schedule.id` to prevent mixing
### 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:
- Click "Delete run" under "Run actions" in the right sidebar
- [x] -> Confirmation dialog should show
- Click "Delete"
- [x] -> run should disappear from list
- [x] -> view should switch to "Draft new run" view
- [x] -> refresh page -> run should not be in list
- Click "Delete" in the menu on a run in the list
- [x] -> Confirmation dialog should show
- Click "Delete"
- [x] -> run should disappear from list
- [x] -> refresh page -> run should not be in list
---------
Co-authored-by: Krzysztof Czerwinski <34861343+kcze@users.noreply.github.com>
Originally we did not allow Blocks to be used as tools due to the
limitations of communicating the correct tool function signatures.
It has however, been decided to allow them to be used knowing that there
are limitations with them.
### Changes 🏗️
- Added ability to execute blocks as tools
### Checklist 📋
<img width="613" alt="Screenshot 2025-02-25 at 12 49 26"
src="https://github.com/user-attachments/assets/e614f56d-2bdc-46c9-8c2c-e56f80343bde"
/>
- create an agent with an SDM block and a block as a tool
- run agent and make sure the block can be called as a tool
#### 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:
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
## Task
The SmartDecisionMakerBlock is a specialized block in a graph-based
system that leverages a language model (LLM) to make intelligent
decisions about which tools or functions to invoke based on a
user-provided prompt. It is designed to process input data, interact
with a language model, and dynamically determine the appropriate tools
to call from a set of available options, making it a powerful component
for AI-driven workflows.
## How It Works in Practice
- **Scenario:** Imagine a workflow where a user inputs, "Send an email
to John about the meeting." The SmartDecisionMakerBlock is connected to
tools like send_email, schedule_meeting, and search_contacts.
- **Execution:**
1. The block receives the prompt and system instructions (e.g., "Choose
a function to call").
2.It identifies the available tools from the graph and constructs their
signatures (e.g., send_email(recipient, subject, body)).
3. The LLM analyzes the prompt and decides to call send_email with
arguments like recipient: "John", subject: "Meeting", body: "Let’s
discuss...".
4. The block yields these tool-specific outputs, which can be picked up
by downstream nodes to execute the email-sending action.
## Changes 🏗️
- Add the Smart Decision Maker (SDM) block.
- Break circular imports in integration code.
## Work in Progress
⚠️ **Important note this is a temporary UX for the system - UX will be
addressed in a future PR** ⚠️
### Current Status
I’m currently focused on the smart decision logic. The main additions in
the ongoing PR include:
- Defining function signatures for OpenAI function-calling schemas based
on node links and the linked blocks.
- Adding tests for function signature generation.
- Force all tool calls to be made via an agent. (Need to uncomment)
- Restrict each tool call entry to a single node.
- simplify the output emission process, to emit each parameter one at a
time.
- Change test to use agents and hardcode output how I think it should
work to test it does actually work
- Hook up openai, in a simplified way, to test the function calling
(mock for testing)
- Once all the above is working, use credentials system and build of
llm.py
### What’s Next
- Review Process
### Reviewers Phase 1
This PR is now ready for review, during the first phase of reviews I'm
looking for comments on approach and logic.
Out of scope: code style and organization at this stage
### Reviewers Phase 2
Once we are all happy with the approach and logic. We can open the
review process to general code quality and nits, to be considered.
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
When opened graph is running and Builder page is refreshed the bottom
says `Run` but should show `Stop` instead.
### Changes 🏗️
- Fix state on refresh, so that the bottom button is `Stop` if graph is
currently running
### 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] Refresh still retrieves past and ongoing execution updates
- [x] Bottom Builder button says `Stop` when page is refreshed and graph
is running
- [x] `Stop` button works and terminates execution
