There is no 100% accurate way of retrying an agent that has been
terminated. And the safest way to avoid executing an agent wrong is
minimizing the chance of an agent execution being terminated. A whole
set of mechanism to make sure the agent is retried on failure is still
in place and improved, this is used as our best-effort reliability
mechanism.
### Changes 🏗️
* Cap SIGINT & SIGTERM to be raised at most once, so the executor can
gracefully handle the stopping.
* SIGINT & SIGTERM will stop the execution request message consumption,
but not agent execution.
* Executor process will only stop if all the in-flight agent executions
are completed or terminated.
* Avoid retrying the agent stop command on AgentExecutorBlock on
timeout.
### 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] Run agent, send SIGTERM to the executor pod, execution should not
be interrupted.
- [x] Run agent, send SIGKILL to the executor pod, execution should be
transferred to another pod.
<!-- Clearly explain the need for these changes: -->
Toran hit an error on reading a snippet incorrectly
### Changes 🏗️
Does fallback getting from dictionary when building email objects
<!-- Concisely describe all of the changes made in this pull request:
-->
### Checklist 📋
#### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] I have made a test plan
- [ ] I have tested my changes according to the test plan:
<!-- Put your test plan here: -->
- [ ] Deploy to dev and have Toran test against his inbox
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
### Changes 🏗️
- Added `_convert_bools()` function to recursively convert string
boolean values ("true"/"false") to actual Python booleans
- Applied boolean conversion to all Airtable API endpoints that send
JSON data to ensure proper type casting
- Fixed parameters that were incorrectly converted to strings (e.g.,
`typecast`, `returnFieldsByFieldId`) to maintain their boolean types
This fix addresses an issue where the Airtable API was not properly
handling boolean values passed as strings, which could cause API calls
to fail or behave unexpectedly.
### 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] Tested boolean field updates with string values "true" and "false"
- [x] Verified that boolean parameters like `typecast` and
`returnFieldsByFieldId` are properly handled
- [x] Confirmed that nested boolean values in records are correctly
converted
- [x] Tested that non-boolean values remain unchanged
[Working Airtable
Example_v56.json](https://github.com/user-attachments/files/21594436/Working.Airtable.Example_v56.json)
Updated login and signup pages to display the Turnstile CAPTCHA and
require verification only when running in a cloud environment. This
prevents unnecessary CAPTCHA prompts in local or non-cloud deployments.
### Changes 🏗️
Locally when you try to login with the wrong password, and you update
and login again, you get a warning about captcha which is wrong, so this
fix makes it so the captcha will only when running in a cloud
### 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] Try to login with the wrong password, get "Invalid login
credentials" and try to login again, you should keep getting "Invalid
login credentials" and it should not mention captcha
Graph evaluation should stop naturally once all the node execution are
stopped.
### Changes 🏗️
Avoid stopping agent node evaluation when stopping graph
### 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] CI
The node execution status can be done before the output persistence,
making the output be persisted when the node execution status is already
completed.
### Changes 🏗️
* Re-order the node execution status & output persistence logic.
* Make agent.py avoid yielding the same node_exec_id twice (that can be
caused by the above issue).
### 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] Existing CI
## Summary
- Removed unused metadata functions from user.py (get_user_metadata,
update_user_metadata)
- Removed unused execution and database functions from database.py and
related imports
- Added NodeExecutionStats validation in execution.py
- Updated CLAUDE.md with PR and commit conventions
## Changes Made
### `/backend/backend/data/user.py`
- Removed `get_user_metadata()` function (unused)
- Removed `update_user_metadata()` function (unused)
- Removed unused import `UserMetadataRaw`
### `/backend/backend/data/execution.py`
- Added `NodeExecutionStats` validation in `from_db()` method
### `/backend/backend/executor/database.py`
- Removed unused imports and function exposures
- Cleaned up DatabaseManagerClient to remove unused client methods
### `/CLAUDE.md`
- Added documentation for creating pull requests
- Added conventional commit types and scopes guide
## Testing
- Existing tests should pass as removed functions were not being used
- No new functionality added
## Checklist
- [x] Code follows the project's style guidelines
- [x] Self-review completed
- [x] Changes are backward compatible
- [x] No new warnings introduced
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
Graph and Node execution can fail due to so many reasons, sometimes this
messes up the stats tracking, giving an inaccurate result. The scope of
this PR is to minimize such issues.
### Changes 🏗️
* Catch BaseException on time_measured decorator to catch
asyncio.CancelledError
* Make sure update node & graph stats are executed on cancellation &
exception.
* Protect graph execution stats update under the thread lock to avoid
race condition.
### 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] Existing automated tests.
---------
Co-authored-by: Claude <noreply@anthropic.com>
This PR refactors the Ayrshare integration to remove the centralized
`get_ayrshare_profile_key` function from the credentials store and
instead retrieve the profile key directly within each Ayrshare block.
This change improves code organization by keeping Ayrshare-specific
logic within the Ayrshare module.
### Changes 🏗️
- **Refactored Ayrshare profile key retrieval**: Moved profile key
fetching logic from the credentials store into the Ayrshare blocks
- **Added `get_profile_key` helper function** in
`autogpt_platform/backend/backend/blocks/ayrshare/_util.py` to fetch the
profile key from user integrations
- **Updated all 15 Ayrshare social media blocks** to use `user_id`
instead of `profile_key` parameter and fetch the profile key internally
- **Removed `get_ayrshare_profile_key` method** from
`autogpt_platform/backend/backend/integrations/credentials_store.py`
- **Removed Ayrshare-specific logic** from
`autogpt_platform/backend/backend/executor/manager.py` that was passing
profile keys to blocks
- **Updated router** in
`autogpt_platform/backend/backend/server/integrations/router.py` to
directly fetch user integrations instead of using the removed method
### 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] Test posting to X/Twitter to check credentials flow
- [x] Verify profile key retrieval works correctly for authenticated
users
- [x] Test Ayrshare SSO URL generation flow
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
This PR helps us bypass the proxy server in server-side requests,
allowing us to directly send requests to the backend and reduce latency.
### Changes 🏗️
- Introduced server-side detection to dynamically set the base URL for
API requests.
- Added error handling for server-side requests to log failures and
throw errors appropriately.
- Updated header management to include authentication tokens when
applicable.
### Checklist 📋
- [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 E2E tests are working.
- [x] I have manually checked the server-side and client-side
components, and both are working perfectly.
- Resolves -
https://github.com/Significant-Gravitas/AutoGPT/issues/10433
- Depends on -
https://github.com/Significant-Gravitas/AutoGPT/pull/10427
- Need to review this pr, once this issue is fixed -
https://github.com/Significant-Gravitas/AutoGPT/issues/10404
I’ve created additional tests for the agents marketplace page
Tests that I have added
- Add to library button works and agent appears in library.
- Download button functionality works.
- Agent page details are visible.
- User can access agent page when logged in.
- User can access agent page when logged out
#### 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] I have done all the tests and they are working perfectly
---------
Co-authored-by: Nicholas Tindle <nicholas.tindle@agpt.co>
Co-authored-by: Lluis Agusti <hi@llu.lu>
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
Co-authored-by: Ubbe <hi@ubbe.dev>
We currently use infinite scroll pagination in multiple places, but our
strategies vary across these locations. This repetitive code writing is
not ideal, and our current methods are also complex. We’re not utilising
React Query’s useInfiniteQuery hooks effectively.
To address these issues, we’re introducing a new component called
`InfiniteScroll` that handles pagination independently.
### How to use it?
- Use React Query’s `useInfiniteHook` to return multiple data points.
For pagination, we only need `fetchNextPage`, `hasNextPage`, and
`isFetchingNextPage`.
```ts
const {
data: agents,
fetchNextPage,
hasNextPage,
isFetchingNextPage,
isLoading: agentLoading,
} = useGetV2ListLibraryAgentsInfinite(
{
page: 1,
page_size: 8,
search_term: searchTerm || undefined,
sort_by: librarySort,
},
);
```
- Simply pass these three data points and the current data length to the
`InfiniteScroll` component. That's it
```tsx
<InfiniteScroll
dataLength={agents.length}
isFetchingNextPage={isFetchingNextPage}
fetchNextPage={fetchNextPage}
hasNextPage={hasNextPage}
loader={<LoadingSpinner />}
>
...
```
### Changes
- Add the `InfiniteScroll.tsx` component for consistency and simplicity
in pagination across the frontend.
- Update the current library page to use the `InfiniteScroll` component.
### Checklist 📋
- [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] I’ve tested everything locally, and it’s working perfectly fine.
## Summary
This PR adds a timeout guard to the `locked_transaction` function used
for credit transactions to prevent indefinite blocking and improve
reliability.
## Changes
- Modified `locked_transaction` in `/backend/backend/data/db.py` to add
proper timeout handling
- Set `lock_timeout` and `statement_timeout` to prevent indefinite
blocking
- Updated function signature to use default timeout parameter
- Added comprehensive docstring explaining the locking mechanism
## Motivation
The previous implementation could potentially block indefinitely if a
lock couldn't be acquired, which could cause issues in production
environments, especially for critical credit transactions.
## Testing
- Existing tests pass
- The timeout mechanism ensures transactions won't hang indefinitely
- Advisory locks are properly released on commit/rollback
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
This PR improves the reliability of the executor system by addressing
several race conditions and improving error handling throughout the
execution pipeline.
### Changes 🏗️
- **Consolidated exception handling**: Now using `BaseException` to
properly catch all types of interruptions including `CancelledError` and
`SystemExit`
- **Atomic stats updates**: Moved node execution stats updates to be
atomic with graph stats updates to prevent race conditions
- **Improved cleanup handling**: Added proper timeout handling (3600s)
for stuck executions during cleanup
- **Fixed concurrent update race conditions**: Node execution updates
are now properly synchronized with graph execution updates
- **Better error propagation**: Improved error type preservation and
status management throughout the execution chain
- **Graph resumption support**: Added proper handling for resuming
terminated and failed graph executions
- **Removed deprecated methods**: Removed `update_node_execution_stats`
in favor of atomic updates
### 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 a graph with multiple nodes and verify stats are updated
correctly
- [x] Cancel a running graph execution and verify proper cleanup
- [x] Simulate node failures and verify error propagation
- [x] Test graph resumption after termination/failure
- [x] Verify no race conditions in concurrent node execution updates
#### For configuration changes:
- [x] `.env.example` is updated or already compatible with my changes
- [x] `docker-compose.yml` is updated or already compatible with my
changes
- [x] I have included a list of my configuration changes in the PR
description (under **Changes**)
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
Sometimes we receive an error where the service is not connected to the
DB, but we have started receiving traffic, making the request fail.
### Changes 🏗️
Make the `/health_check` endpoint also check the database connection.
### 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] Existing CI, manual test
- Resolves -
https://github.com/Significant-Gravitas/AutoGPT/issues/10428
- Depends on -
https://github.com/Significant-Gravitas/AutoGPT/pull/10427
- Need to review this pr, once this issue is fixed -
https://github.com/Significant-Gravitas/AutoGPT/issues/10404
I’ve created additional tests for the creators marketplace page
Tests that I have added
- User can access creator's page when logged out.
- User can access creator's page when logged in.
- Creator page details are visible.
- Agents in agent by sections navigation works.
#### 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] I have done all the tests and they are working perfectly
With this PR, we’re changing the data fetching strategy on the
marketplace page. We’re now using autogenerated React queries.
### Changes
- Splits separate render logic and hook logic.
- Update the data fetching strategy.
- Currently, we’re seeing agents in the featured section and creators in
the featured creators section, even if they’re not set to “isFeatured”
true. I’ve fixed that also.
### Checklist 📋
- [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 marketplace E2E tests are working.
- [x] I’ve tested all the links and checked if everything renders
perfectly on the marketplace page.
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
## Summary
- Adds AI-generated activity status summaries for agent execution
results
- Provides users with conversational, non-technical summaries of what
their agents accomplished
- Includes comprehensive execution data analysis with honest failure
reporting
## Changes Made
- **Backend**: Added `ActivityStatusGenerator` module with async LLM
integration
- **Database**: Extended `GraphExecutionStats` and `Stats` models with
`activity_status` field
- **Frontend**: Added "Smart Agent Execution Summary" display with
disclaimer tooltip
- **Settings**: Added `execution_enable_ai_activity_status` toggle
(disabled by default)
- **Testing**: Comprehensive test suite with 12 test cases covering all
scenarios
## Key Features
- Collects execution data including graph structure, node relations,
errors, and I/O samples
- Generates user-friendly summaries from first-person perspective
- Honest reporting of failures and invalid inputs (no sugar-coating)
- Payload optimization for LLM context limits
- Full async implementation with proper error handling
## Test Plan
- [x] All existing tests pass
- [x] New comprehensive test suite covers success/failure scenarios
- [x] Feature toggle testing (enabled/disabled states)
- [x] Frontend integration displays correctly
- [x] Error handling and edge cases covered
🤖 Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: Claude <noreply@anthropic.com>
This PR updates the existing E2E test data script to support the
creation of featured creators and featured agents. Previously, these
entities were not included, which limited our ability to fully test
certain flows during Playwright E2E testing.
### Changes
- Added logic to create featured creators
- Added logic to create featured agents
### 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 tests are passing locally after updating the data script.
- Resolves -
https://github.com/Significant-Gravitas/AutoGPT/issues/10426
- Need to review this pr, once this issue is fixed -
https://github.com/Significant-Gravitas/AutoGPT/issues/10404
I’ve created additional tests for the main page, divided into two parts:
one for basic functionality and the other for edge cases.
**Basic functionality:**
- Users can access the marketplace page when logged out.
- Users can access the marketplace page when logged in.
- Featured agents, top agents, and featured creators are visible.
- Users can navigate and interact with marketplace elements.
- The complete search flow works correctly.
**Edge cases:**
- Searching for a non-existent item shows no results.
### Changes
- Introduced a new test suite for the marketplace, covering basic
functionality and edge cases.
- Implemented the MarketplacePage class to encapsulate interactions with
the marketplace page.
- Added utility functions for assertions, including visibility checks
and URL matching.
- Enhanced the LoginPage class with a goto method for navigation.
- Established a comprehensive search flow test to validate search
functionality.
#### 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] I have done all the tests and they are working perfectly
---------
Co-authored-by: Nicholas Tindle <nicholas.tindle@agpt.co>
Co-authored-by: Lluis Agusti <hi@llu.lu>
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
Co-authored-by: Ubbe <hi@ubbe.dev>
The notification service was running an inefficient polling loop that
constantly
checked each queue sequentially with 1-second timeouts, even when queues
were
empty. This caused:
- High CPU usage from continuous polling
- Sequential processing that blocked queues from being processed in
parallel
- Unnecessary delays from timeout-based polling instead of event-driven
consumption
- Poor throughput (500-2,000 messages/second) compared to potential
(8,000-12,000
messages/second)
## Changes 🏗️
- Replaced polling-based _run_queue() with event-driven _consume_queue()
using
async iterators
- Implemented concurrent queue consumption using asyncio.gather()
instead of
sequential processing
- Added QoS settings (prefetch_count=10) to control memory usage
- Improved error handling with message.process() context manager for
automatic
ack/nack
- Added graceful shutdown that properly cancels all consumer tasks
- Removed unused QueueEmpty import
## Checklist 📋
### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] I have made a test plan
- [ ] I have tested my changes according to the test plan:
- [ ] Deploy to test environment and monitor CPU usage
- [ ] Verify all queue types (immediate, admin, batch, summary) process
messages
correctly
- [ ] Test graceful shutdown with messages in flight
- [ ] Monitor that database management service remains stable
- [ ] Check logs for proper consumer startup messages
- [ ] Verify messages are properly acked/nacked on success/failure
---------
Co-authored-by: Claude <claude@users.noreply.github.com>
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
Some failure on DB RPC can cause agent execution failure. This change
makes sure the error chance is minimized.
### Changes 🏗️
* Enable request retry
* Increase transaction timeout
* Use better typing on the DB query
* Gracefully handles insufficient balance
### 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
## Changes 🏗️
- Moved API call from `usePublishAgentModal` to `useAgentInfoStep` for
better encapsulation
- overall cleaner state management + [state
colocation](https://kentcdodds.com/blog/state-colocation-will-make-your-react-app-faster)
- Added loading states with a spinner to the submit button during API
call
- Removed redundant validation: now relies entirely on zod schema
validation
- All thumbnails now use 16:9 (`aspect-video`) aspect ratio for
consistency
- Highlight selected thumbnails with blue border
- Table alignment fixes
- Rename `Edit` action to `View` to better reflect the content of the
modal that appears when clicked...
## 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] API calls work with loading states in Agent Info Step
- [x] Image aspect ratios are consistent across all components
- [x] Form validation works through zod schema only
### For configuration changes:
None
## Summary
This adds 10 new LLM's to the platform!
I have added the model names, metadata like max input and output and the
price for each model!
- GROK_4
- KIMI_K2
- QWEN3_235B_A22B_THINKING
- QWEN3_CODER
- GEMINI_2_5_FLASH
- GEMINI_2_0_FLASH
- GEMINI_2_5_FLASH_LITE_PREVIEW
- GEMINI_2_0_FLASH_LITE
- DEEPSEEK_R1_0528
- GPT41_MINI
## 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] Test and verify all the models work!
## Changes 🏗️
### Why these changes
We have a high-priority bug where the publish agent modal wouldn't open
when clicking `Edit` on the Dashboard Creator page table. The create
form was also buggy.
When looking into the code, I noticed it was pretty messy. I went ahead
and refactored it:
- [x] separation of concerns ( _split render / hook logic_ )
- [x] split into sub-components ( `PublishAgentModal/components` )
- [x] colocated state ( moved state to the modal steps rather than
having everything top-level )
- [x] used the new Design System components
Overall, we end up with a cleaner and stable experience ✨
### E2E tests
I also added E2E tests 🤖 to make sure we catch regressions in the future
in this modal. For now, it tests the first 2 steps. It does not do image
upload and publish as that wasn't working locally ( _might iterate on
that later_ )
### Step 1 – Select Agent
<img width="1161" height="859" alt="Screenshot 2025-07-29 at 16 12 46"
src="https://github.com/user-attachments/assets/a4949fb0-1a44-4926-a374-51eefadef063"
/>
### Step 2 – Agent Info Form
<img width="1061" height="804" alt="Screenshot 2025-07-29 at 16 03 11"
src="https://github.com/user-attachments/assets/b9a45bda-18ea-4844-b52c-db499f45193e"
/>
### Step 3 – Agent Review
<img width="1480" height="867" alt="Screenshot 2025-07-29 at 16 11 07"
src="https://github.com/user-attachments/assets/248bdf58-886d-43f3-a37a-35fd1a83e566"
/>
## 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] open the modal through the Account menu → ( `Publish Agent` )
- [x] complete the form and check validation errors
- [x] add images and generate image
- [x] publish the agent
- [x] the agent shows up on the table
- [x] Open an agent under review in the table ( _click `Edit` on the
actions_ )
- [x] it opens the modal on the 3rd step ( _review step_ )
### For configuration changes:
None
---------
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Co-authored-by: Abhimanyu Yadav <122007096+Abhi1992002@users.noreply.github.com>
I have gone through and tested all 59 llm's on the platform, found 5
where deprecated/aren't available any more so i removed them.
I made a agent with 59 llm call blocks, set each llm and ran it, i got
several returned replies saying that models where deprecated so i
removed those models.
<img width="1804" height="887" alt="image"
src="https://github.com/user-attachments/assets/907776e1-b491-465d-8219-e86c98559e41"
/>
Models removed:
- O1_PREVIEW
- MIXTRAL_8X7B
- EVA_QWEN_2_5_32B
- PERPLEXITY_LLAMA_3_1_SONAR_LARGE_128K_ONLINE
- QWEN_QWQ_32B_PREVIEW
### Changes 🏗️
This PR adds Firecrawl integration to AutoGPT, providing powerful web
scraping and data extraction capabilities:
**New Blocks Added:**
⚠️ All these blocks are synchronous so take a while to finish, this
allows a simpler agent workflow
- **Firecrawl Scrape Block**: Scrapes single web pages with various
output formats (Markdown, HTML, JSON, screenshots)
- **Firecrawl Crawl Block**: Crawls entire websites following links with
customizable depth and filters
- **Firecrawl Extract Block**: Extracts structured data from web pages
using AI-powered prompts
- **Firecrawl Map Block**: Maps website structure and returns a list of
all discovered URLs
- **Firecrawl Search Block**: Searches Google and scrapes the results
**Key Features:**
- Advanced anti-blocking technology to bypass scraping protections
- Multiple output formats including Markdown, HTML, JSON, and
screenshots
- AI-powered data extraction with custom prompts and schemas
- Configurable crawling depth and URL filtering
- Built-in caching and rate limiting
- Google search integration for discovering relevant content
**Use Cases:**
- Web data extraction for research and analysis
- Content monitoring and change tracking
- Competitive intelligence gathering
- SEO analysis and website mapping
- Automated data collection workflows
### 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] Verified all Firecrawl blocks appear in the UI
- [x] Tested scraping various websites with different formats
- [x] Tested crawling with depth limits and URL filters
- [x] Tested data extraction with custom prompts
- [x] Verified error handling for invalid URLs and API failures
- [x] Tested authentication with Firecrawl API key
- [x] Confirmed proper rate limiting and caching behavior
<img width="1025" height="1027" alt="Screenshot 2025-07-30 at 15 20 28"
src="https://github.com/user-attachments/assets/7b94d3cf-7a0e-4d09-a9c5-24c4e8a3b660"
/>
# Example Agent
[FC
Testing_v12.json](https://github.com/user-attachments/files/21510608/FC.Testing_v12.json)
## Summary
- Enabled the Instagram posting block that was previously disabled
- The block provides comprehensive Instagram-specific posting options
including stories, reels, posts, user tagging, and location tagging
- Improved parameter types and validation for better user experience
## Changes 🏗️
- Removed `disabled=True` from Instagram posting block to enable
functionality
- Updated parameter types from required to optional with proper None
defaults for better flexibility
- Added validation for Instagram reel options to ensure all required
fields are provided together
- Improved user tag validation with better error messages
- Added support for:
- Instagram Stories (24-hour expiration)
- Instagram Reels with audio, thumbnails, and feed sharing options
- Alt text for accessibility
- Location tagging via Facebook Page ID
- User tagging with coordinate support
- Collaborator tagging
- Auto-resize functionality
## 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] Verified Instagram block is now available in the block list
## Summary
- Enabled the YouTube posting block that was previously disabled
- The block provides comprehensive YouTube-specific posting options
including titles, visibility settings, thumbnails, playlists, tags, and
more
## Changes 🏗️
- Removed `disabled=True` from YouTube posting block to enable
functionality
- Added full YouTube API integration with all supported options:
- Video title and description
- Visibility settings (private/public/unlisted)
- Thumbnail support
- Playlist management
- Video tags and categories
- YouTube Shorts support
- Subtitle/caption support
- Country-based targeting
- Synthetic media disclosure
## 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] Verified YouTube block is now available in the block list
https://github.com/user-attachments/assets/d4459f15-fe57-47bf-8459-f06f1af45ad6
<img width="374" height="593" alt="Screenshot 2025-07-31 at 11 26 29"
src="https://github.com/user-attachments/assets/4dcf30dd-439c-4a44-b56a-640832d6c550"
/>
## Changes 🏗️
My previous PR,
https://github.com/Significant-Gravitas/AutoGPT/pull/10480, didn't fully
resolve the issue of broken links sometimes appearing for some runs in
the Agent Activity dropdown.
- Fixed the logic ( verified with a deployment in dev... )
- Simplified logic, making less API calls
- If we have an execution without a clear agent ID, we display it but
don't link to it
- Re-generated API types ( _had to update call in dashboard agents
because of it_ )
## 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 agents
- [x] Runs appear correctly in the activity dropdown without broken
links
### For configuration changes:
None
- Resolves#10489
### Changes 🏗️
- Fix Google OAuth token revocation
- Fix credentials object conversion in `GoogleOAuthHandler`
### 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] Google OAuth flow still works
- [x] Deleting Google OAuth credentials works, token revocation doesn't
error
### Changes 🏗️
This PR adds a new Wolfram Alpha block that integrates with Wolfram's
LLM API endpoint:
- **Ask Wolfram Block**: Allows users to ask questions to Wolfram Alpha
and get structured answers
- **API Integration**: Implements the Wolfram LLM API endpoint
(`/api/v1/llm-api`) for natural language queries
- **Simple Authentication**: Uses App ID based authentication via API
key credentials
- **Error Handling**: Proper error handling for API failures with
descriptive error messages
The block enables users to leverage Wolfram Alpha's computational
knowledge engine for:
- Mathematical calculations and explanations
- Scientific data and facts
- Unit conversions
- Historical information
- And many other knowledge-based queries
### 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] Verified the block appears in the UI and can be added to workflows
- [x] Tested API authentication with valid Wolfram App ID
- [x] Tested various query types (math, science, general knowledge)
- [x] Verified error handling for invalid credentials
- [x] Confirmed proper response formatting
**Configuration changes:**
- Users need to add `WOLFRAM_APP_ID` to their environment variables or
provide it through the UI credentials field
### Changes 🏗️
This PR adds Airtable integration to AutoGPT with the following blocks:
- **List Bases Block**: Lists all Airtable bases accessible to the
authenticated user
- **Create Base Block**: Creates new Airtable bases with specified
workspace and name
<img width="1294" height="879" alt="Screenshot 2025-07-30 at 11 03 43"
src="https://github.com/user-attachments/assets/0729e2e8-b254-4ed6-9481-1c87a09fb1c8"
/>
### Checklist 📋
#### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] Tested create base block
- [x] Tested list base block
Need: The Gmail integration had several parsing issues that were causing
data loss and
workflow incompatibilities:
1. Email recipient parsing only captured the first recipient, losing
CC/BCC and multiple TO
recipients
2. Email body parsing was inconsistent between blocks, sometimes showing
"This email does
not contain a readable body" for valid emails
3. Type mismatches between blocks caused serialization issues when
connecting them in
workflows (lists being converted to string representations like
"[\"email@example.com\"]")
# Changes 🏗️
1. Enhanced Email Model:
- Added cc and bcc fields to capture all recipients
- Changed to field from string to list for consistency
- Now captures all recipients instead of just the first one
2. Improved Email Parsing:
- Updated GmailReadBlock and GmailGetThreadBlock to parse all recipients
using
getaddresses()
- Unified email body parsing logic across blocks with robust multipart
handling
- Added support for HTML to plain text conversion
- Fixed handling of emails with attachments as body content
3. Fixed Block Compatibility:
- Updated GmailSendBlock and GmailCreateDraftBlock to accept lists for
recipient fields
- Added validation to ensure at least one recipient is provided
- All blocks now consistently use lists for recipient fields, preventing
serialization
issues
4. Updated Test Data:
- Modified all test inputs/outputs to use the new list format for
recipients
- Ensures tests reflect the new data structure
# Checklist 📋
For code changes:
- I have clearly listed my changes in the PR description
- I have made a test plan
- I have tested my changes according to the test plan:
- Run existing Gmail block unit tests with poetry run test
- Create a workflow that reads emails with multiple recipients and
verify all TO, CC, BCC
recipients are captured
- Test email body parsing with plain text, HTML, and multipart emails
- Connect GmailReadBlock → GmailSendBlock in a workflow and verify
recipient data flows
correctly
- Connect GmailReplyBlock → GmailSendBlock and verify no serialization
errors occur
- Test sending emails with multiple recipients via GmailSendBlock
- Test creating drafts with multiple recipients via
GmailCreateDraftBlock
- Verify backwards compatibility by testing with single recipient
strings (should now
require lists)
- Create from scratch and execute an agent with at least 3 blocks
- Import an agent from file upload, and confirm it executes correctly
- Upload agent to marketplace
- Import an agent from marketplace and confirm it executes correctly
- Edit an agent from monitor, and confirm it executes correctly
# Breaking Change Note: The to field in GmailSendBlock and
GmailCreateDraftBlock now requires
a list instead of accepting both string and list. Existing workflows
using strings will
need to be updated to use lists (e.g., ["email@example.com"] instead of
"email@example.com").
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
## Changes 🏗️
Fix the issue where sometimes the agent activity would show a link to
agent runs that are not available in the library. So only show runs that
can be verified in the library. Improve the display of the agent name as
well 🤔
## 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 agents
- [x] There are no empty links on the activity dropdown
- [x] Agent names look good 💅🏽
### For configuration changes:
None
### Changes 🏗️
This PR fixes an issue where LLM blocks (particularly
AITextSummarizerBlock) were not properly tracking `llm_call_count` in
their execution statistics, despite correctly tracking token counts.
**Root Cause**: The `finally` block in
`AIStructuredResponseGeneratorBlock.run()` that sets `llm_call_count`
was executing after the generator returned, meaning the stats weren't
available when `merge_llm_stats()` was called by dependent blocks.
**Changes made**:
- **Fixed stats tracking timing**: Moved `llm_call_count` and
`llm_retry_count` tracking to execute before successful return
statements in `AIStructuredResponseGeneratorBlock.run()`
- **Removed problematic finally block**: Eliminated the finally block
that was setting stats after function return
- **Added comprehensive tests**: Created extensive test suite for LLM
stats tracking across all AI blocks
- **Added SmartDecisionMaker stats tracking**: Fixed missing LLM stats
tracking in SmartDecisionMakerBlock
- **Fixed type errors**: Added appropriate type ignore comments for test
mock objects
**Files affected**:
- `backend/blocks/llm.py`: Fixed stats tracking timing in
AIStructuredResponseGeneratorBlock
- `backend/blocks/smart_decision_maker.py`: Added missing LLM stats
tracking
- `backend/blocks/test/test_llm.py`: Added comprehensive LLM stats
tracking tests
- `backend/blocks/test/test_smart_decision_maker.py`: Added LLM stats
tracking test and fixed circular imports
### 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] Created comprehensive unit tests for all LLM blocks stats tracking
- [x] Verified AITextSummarizerBlock now correctly tracks llm_call_count
(was 0, now shows actual call count)
- [x] Verified AIStructuredResponseGeneratorBlock properly tracks stats
with retries
- [x] Verified SmartDecisionMakerBlock now tracks LLM usage stats
- [x] Verified all existing tests still pass
- [x] Ran `poetry run format` to ensure code formatting
- [x] All 11 LLM and SmartDecisionMaker tests pass
#### For configuration changes:
- [x] `.env.example` is updated or already compatible with my changes
- [x] `docker-compose.yml` is updated or already compatible with my
changes
- [x] I have included a list of my configuration changes in the PR
description (under **Changes**)
**Note**: No configuration changes were needed for this fix.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-authored-by: Claude <noreply@anthropic.com>
HTTP requests can fail when the DNS is messed up. Sometimes this kind of
issue requires a client reset.
### Changes 🏗️
Introduce HTTP client refresh on repeated error
### 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 run, added tests
To allow for a simpler dev experience, the new SDK now auto discovers
providers and registers them. However, the OAuth system was still
requiring these credentials to be hardcoded in the settings object.
This PR changes that to verify the env var is present during
registration and then allows the OAuth system to load them from the env.
### Changes 🏗️
- **OAuth Registration**: Modified `ProviderBuilder.with_oauth(..)` to
check OAuth env vars exist during registration
- **OAuth Loading**: Updated OAuth system to load credentials from env
vars if not using secrets
- **Block Filtering**: Added `is_block_auth_configured()` function to
check if a block has valid authorization options configured at runtime
- **Test Updates**: Fixed failing SDK registry tests to properly mock
environment variables for OAuth registration
### 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] Verified OAuth system checks that env vars exist during provider
registration
- [x] Confirmed OAuth system can use env vars directly without requiring
hardcoded secrets
- [x] Tested that blocks with unconfigured OAuth providers are filtered
out
- [x] All SDK registry tests pass with proper env var mocking
#### For configuration changes:
- [x] `.env.example` is updated or already compatible with my changes
- [x] `docker-compose.yml` is updated or already compatible with my
changes
- [x] I have included a list of my configuration changes in the PR
description (under **Changes**)
- OAuth providers now require their client ID and secret env vars to be
set for registration
- No changes required to `.env.example` or `docker-compose.yml`
---------
Co-authored-by: Reinier van der Leer <pwuts@agpt.co>
## Changes 🏗️
There is a bug where the agent activity dropdown bubble only shows up to
`6` even if there are `50 running agents`. We only display the last 6
runs in the dropdown, but the bubble badge count should show the correct
all running agents.
On top of that we added the option to search runs by agent name when you
have more than 6 recent runs:
https://github.com/user-attachments/assets/931e3db7-5715-48d1-b4df-22490fae9de0
- Also make the dropdown items a link ( `a` ) so that you can command
click them to open runs in new tabs.
- Keep up to `400` executions on the state ( worse case load test )
- Each execution object is relatively small (ID, status, timestamps,
agent info)
- 400 objects × ~`1KB` each = negligible memory footprint `400kb`
- Always display running agents at the top
- Only display runs from the last week on the dropdown
- the agent library page contains the historical runs, this is just to
show the recent ones
### Code changes
- **Added count tracking**
- the `NotificationState` interface now includes separate count fields
(`activeCount`, `recentCompletionsCount`, `recentFailuresCount`) to
track the actual numbers independent of display limits.
- **Dual array system:**
- the `categorizeExecutions` function now creates:
- unlimited arrays for counting all executions
- limited arrays (sliced to 6 items) for dropdown display
- Updated all helper functions to properly maintain both the display
arrays and the count fields.
- Component uses actual counts
- `<AgentActivityDropdown />` component now uses `activeCount` for the
badge and hover hint instead of `activeExecutions.length`
## 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] Login and navigate to library or build
- [x] Start running agents like there is no tomorrow
- [x] The badge shows the correct agent execution count ( .i.e 10 )
- [x] The dropdown only displays the 6 most recent
- [x] You can command click on the runs and they open in new tabs
### For configuration changes
None
The heartbeat mechanism doesn't seem to work at the moment.
### Changes 🏗️
Revert the RabbitMQ consumer heartbeat mechanism
### 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] Run agents
This PR adds WordPress integration to AutoGPT platform, enabling users
to create posts on WordPress.com and Jetpack-enabled sites.
### Changes 🏗️
**OAuth Implementation:**
- Added WordPress OAuth2 handler (`_oauth.py`) supporting both single
blog and global access tokens
- Implemented OAuth flow without PKCE (as WordPress doesn't require it)
- Added token validation endpoint support
- Server-side tokens don't expire, eliminating the need for refresh in
most cases
**API Integration:**
- Created WordPress API client (`_api.py`) with Pydantic models for type
safety
- Implemented `create_post` function with full support for WordPress
post features
- Added helper functions for token validation and generic API requests
- Fixed response models to handle WordPress API's mixed data types
**WordPress Block:**
- Created `WordPressCreatePostBlock` in `blog.py` with minimal
user-facing options
- Exposed fields: site, title, content, excerpt, slug, author,
categories, tags, featured_image, media_urls
- Posts are published immediately by default
- Integrated with platform's OAuth credential system
### 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] OAuth URL generation works correctly for single blog and global
access
- [x] Token exchange and validation functions handle WordPress API
responses
- [x] Create post block properly transforms input data to API format
- [x] Response models handle mixed data types from WordPress API
The WordPress OAuth provider needs to be configured with client ID and
secret from WordPress.com application settings.
<!-- Clearly explain the need for these changes: -->
I'm working with these blocks and found some much needed improvements
### Changes 🏗️
- Outputs labels from emails
- Types the outputs
- reorder the yielding of the smart decision blokc
<!-- Concisely describe all of the changes made in this pull request:
-->
### 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] Build a test agent with labels, sending, reading emails + smart
decision maker
When we run multiple instances of the executor, some of the executors
can oversubscribe the messages and end up queuing the agent execution
request instead of letting another executor handle the job. This change
solves the problem.
### Changes 🏗️
* Reject execution request when the executor is full.
* Improve `active_graph_runs` tracking for better horizontal scaling
heuristics.
### 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 graph execution & CI
### Changes 🏗️
1. Json columns have to be json serializable, but sometimes the data is
not, so `SafeJson` is introduced to make sure that the data being loaded
can be string serialized and back before persisting into the database.
2. Locks & transactions seem to be used in the case where it's not
needed, this reduces database & redis performance.
### 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] CI tests
