* Switch to new "RunAgent" method name.
* Try to disable false positive naming warning.
* Add comment about disabled warnings.
* Rename `RunAgent` to just `Run`.
* Update CHANGELOG.
* Cosmos DB UT Fast Skip (Non-Configured Local envs) + Long running UT skip in pipeline when no CosmosDB changes happened
* Force a CosmosDB source code change to trigger the pipeline
* Address possible string boolean mismatch
* Add debug
* Enabling emulator always when running IT
* Update to latest Azure.AI.*, OpenAI, and M.E.AI*
Absorb breaking changes in Responses surface area
* Update dotnet/samples/AgentWebChat/AgentWebChat.AgentHost/Utilities/ChatClientExtensions.cs
* Update dotnet/samples/AgentWebChat/AgentWebChat.AgentHost/Utilities/ChatClientExtensions.cs
* Update dotnet/samples/AgentWebChat/AgentWebChat.AgentHost/Utilities/ChatClientExtensions.cs
* Update dotnet/samples/GettingStarted/AgentWithOpenAI/Agent_OpenAI_Step04_CreateFromOpenAIResponseClient/Program.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Using patch to remove the model is necessary, updated the response client to actually use the the ForAgent
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Roger Barreto <19890735+rogerbarreto@users.noreply.github.com>
## Summary
Enhanced `HandoffBuilder._apply_auto_tools` to use the target agent's
description when creating handoff tools, providing more informative tool
descriptions for LLMs.
## Changes
- Modified `_apply_auto_tools` to extract `description` from
`AgentExecutor._agent` when available
- Updated iteration to use `.items()` for more efficient dict traversal
- Handoff tools now use agent descriptions instead of generic placeholders
## Example
Before: "Handoff to the refund_agent agent."
After: "You handle refund requests. Ask for order details and process refunds."
## Testing
- All handoff tests pass (20/20)
- No breaking changes to existing API
Fixes#2713
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
* Added an example of using kwargs in ai_function
* Added thread object to ai_function kwargs
* Updated docs
* Small fix
* Added thread parameter filtering
* adds support for labels in edges, fixes rendering of labels in dot and mermaid, adds rendering of labels in edges
* Update dotnet/src/Microsoft.Agents.AI.Workflows/Visualization/WorkflowVisualizer.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* escaping edge labels, adding tests for labels containing strange characters that would break the diagram and enabling the previous signature so the API has backwards compatibility.
* Unify label in EdgeData
* Edge API adjustments, removed useless "sanitizer"
* fixed test
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Jacob Alber <jaalber@microsoft.com>
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
* added more complete parsing for mcp tool arguments
* fixed mypy
* added nonlocal model counter, and some fixes
* fixes in naming logic
* extracted json parsing function, added parametrized test and checked coverage
* Add factory pattern to sequential orchestration builder
* Use temp list to avoid override
* Add sample and some other fixes
* Fix comments
* Small fix
* Update readme
* Support HITL for orchestration patterns
* Cleanup around naming
* Fix typing issues
* Clean up
* Naming clean up
* Updates to HITL to make it cleaner
* Rename human input hook to orchestration request info
* Clean up per PR feedback
- Replace OPENAI_APIKEY with OPENAI_API_KEY across all samples
- Replace AZURE_FOUNDRY_OPENAI_APIKEY with AZURE_FOUNDRY_OPENAI_API_KEY
- Ensures consistency with OpenAI's standard naming convention
- Applies to .NET and Python samples
Fixes#1001
Co-authored-by: Alexander Zarei <alzarei@users.noreply.github.com>
* .NET: [Durable Agents] Update CHANGELOG with release notes for past releases
Backfills the CHANGELOG.md files with the last several updates.
* Update dotnet/src/Microsoft.Agents.AI.DurableTask/CHANGELOG.md
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update the Azure Functions changelog and add GHCP changelog instructions for these projects
* Tweak instructions
* Remove the timestamp requirement
* Rename instructions file
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* change namespaces for agents and extension methods of the Microsoft.Agents.AI.OpenAI package
* remove unnecessary namespace
* remove unused namespaces
* fix compilation issues and rrolled back removed run methods
* sort usings
* add extension methods for AIAgent to work with OpenAI Responses primitives
* Move OpenAIChatClientAgent and OpenAIResponseClientAgent to samples
* sort usings
* sort usings
* Title: Fix WorkflowFailedEvent error extraction to use details instead of error Body:
Summary
Fixed WorkflowFailedEvent mapping to extract error message from details.message instead of non-existent error attribute
Added support for including details.extra context in error messages when present
Problem
The WorkflowFailedEvent handler in _mapper.py was reading event.error, but WorkflowFailedEvent uses a details attribute (of type WorkflowErrorDetails), not error. This caused all workflow failures to display "Unknown error" in the UI instead of the actual error message.
Fix
Updated the handler to match the pattern already used by ExecutorFailedEvent:
Read from event.details instead of event.error
Extract details.message for the error text
Include details.extra context when available
* improve error handling consistency
* Provide way for HITL with magentic
* support tool call approvals and hitl stall replan
* human plan intervention sample
* Clean up
* Improve loging
* updates
* Add type annotations for AgentRunEvent and AgentRunUpdateEvent data attributes
* Fix unnecessary cast after typing improvement
* Mcp pkg update introduced type change. Fix it.
* update opentelemetry deps to 1.39.0 and use LogRecordExporter for type compatibility
* Added unit test for RetrieveConversationMessagesExecutor
* Unit test for declarative object model AddConversationMessageExecutor
* Remove unnecessary test.
* Fix test.
* show app version in devui .NET: Python: Improved Versioning for DevUI
Fixes#2059
* feat: Add multimodal input support for workflows and refactor chat input
This PR adds support for multimodal content (images, files) in workflow
inputs and refactors the chat input into a reusable component.
## Multimodal Workflow Support
- Add `isChatMessageSchema()` to detect ChatMessage input schemas
- Update `RunWorkflowButton` to use `ChatMessageInput` for ChatMessage workflows
- Wrap multimodal content in OpenAI message format for backend processing
- Add `_is_openai_multimodal_format()` to detect OpenAI ResponseInputParam
- Update `_parse_workflow_input()` to route multimodal input through
existing `_convert_input_to_chat_message()` converter
## Reusable ChatMessageInput Component
- Extract chat input logic from agent-view into `ChatMessageInput` component
- Support file upload, drag & drop, paste handling, and attachments
- Add `useDragDrop` hook for parent-level drag handling with full-area
drop zones
- Refactor agent-view to use the new shared component
## Other Improvements
- Add `isStreaming` prop to executor nodes for animation control
- Clean up unused imports and state variables in agent-view
- Add tests for multimodal workflow input handling
Fixes workflow input not receiving images when using AgentExecutor nodes.
* add self loop edge, fix#2470
* fix test
* Fix AG-UI forwardedProps JSON property name
The RunAgentInput.ForwardedProperties property was using the wrong JSON property
name 'forwardedProperties' instead of 'forwardedProps' per the AG-UI protocol
specification.
This fix:
- Changes JsonPropertyName from 'forwardedProperties' to 'forwardedProps'
- Adds comprehensive integration tests for forwarded properties
Fixes#2468
* Fix formatting
* update .net sdk and runtime images and update nuget packages to the latests versions
* align version of roslyn analyzers
* unlock roslyn analyzer packages
* Add sample to show handoff as agent with HITL
* Update uv.lock with latest pkg versions. Fix lint error.
* Upgrade grpcio to 1.76.0
* Handle grpcio versions
* Case insensitive compare for declarative
* Update the declarative agent samples
* Add the Microsoft.Agents.AI.Declarative project
* Make the package non packable
* Use the RecalcEngine when creating the ChatOptions
* Ignore VSTHRD200
* Add geting started samples
* Address code review feedback
* prevent stremed updates loss when resuming streaming with non-agent managed store or/and context provider
* Update dotnet/src/Microsoft.Agents.AI/ChatClient/ChatClientAgent.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* throw not supported exception instead invalid operation
* Update dotnet/src/Microsoft.Agents.AI/ChatClient/ChatClientAgent.cs
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
* Update dotnet/src/Microsoft.Agents.AI/ChatClient/ChatClientAgent.cs
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
* use conversation id to check if chat history is managed by agent service or not
* extract background responses tests into a separate file
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
* docs: Update Python orchestration documentation
Remove outdated 'coming soon' statements for GroupChat, Sequential,
and Concurrent orchestrations in core package README and transparency FAQ.
Add links to existing samples in python/samples/getting_started/workflows/orchestration/.
Note: python/README.md was already updated in PR #1914 (2499262f).
Fixes documentation inconsistency found in Issue #1899.
* Address review feedback: make TRANSPARENCY_FAQ language-neutral
Remove Python-specific sample links from TRANSPARENCY_FAQ.md as it should
pertain to all MAF languages (Python and .NET), not strictly Python.
The python/packages/core/README.md retains the specific sample links as
that is Python-specific documentation.
* Update Learn documentation link in README
---------
Co-authored-by: kishikawa-hayato <84244732+HerBest-max@users.noreply.github.com>
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
* Propagate orchestration ID (if any).
* Add integration test for orchestration ID in entity state.
* Update schema.
* Fixup formatting issues.
* Fix more formatting issues.
* draft commit
* Added Cosmos agent thread and tests
* revert unnecessary changes and fix tests
* add multi-tenant support with hierarchical partition keys (and tests).
* enhance transactional batch
* address review comments
* Address PR review comments from @westey-m
* Merge upstream/main - resolve slnx conflicts
* use param validation helpers
* Replace useManagedIdentity boolean with TokenCredential parameter
* Remove redundant suppressions and fix tests
* Rename project from Microsoft.Agents.AI.Abstractions.CosmosNoSql to Microsoft.Agents.AI.CosmosNoSql
* Refactor constructors to use chaining pattern
* Reorder deserialization constructor parameters for consistency
* Remove database/container IDs from serialized state
* Remove auto-generation of MessageId
* Optimize AddMessagesAsync to avoid enumeration when possible
* Add MaxMessagesToRetrieve to limit context window
* Make Role nullable instead of defaulting
* Fix net472 build without rebasing 19 commits
* Add Cosmos DB emulator to CI workflow
* Fix Cosmos DB emulator tests: use Skip.If instead of Assert.Fail and start emulator before unit tests
* Replace Skip.If() with conditional return to fix compilation
* Use env var to skip Cosmos tests on non-Windows CI
* Add Xunit.SkippableFact package to properly skip Cosmos tests on Linux
* Change [Fact] to [SkippableFact] for proper test skipping behavior
* Remove stale Microsoft.Agents.AI.Abstractions.CosmosNoSql directory
* Fix code formatting: add braces, this. qualifications, and final newlines
* Fix file encoding to UTF-8 with BOM, fix import ordering, and remove unnecessary using directives
* Convert backing fields to auto-properties and remove Azure.Identity using directive
* Fix CosmosChatMessageStore.cs encoding back to UTF-8 with BOM
* Fix test file formatting: indentation, encoding, imports, this. qualifications, naming conventions, and simplify new expressions
* Fix const field naming violations: Remove s_ prefix from const fields and add this. qualification to Dispose call
* Add local .editorconfig for Cosmos DB tests to suppress IDE0005 false positives from multi-targeting
* Fix IDE1006 naming violations: Rename TestDatabaseId to s_testDatabaseId and add final newlines
* Address PR review comments
Address Wesley's review comments:
- Remove Cosmos DB package references from core projects
- Delete duplicate test files from old package structure
- Remove redundant parameter validation from extension methods
Address Kiran's review comments:
- Remove redundant 429 retry logic (SDK handles automatically)
- Add explicit RequestEntityTooLarge error handling
- Remove dead code in GetMessageCountAsync
- Add defensive partition key validation comments
* Fix IDE0001 formatting error in AgentProviderExtensions.cs. Use type alias to resolve namespace conflict between Azure.AI.Agents.Persistent.RunStatus and Microsoft.Agents.AI.Workflows.RunStatus. This eliminates the need for global:: qualifier which triggered the formatter warning.
* Update package versions for Aspire 13.0.0 compatibility
* Fix TargetFrameworks in Cosmos DB projects
- Replace with which is defined in Directory.Build.props
- Fix package reference from System.Linq.Async to System.Linq.AsyncEnumerable to match Directory.Packages.props
* Remove redundant counter, add partition key validation, use factory pattern for deserialization
* Standardize orchestration outputs as list of chatmessage. Add chat options to group chat prompt manager
* refactor group chat
* Improve group chat manager
* README Update
* Cleanup
* Add comment
* More cleanup
* Standardize termination condition for group chat
* Improvements on termination logic
* Fix tests
* Fix new line
* PR feedback
* Update ChatKit based on OpenAI type change
* Raise error if response format is not expected type
* Only one starting executor required. Add tests.
* Add magentic start executor test
* WIP
* WIP
* Simple call working
* Update Thinking sample
* Non-Streaming Function calling working
* Update Anthropic Impl
* Public Preps
* UT + IT working
* Update documentation + samples
* Update variable
* Revert nuget.config
* Add IT for BetaService implementation
* Remove polyfill + enable IT to run for netstandard 2.0
* Skipping Anthropic IT's for manual execution and avoid pipeline execution
* Fix compilation error
* Address error in UT
* Apply suggestions from code review
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix warning
* Net 10 update
* Update for NET 10, remove Anthropic.Foundry due to vulnerability
* Final missing adjustments for NET 10
* Address PR comments
* Remove unused code
* Address feedback
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* make tool call view optional in devui + other link fixes
* fix#2310, ensure correct port is shown in command
* fix dialog bug
* ensure executor ids are tracked per items, fix bug where data from concurrent executors where not seperated properly fix#2351
* fix: Enable multi-round human-in-the-loop (HIL) in DevUI workflows
- Backend: Enrich RequestInfoEvents with response schemas in send_responses_streaming path
- Frontend: Replace old HIL requests with new ones instead of accumulating them
- Frontend: Fix HIL response state management to prevent sending stale request responses
This allows workflows to properly handle sequential HIL requests, showing only the
current request to users and progressing through multiple input rounds correctly.
fixes#2334
* fix bug to ensure in memory entities cannot be reloaded in ui
* Upgrade to .NET 10
- Require .NET 10 SDK
- Include net10.0 assets in all assemblies
- Move net9.0-only targets to net10.0
- Update LangVersion to latest
- Remove complicated distinctions between debug target TFMs and release target TFMs
- Remove unnecessary package dependencies when built into netcoreapp
- Clean up some ifdefs
- Clean up some analyzer warnings
* Fix CI
* feat(mcp): add full _meta field support for CallToolResult objects
- Extract and preserve complete _meta field from MCP CallToolResult responses
- Merge metadata into additional_properties of converted content items
- Handle isError field for proper error state integration
- Support arbitrary metadata like token usage, costs, and performance metrics
- Maintain backward compatibility with existing tool execution workflows
- Add comprehensive test coverage for all metadata scenarios including edge cases
- Update documentation with metadata handling examples and patterns
Fixes protocol compliance violation where _meta fields were being dropped,
enables proper monitoring and cost tracking of MCP tool usage.
* Update python/packages/core/agent_framework/_mcp.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Clarify MCP _meta field test to use generic example metadata
- Updated test_mcp_call_tool_result_with_meta_arbitrary_data to use arbitrary metadata fields
- Added comments to emphasize that _meta structure is server-specific and not standardized
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Python: Fix pyright errors and move search provider to core (#1546)
* address pablo coments
* update azure ai search pypi version to latest prev
* init update
* Fix MyPy type annotation errors in search provider
- Add type annotation to DEFAULT_CONTEXT_PROMPT
- Add type annotation to vectorizable_fields
- Add union type annotation to vector_queries
* Fix DEFAULT_CONTEXT_PROMPT MyPy error and update test
- Rename DEFAULT_CONTEXT_PROMPT to _DEFAULT_SEARCH_CONTEXT_PROMPT to avoid conflict with base class Final variable
- Update test to use new constant name
- All core package tests passing (1123 passed)
* Python: Move Azure AI Search to separate package per PR feedback
Addresses reviewer feedback from PR #1546 by isolating the beta dependency
(azure-search-documents==11.7.0b2) into a new agent-framework-aisearch package.
Changes:
- Created new agent-framework-aisearch package with complete structure
- Moved AzureAISearchContextProvider from core to aisearch package
- Added AzureAISearchSettings class for environment variable auto-loading
- Added support for direct API key string (auto-converts to AzureKeyCredential)
- Added azure_openai_api_key parameter for Knowledge Base authentication
- Updated embedding_function type to Callable[[str], Awaitable[list[float]]]
- Moved Role import to top-level imports
- Maintained lazy loading through agent_framework.azure module
- Removed beta dependency from core package
- Updated all tests to use new package location
- All quality checks pass: ruff format/lint, pyright, mypy (0 errors)
- All 21 unit tests pass with 59% coverage
Semantic search mode verified working with both API key and managed identity authentication.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Python: Clarify top_k parameter only applies to semantic mode
Updated documentation to clarify that the top_k parameter only affects
semantic search mode. In agentic mode, the server-side Knowledge Base
determines retrieval based on query complexity and reasoning effort.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Python: Add Knowledge Base output mode and retrieval reasoning effort parameters
Added support for configurable Knowledge Base behavior in agentic mode:
- knowledge_base_output_mode: "extractive_data" (default) or "answer_synthesis"
Some knowledge sources require answer_synthesis mode for proper functionality.
- retrieval_reasoning_effort: "minimal" (default), "medium", or "low"
Controls query planning complexity and multi-hop reasoning depth.
These parameters give users fine-grained control over Knowledge Base behavior
and enable support for knowledge sources that require answer synthesis.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* effort and outputmode query params
* Address PR review feedback for Azure AI Search context provider
* comments eduward
* ed latest comments
---------
Co-authored-by: Farzad Sunavala <farzad.sunavala.enovate.ai>
Co-authored-by: farzad528 <farzad528@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
* Add unit tests for create conversation executor
* Update indentation and comment typo.
* Added unit tests for declarative executor SetMultipleVariablesExecutor
* Updated comments and syntactic sugar
* Add unit test for declarative executor RetrieveConversationMessageExecutor
* Removed irrelevant code statements
* Updated based on copilot feedback.
Fixes#2219
Adds default=str to json.dumps() calls to handle non-JSON-serializable
types like datetime objects in tool function results.
Co-authored-by: kishikawa-hayato <84244732+HerBest-max@users.noreply.github.com>
* Deep copy the agent chat options to avoid mutations
* avoiding _thread.RLock pickling errors
2025-11-20 08:06:14 +00:00
Evan MattsonGitHubCopilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
* Potential fix for code scanning alert no. 18: Information exposure through an exception
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
* Fix test
---------
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
* First working version
* Simplify the implementations
* Remove unused env var
* Update Python syntax
* Address feedbacks
* Fix a typo
* Update names as review suggestions
* Citation for self-reflection
* Move to independent folder
* Update python/samples/getting_started/evaluation/azure_ai_foundry/evaluation/README.md
Co-authored-by: Eduard van Valkenburg <eavanvalkenburg@users.noreply.github.com>
* Updated from parquet to JSONL and hide the default environment variables
* As review feedback, remove the purpose of using `run_self_reflection_batch` as a library, only use it as sample code
* Update python/samples/getting_started/evaluation/azure_ai_foundry/evaluation/self_reflection.py
Co-authored-by: Eduard van Valkenburg <eavanvalkenburg@users.noreply.github.com>
---------
Co-authored-by: Eduard van Valkenburg <eavanvalkenburg@users.noreply.github.com>
* first work on declarative
* initial version of the declarative support
* fix tests and mypy
* fix parameters of functiontool
* slight logic improvement
* remove path until merge
* updates from comments
* create dispatcher and spec type, json_schema method
* fix mypy, skipping model
* updated lock
* fixed declarative tests and renamed some other test files
* refined loader
* updated lock
* fix mypy
* added readme to samples folder
* fixes from review
* undid test file rename
* fix: resolve string annotations in FunctionExecutor
Enhance type hint validation in FunctionExecutor by importing `typing` and
using `get_type_hints` to correctly resolve annotations.
This fixes validation failures when `from __future__ import annotations`
is enabled, which stores annotations as strings.
Fixes#1808
* Update python/packages/core/tests/workflow/test_function_executor_future.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* ran pre commit
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Fix bug where ChatAgent system instructions were not captured in Langfuse
traces due to incorrect attribute access.
The observability code was attempting to retrieve instructions using
getattr(self, "instructions", None), but ChatAgent stores instructions
in self.chat_options.instructions. This caused system_instructions to
always be None in Langfuse traces.
Changed both _trace_agent_run and _trace_agent_run_stream functions
to correctly retrieve instructions from chat_options.instructions.
Fixes affect:
- Line 1123: _trace_agent_run (non-streaming)
- Line 1192: _trace_agent_run_stream (streaming)
2025-11-19 07:08:27 +00:00
Eduard van ValkenburgGitHubCopilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>eavanvalkenburgcopilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>CopilotVictor Dibia
Update XML documentation to clarify exception behavior.
See `ChatClientAgentThreadTests.SetConversationIdThrowsWhenMessageStoreIsSet` which already verifies this is the actual behavior.
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
* Fix: Prevent duplicate MCP tools and prompts (#1876)
- Added deduplication logic in MCPTool.load_tools() method
- Added deduplication logic in MCPTool.load_prompts() method
- Track existing function names before loading from MCP server
- Skip tools/prompts that are already registered in _functions list
- Prevents 400 error from Azure AI Foundry caused by duplicate tool names
The issue occurred because load_tools() was being called multiple times
(during connect() and by notification handlers), causing tools to be
appended without duplicate checking.
Changes made:
1. In load_tools(): Added existing_names set to track registered functions
2. In load_tools(): Added check to skip tools already in existing_names
3. In load_prompts(): Applied same deduplication pattern
Testing:
- Created unit test verifying deduplication logic
- Confirmed duplicates are skipped correctly
- Confirmed new functions are added correctly
- Prevents duplicate tool names being sent to LLM
Fixes#1876
* Address review feedback: Prevent multiple calls to load_tools and load_prompts
- Added _tools_loaded and _prompts_loaded flags to MCPTool class
- Modified load_tools() to check if already loaded and return early
- Modified load_prompts() to check if already loaded and return early
- Moved test cases from test_mcp_fix.py to test_mcp.py
- Added tests for multiple call prevention
- Deleted separate test_mcp_fix.py file
Addresses review feedback from @eavanvalkenburg:
- Prevents accidental multiple calls to load_tools()
- Prevents accidental multiple calls to load_prompts()
- Test file now in proper location (test_mcp.py)
* Address review feedback: Move flag checks to connect() and remove comments
- Removed verbose comments from code
- Moved _tools_loaded and _prompts_loaded checks to connect() method
- Allows manual calls to load_tools() and load_prompts() for updates
- Updated tests to reflect new behavior
- connect() now prevents duplicate loading during connection
- Users can still manually call load_tools()/load_prompts() to refresh
Addresses feedback from @eavanvalkenburg
* Fix: Code quality and formatting issues
- Applied black formatting
- Fixed ruff linting issues
- All tests passing locally
* chore: Re-run uv lock per review request
* Apply pre-commit formatting: consolidate type annotations
- Consolidate multi-line type annotations to single line
- Remove unnecessary parentheses
- Apply ruff format and security checks
* Move Purview integration logic into middleware
* Improve error handling and user id management
* Rename purview package
* Handle 402s more explicitly; add Middleware generation methods; don't ignore exceptions
* Use DI container; pass scope id to PC
* Add protection scope caching
* Wrap more exceptions in PurviewClient
* Remove block check dedup; add tests
* Refactor PurviewWrapper intialization; Add unit tests
* Use different .Use method and add IDisposable stub
* Add background job processing for Purview
* Misc comment cleanup
* Apply copilot comments
* Fix formatting
* Formatting other files to fix pipeline
* Small updates to settings and exceptions
* Add README
* Move Purview sample
* Address review comments and update XML comments
* Newline after namespace
* Move public Purview classes to single namespace; Clean up csproj and slnx
* Commit the renames
* Remove unused openAI dependency
---------
Co-authored-by: Dmytro Struk <13853051+dmytrostruk@users.noreply.github.com>
* fix devui regression from #2021 where all input is stringified but devui HIL input does not handle stringified json strings correctly.
* update incorrect test
* add devui hil input tests
This commit fixes three issues in the security_filter_middleware:
1. Missing context.terminate flag - Without this, middleware continues processing after setting blocked response
2. No streaming support - When context.is_streaming is True, middleware now returns async generator with ChatResponseUpdate
3. Checks all messages - Changed to check only context.messages[-1] (most recent user message) instead of iterating through conversation history
Changes:
- Added AsyncIterable import
- Added ChatResponseUpdate and TextContent imports
- Modified security_filter_middleware to handle both streaming and non-streaming modes
- Added context.terminate = True to properly stop execution
- Changed message checking logic to only inspect the last user message
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
* Added changes (#1909)
* Python: [Feature Branch] Renamed Azure AI agent and small fixes (#1919)
* Renaming
* Small fixes
* Update python/packages/core/agent_framework/openai/_shared.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Small fix
* Python: [Feature Branch] Added use_latest_version parameter to AzureAIClient (#1959)
* Added use_latest_version parameter to AzureAIClient
* Added unit tests
* Update python/samples/getting_started/agents/azure_ai/azure_ai_use_latest_version.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update python/packages/azure-ai/agent_framework_azure_ai/_client.py
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
* Python: [Feature Branch] Structured Outputs and more examples for AzureAIClient (#1987)
* Small updates
* Added support for structured outputs
* Added code interpreter example
* More examples and fixes
* Added more examples and README
* Small fix
* Addressed PR feedback
* Removed optional ID from FunctionResultContent (#2011)
* Added hosted MCP support (#2018)
* Python: [Feature Branch] Fixed "store" parameter handling (#2069)
* Fixed store parameter handling
* Small fix
* Python: [Feature Branch] Added more examples and fixes for Azure AI agent (#2077)
* Updated azure-ai-projects package version
* Added an example of hosted MCP with approval required
* Updated code interpreter example
* Added file search example
* Update python/samples/getting_started/agents/azure_ai/azure_ai_with_file_search.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update python/samples/getting_started/agents/azure_ai/azure_ai_with_file_search.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Small fix
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Added handling for conversation_id (#2098)
* Merge from main
* Revert "Merge from main"
This reverts commit b8206a85d7.
* Python: [Feature Branch] Merge from main to Azure AI branch (#2111)
* Do not build DevUI assets during .NET project build (#2010)
* .NET: Add unit tests for declarative executor SetMultipleVariables (#2016)
* Add unit tests for create conversation executor
* Update indentation and comment typo.
* Added unit tests for declarative executor SetMultipleVariablesExecutor
* Updated comments and syntactic sugar
* Python: DevUI: Use metadata.entity_id instead of model field (#1984)
* DevUI: Use metadata.entity_id for agent/workflow name instead of model field
* OpenAI Responses: add explicit request validation
* Review feedback
* .NET: DevUI - Do not automatically add/map OpenAI services/endpoints (#2014)
* Don't add OpenAIResponses as part of Dev UI
You should be able to add and remove Dev UI without impacting your other production endpoints.
* Remove `AddDevUI()` and do not map OpenAI endpoints from `MapDevUI()`
* Fix comment wording
* Revise documentation
---------
Co-authored-by: Daniel Roth <daroth@microsoft.com>
* Python: DevUI: Add OpenAI Responses API proxy support + HIL for Workflows (#1737)
* DevUI: Add OpenAI Responses API proxy support with enhanced UI features
This commit adds support for proxying requests to OpenAI's Responses API,
allowing DevUI to route conversations to OpenAI models when configured to enable testing.
Backend changes:
- Add OpenAI proxy executor with conversation routing logic
- Enhance event mapper to support OpenAI Responses API format
- Extend server endpoints to handle OpenAI proxy mode
- Update models with OpenAI-specific response types
- Remove emojis from logging and CLI output for cleaner text
Frontend changes:
- Add settings modal with OpenAI proxy configuration UI
- Enhance agent and workflow views with improved state management
- Add new UI components (separator, switch) for settings
- Update debug panel with better event filtering
- Improve message renderers for OpenAI content types
- Update types and API client for OpenAI integration
* update ui, settings modal and workflow input form, add register cleanup hooks.
* add workflow HIL support, user mode, other fixes
* feat(devui): add human-in-the-loop (HIL) support with dynamic response schemas
Implement HIL workflow support allowing workflows to pause for user input
with dynamically generated JSON schemas based on response handler type hints.
Key Features:
- Automatic response schema extraction from @response_handler decorators
- Dynamic form generation in UI based on Pydantic/dataclass response types
- Checkpoint-based conversation storage for HIL requests/responses
- Resume workflow execution after user provides HIL response
Backend Changes:
- Add extract_response_type_from_executor() to introspect response handlers
- Enrich RequestInfoEvent with response_schema via _enrich_request_info_event_with_response_schema()
- Map RequestInfoEvent to response.input.requested OpenAI event format
- Store HIL responses in conversation history and restore checkpoints
Frontend Changes:
- Add HILInputModal component with SchemaFormRenderer for dynamic forms
- Support Pydantic BaseModel and dataclass response types
- Render enum fields as dropdowns, strings as text/textarea, numbers, booleans, arrays, objects
- Display original request context alongside response form
Testing:
- Add tests for checkpoint storage (test_checkpoints.py)
- Add schema generation tests for all input types (test_schema_generation.py)
- Validate end-to-end HIL flow with spam workflow sample
This enables workflows to seamlessly pause execution and request structured user input
with type-safe, validated forms generated automatically from response type annotations.
* improve HIL support, improve workflow execution view
* ui updates
* ui updates
* improve HIL for workflows, add auth and view modes
* update workflow
* security improvements , ui fixes
* fix mypy error
* update loading spinner in ui
---------
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
* .NET: Remove launchSettings.json from .gitignore in dotnet/samples (#2006)
* Remove launchSettings.json from .gitignore in dotnet/samples
* Update dotnet/samples/GettingStarted/DevUI/DevUI_Step01_BasicUsage/Properties/launchSettings.json
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update dotnet/samples/AGUIClientServer/AGUIServer/Properties/launchSettings.json
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* DevUI: Serialize workflow input as string to maintain conformance with OpenAI Responses format (#2021)
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
* Add Microsoft Agent Framework logo to assets (#2007)
* Updated package versions (#2027)
* DevUI: Prevent line breaks within words in the agent view (#2024)
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
* .NET [AG-UI]: Adds support for shared state. (#1996)
* Product changes
* Tests
* Dojo project
* Cleanups
* Python: Fix underlying tool choice bug and all for return to previous Handoff subagent (#2037)
* Fix tool_choice override bug and add enable_return_to_previous support
* Add unit test for handoff checkpointing
* Handle tools when we have them
* added missing chatAgent params (#2044)
* .NET: fix ChatCompletions Tools serialization (#2043)
* fix serialization in chat completions on tools
* nit
* .NET: assign AgentCard's URL to mapped-endpoint if not defined explicitly (#2047)
* fix serialization in chat completions on tools
* nit
* write e2e test for agent card resolve + adjust behavior
* nit
* Version 1.0.0-preview.251110.1 (#2048)
* .NET: Remove moved OpenAPI sample and point to SK one. (#1997)
* Remove moved OpenAPI sample and point to SK one.
* Update dotnet/samples/GettingStarted/Agents/README.md
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Bump AWSSDK.Extensions.Bedrock.MEAI from 4.0.4.2 to 4.0.4.6 (#2031)
---
updated-dependencies:
- dependency-name: AWSSDK.Extensions.Bedrock.MEAI
dependency-version: 4.0.4.6
dependency-type: direct:production
update-type: version-update:semver-patch
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* .NET: Separate all memory and rag samples into their own folders (#2000)
* Separate all memory and rag samples into their own folders
* Fix broken link.
* Python: .Net: Dotnet devui compatibility fixes (#2026)
* DevUI: Add OpenAI Responses API proxy support with enhanced UI features
This commit adds support for proxying requests to OpenAI's Responses API,
allowing DevUI to route conversations to OpenAI models when configured to enable testing.
Backend changes:
- Add OpenAI proxy executor with conversation routing logic
- Enhance event mapper to support OpenAI Responses API format
- Extend server endpoints to handle OpenAI proxy mode
- Update models with OpenAI-specific response types
- Remove emojis from logging and CLI output for cleaner text
Frontend changes:
- Add settings modal with OpenAI proxy configuration UI
- Enhance agent and workflow views with improved state management
- Add new UI components (separator, switch) for settings
- Update debug panel with better event filtering
- Improve message renderers for OpenAI content types
- Update types and API client for OpenAI integration
* update ui, settings modal and workflow input form, add register cleanup hooks.
* add workflow HIL support, user mode, other fixes
* feat(devui): add human-in-the-loop (HIL) support with dynamic response schemas
Implement HIL workflow support allowing workflows to pause for user input
with dynamically generated JSON schemas based on response handler type hints.
Key Features:
- Automatic response schema extraction from @response_handler decorators
- Dynamic form generation in UI based on Pydantic/dataclass response types
- Checkpoint-based conversation storage for HIL requests/responses
- Resume workflow execution after user provides HIL response
Backend Changes:
- Add extract_response_type_from_executor() to introspect response handlers
- Enrich RequestInfoEvent with response_schema via _enrich_request_info_event_with_response_schema()
- Map RequestInfoEvent to response.input.requested OpenAI event format
- Store HIL responses in conversation history and restore checkpoints
Frontend Changes:
- Add HILInputModal component with SchemaFormRenderer for dynamic forms
- Support Pydantic BaseModel and dataclass response types
- Render enum fields as dropdowns, strings as text/textarea, numbers, booleans, arrays, objects
- Display original request context alongside response form
Testing:
- Add tests for checkpoint storage (test_checkpoints.py)
- Add schema generation tests for all input types (test_schema_generation.py)
- Validate end-to-end HIL flow with spam workflow sample
This enables workflows to seamlessly pause execution and request structured user input
with type-safe, validated forms generated automatically from response type annotations.
* improve HIL support, improve workflow execution view
* ui updates
* ui updates
* improve HIL for workflows, add auth and view modes
* update workflow
* security improvements , ui fixes
* fix mypy error
* update loading spinner in ui
* DevUI: Serialize workflow input as string to maintain conformance with OpenAI Responses format
* Phase 1: Add /meta endpoint and fix workflow event naming for .NET DevUI compatibility
* additional fixes for .NET DevUI workflow visualization item ID tracking
**Problem:**
.NET DevUI was generating different item IDs for ExecutorInvokedEvent and
ExecutorCompletedEvent, causing only the first executor to highlight in the
workflow graph. Long executor names and error messages also broke UI layout.
**Changes:**
- Add ExecutorActionItemResource to match Python DevUI implementation
- Track item IDs per executor using dictionary in AgentRunResponseUpdateExtensions
- Reuse same item ID across invoked/completed/failed events for proper pairing
- Add truncateText() utility to workflow-utils.ts
- Truncate executor names to 35 chars in execution timeline
- Truncate error messages to 150 chars in workflow graph nodes
** Details:**
- ExecutorActionItemResource registered with JSON source generation context
- Dictionary cleaned up after executor completion/failure to prevent memory leaks
- Frontend item tracking by unique item.id supports multiple executor runs
- All changes follow existing codebase patterns and conventions
Tested with review-workflow showing correct executor highlighting and state
transitions for sequential and concurrent executors.
* format fixes, remove cors tests
* remove unecessary attributes
---------
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
Co-authored-by: Reuben Bond <reuben.bond@gmail.com>
* DevUI: support having both an agent and a workflow with the same id in discovery (#2023)
* Python: Fix Model ID attribute not showing up in `invoke_agent` span (#2061)
* Best effort to surface the model id to invoke agent span
* Fix tests
* Fix tests
* Version 1.0.0-preview.251107.2 (#2065)
* Version 1.0.0-preview.251110.2 (#2067)
* Update README.md to change Grafana links to Azure portal links for dashboard access (#1983)
* .NET - Enable build & test on branch `feature-foundry-agents` (#2068)
* Tests good, mkay
* Update .github/workflows/dotnet-build-and-test.yml
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Enable feature build pipelines
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Roger Barreto <19890735+rogerbarreto@users.noreply.github.com>
* Python: Add concrete AGUIChatClient (#2072)
* Add concrete AGUIChatClient
* Update logging docstrings and conventions
* PR feedback
* Updates to support client-side tool calls
* .NET: Move catalog samples to the HostedAgents folder (#2090)
* move catalog samples to the HostedAgents folder
* move the catalog samples' projects to the HostedAgents folder
* Bump OpenTelemetry.Instrumentation.Runtime from 1.12.0 to 1.13.0 (#1856)
---
updated-dependencies:
- dependency-name: OpenTelemetry.Instrumentation.Runtime
dependency-version: 1.13.0
dependency-type: direct:production
update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* .NET: Bump Microsoft.SemanticKernel.Agents.Abstractions from 1.66.0 to 1.67.0 (#1962)
* Bump Microsoft.SemanticKernel.Agents.Abstractions from 1.66.0 to 1.67.0
---
updated-dependencies:
- dependency-name: Microsoft.SemanticKernel.Agents.Abstractions
dependency-version: 1.67.0
dependency-type: direct:production
update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
* .NET: Bump all Microsoft.SemanticKernel packages from 1.66.* to 1.67.* (#1969)
* Initial plan
* Update all Microsoft.SemanticKernel packages to 1.67.*
Co-authored-by: rogerbarreto <19890735+rogerbarreto@users.noreply.github.com>
* Remove unrelated changes to package-lock.json and yarn.lock
Co-authored-by: markwallace-microsoft <127216156+markwallace-microsoft@users.noreply.github.com>
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: rogerbarreto <19890735+rogerbarreto@users.noreply.github.com>
Co-authored-by: markwallace-microsoft <127216156+markwallace-microsoft@users.noreply.github.com>
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
Co-authored-by: rogerbarreto <19890735+rogerbarreto@users.noreply.github.com>
Co-authored-by: markwallace-microsoft <127216156+markwallace-microsoft@users.noreply.github.com>
* .NET: fix: WorkflowAsAgent Sample (#1787)
* fix: WorkflowAsAgent Sample
* Also makes ChatForwardingExecutor public
* feat: Expand ChatForwardingExecutor handled types
Make ChatForwardingExecutor match the input types of ChatProtocolExecutor.
* fix: Update for the new AgentRunResponseUpdate merge logic
AIAgent always sends out List<ChatMessage> now.
* Updated (#2076)
* Bump vite in /python/samples/demos/chatkit-integration/frontend (#1918)
Bumps [vite](https://github.com/vitejs/vite/tree/HEAD/packages/vite) from 7.1.9 to 7.1.12.
- [Release notes](https://github.com/vitejs/vite/releases)
- [Changelog](https://github.com/vitejs/vite/blob/v7.1.12/packages/vite/CHANGELOG.md)
- [Commits](https://github.com/vitejs/vite/commits/v7.1.12/packages/vite)
---
updated-dependencies:
- dependency-name: vite
dependency-version: 7.1.12
dependency-type: direct:development
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* Bump Roslynator.Analyzers from 4.14.0 to 4.14.1 (#1857)
---
updated-dependencies:
- dependency-name: Roslynator.Analyzers
dependency-version: 4.14.1
dependency-type: direct:production
update-type: version-update:semver-patch
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* Bump MishaKav/pytest-coverage-comment from 1.1.57 to 1.1.59 (#2034)
Bumps [MishaKav/pytest-coverage-comment](https://github.com/mishakav/pytest-coverage-comment) from 1.1.57 to 1.1.59.
- [Release notes](https://github.com/mishakav/pytest-coverage-comment/releases)
- [Changelog](https://github.com/MishaKav/pytest-coverage-comment/blob/main/CHANGELOG.md)
- [Commits](https://github.com/mishakav/pytest-coverage-comment/compare/v1.1.57...v1.1.59)
---
updated-dependencies:
- dependency-name: MishaKav/pytest-coverage-comment
dependency-version: 1.1.59
dependency-type: direct:production
update-type: version-update:semver-patch
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
* Python: Handle agent user input request in AgentExecutor (#2022)
* Handle agent user input request in AgentExecutor
* fix test
* Address comments
* Fix tests
* Fix tests
* Address comments
* Address comments
* Python: OpenAI Responses Image Generation Stream Support, Sample and Unit Tests (#1853)
* support for image gen streaming
* small fixes
* fixes
* added comment
* Python: Fix MCP Tool Parameter Descriptions Not Propagated to LLMs (#1978)
* mcp tool description fix
* small fix
* .NET: Allow extending agent run options via additional properties (#1872)
* Allow extending agent run options via additional properties
This mirrors the M.E.AI model in ChatOptions.AdditionalProperties which is very useful when building functionality pipelines.
Fixes https://github.com/microsoft/agent-framework/issues/1815
* Expand XML documentation
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Add AdditionalProperties tests to AgentRunOptions
Co-authored-by: kzu <169707+kzu@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: kzu <169707+kzu@users.noreply.github.com>
* Python: Use the last entry in the task history to avoid empty responses (#2101)
* Use the last entry in the task history to avoid empty responses
* History only contains Messages
* Updated package versions (#2104)
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Reuben Bond <203839+ReubenBond@users.noreply.github.com>
Co-authored-by: Peter Ibekwe <109177538+peibekwe@users.noreply.github.com>
Co-authored-by: Jeff Handley <jeffhandley@users.noreply.github.com>
Co-authored-by: Daniel Roth <daroth@microsoft.com>
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Shawn Henry <sphenry@gmail.com>
Co-authored-by: Javier Calvarro Nelson <jacalvar@microsoft.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
Co-authored-by: Eduard van Valkenburg <eavanvalkenburg@users.noreply.github.com>
Co-authored-by: Korolev Dmitry <deagle.gross@gmail.com>
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Reuben Bond <reuben.bond@gmail.com>
Co-authored-by: Tao Chen <taochen@microsoft.com>
Co-authored-by: wuweng <wuweng@microsoft.com>
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
Co-authored-by: Roger Barreto <19890735+rogerbarreto@users.noreply.github.com>
Co-authored-by: SergeyMenshykh <68852919+SergeyMenshykh@users.noreply.github.com>
Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Jacob Alber <jaalber@microsoft.com>
Co-authored-by: Giles Odigwe <79032838+giles17@users.noreply.github.com>
Co-authored-by: Daniel Cazzulino <daniel@cazzulino.com>
Co-authored-by: kzu <169707+kzu@users.noreply.github.com>
* Updated azure-ai-projects package version and small fixes (#2139)
* Python: [Feature Branch] Resolve CI issues (#2143)
* Small documentation and code fixes
* Small fix in documentation
* Addressed PR feedback
* Added AI Search example
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
Co-authored-by: Reuben Bond <203839+ReubenBond@users.noreply.github.com>
Co-authored-by: Peter Ibekwe <109177538+peibekwe@users.noreply.github.com>
Co-authored-by: Jeff Handley <jeffhandley@users.noreply.github.com>
Co-authored-by: Daniel Roth <daroth@microsoft.com>
Co-authored-by: Victor Dibia <chuvidi2003@gmail.com>
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
Co-authored-by: Shawn Henry <sphenry@gmail.com>
Co-authored-by: Javier Calvarro Nelson <jacalvar@microsoft.com>
Co-authored-by: Eduard van Valkenburg <eavanvalkenburg@users.noreply.github.com>
Co-authored-by: Korolev Dmitry <deagle.gross@gmail.com>
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Reuben Bond <reuben.bond@gmail.com>
Co-authored-by: Tao Chen <taochen@microsoft.com>
Co-authored-by: wuweng <wuweng@microsoft.com>
Co-authored-by: Roger Barreto <19890735+rogerbarreto@users.noreply.github.com>
Co-authored-by: SergeyMenshykh <68852919+SergeyMenshykh@users.noreply.github.com>
Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Jacob Alber <jaalber@microsoft.com>
Co-authored-by: Giles Odigwe <79032838+giles17@users.noreply.github.com>
Co-authored-by: Daniel Cazzulino <daniel@cazzulino.com>
Co-authored-by: kzu <169707+kzu@users.noreply.github.com>
* move catalog samples to the HostedAgents folder
* move the catalog samples' projects to the HostedAgents folder
* move deep research sample out of the HostedAgents folder
* fix: WorkflowAsAgent Sample
* Also makes ChatForwardingExecutor public
* feat: Expand ChatForwardingExecutor handled types
Make ChatForwardingExecutor match the input types of ChatProtocolExecutor.
* fix: Update for the new AgentRunResponseUpdate merge logic
AIAgent always sends out List<ChatMessage> now.
* DevUI: Add OpenAI Responses API proxy support with enhanced UI features
This commit adds support for proxying requests to OpenAI's Responses API,
allowing DevUI to route conversations to OpenAI models when configured to enable testing.
Backend changes:
- Add OpenAI proxy executor with conversation routing logic
- Enhance event mapper to support OpenAI Responses API format
- Extend server endpoints to handle OpenAI proxy mode
- Update models with OpenAI-specific response types
- Remove emojis from logging and CLI output for cleaner text
Frontend changes:
- Add settings modal with OpenAI proxy configuration UI
- Enhance agent and workflow views with improved state management
- Add new UI components (separator, switch) for settings
- Update debug panel with better event filtering
- Improve message renderers for OpenAI content types
- Update types and API client for OpenAI integration
* update ui, settings modal and workflow input form, add register cleanup hooks.
* add workflow HIL support, user mode, other fixes
* feat(devui): add human-in-the-loop (HIL) support with dynamic response schemas
Implement HIL workflow support allowing workflows to pause for user input
with dynamically generated JSON schemas based on response handler type hints.
Key Features:
- Automatic response schema extraction from @response_handler decorators
- Dynamic form generation in UI based on Pydantic/dataclass response types
- Checkpoint-based conversation storage for HIL requests/responses
- Resume workflow execution after user provides HIL response
Backend Changes:
- Add extract_response_type_from_executor() to introspect response handlers
- Enrich RequestInfoEvent with response_schema via _enrich_request_info_event_with_response_schema()
- Map RequestInfoEvent to response.input.requested OpenAI event format
- Store HIL responses in conversation history and restore checkpoints
Frontend Changes:
- Add HILInputModal component with SchemaFormRenderer for dynamic forms
- Support Pydantic BaseModel and dataclass response types
- Render enum fields as dropdowns, strings as text/textarea, numbers, booleans, arrays, objects
- Display original request context alongside response form
Testing:
- Add tests for checkpoint storage (test_checkpoints.py)
- Add schema generation tests for all input types (test_schema_generation.py)
- Validate end-to-end HIL flow with spam workflow sample
This enables workflows to seamlessly pause execution and request structured user input
with type-safe, validated forms generated automatically from response type annotations.
* improve HIL support, improve workflow execution view
* ui updates
* ui updates
* improve HIL for workflows, add auth and view modes
* update workflow
* security improvements , ui fixes
* fix mypy error
* update loading spinner in ui
* DevUI: Serialize workflow input as string to maintain conformance with OpenAI Responses format
* Phase 1: Add /meta endpoint and fix workflow event naming for .NET DevUI compatibility
* additional fixes for .NET DevUI workflow visualization item ID tracking
**Problem:**
.NET DevUI was generating different item IDs for ExecutorInvokedEvent and
ExecutorCompletedEvent, causing only the first executor to highlight in the
workflow graph. Long executor names and error messages also broke UI layout.
**Changes:**
- Add ExecutorActionItemResource to match Python DevUI implementation
- Track item IDs per executor using dictionary in AgentRunResponseUpdateExtensions
- Reuse same item ID across invoked/completed/failed events for proper pairing
- Add truncateText() utility to workflow-utils.ts
- Truncate executor names to 35 chars in execution timeline
- Truncate error messages to 150 chars in workflow graph nodes
** Details:**
- ExecutorActionItemResource registered with JSON source generation context
- Dictionary cleaned up after executor completion/failure to prevent memory leaks
- Frontend item tracking by unique item.id supports multiple executor runs
- All changes follow existing codebase patterns and conventions
Tested with review-workflow showing correct executor highlighting and state
transitions for sequential and concurrent executors.
* format fixes, remove cors tests
* remove unecessary attributes
---------
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
Co-authored-by: Reuben Bond <reuben.bond@gmail.com>
* DevUI: Add OpenAI Responses API proxy support with enhanced UI features
This commit adds support for proxying requests to OpenAI's Responses API,
allowing DevUI to route conversations to OpenAI models when configured to enable testing.
Backend changes:
- Add OpenAI proxy executor with conversation routing logic
- Enhance event mapper to support OpenAI Responses API format
- Extend server endpoints to handle OpenAI proxy mode
- Update models with OpenAI-specific response types
- Remove emojis from logging and CLI output for cleaner text
Frontend changes:
- Add settings modal with OpenAI proxy configuration UI
- Enhance agent and workflow views with improved state management
- Add new UI components (separator, switch) for settings
- Update debug panel with better event filtering
- Improve message renderers for OpenAI content types
- Update types and API client for OpenAI integration
* update ui, settings modal and workflow input form, add register cleanup hooks.
* add workflow HIL support, user mode, other fixes
* feat(devui): add human-in-the-loop (HIL) support with dynamic response schemas
Implement HIL workflow support allowing workflows to pause for user input
with dynamically generated JSON schemas based on response handler type hints.
Key Features:
- Automatic response schema extraction from @response_handler decorators
- Dynamic form generation in UI based on Pydantic/dataclass response types
- Checkpoint-based conversation storage for HIL requests/responses
- Resume workflow execution after user provides HIL response
Backend Changes:
- Add extract_response_type_from_executor() to introspect response handlers
- Enrich RequestInfoEvent with response_schema via _enrich_request_info_event_with_response_schema()
- Map RequestInfoEvent to response.input.requested OpenAI event format
- Store HIL responses in conversation history and restore checkpoints
Frontend Changes:
- Add HILInputModal component with SchemaFormRenderer for dynamic forms
- Support Pydantic BaseModel and dataclass response types
- Render enum fields as dropdowns, strings as text/textarea, numbers, booleans, arrays, objects
- Display original request context alongside response form
Testing:
- Add tests for checkpoint storage (test_checkpoints.py)
- Add schema generation tests for all input types (test_schema_generation.py)
- Validate end-to-end HIL flow with spam workflow sample
This enables workflows to seamlessly pause execution and request structured user input
with type-safe, validated forms generated automatically from response type annotations.
* improve HIL support, improve workflow execution view
* ui updates
* ui updates
* improve HIL for workflows, add auth and view modes
* update workflow
* security improvements , ui fixes
* fix mypy error
* update loading spinner in ui
---------
Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
* Don't add OpenAIResponses as part of Dev UI
You should be able to add and remove Dev UI without impacting your other production endpoints.
* Remove `AddDevUI()` and do not map OpenAI endpoints from `MapDevUI()`
* Fix comment wording
* Revise documentation
---------
Co-authored-by: Daniel Roth <daroth@microsoft.com>
* Add unit tests for create conversation executor
* Update indentation and comment typo.
* Added unit tests for declarative executor SetMultipleVariablesExecutor
* Updated comments and syntactic sugar
* Add ChatHistoryMemoryProvider with unit tests
* Set new project to not packable.
* Fix bugs
* Add serialization support.
* Update dotnet/src/Microsoft.Agents.AI.VectorDataMemory/ChatHistoryMemoryProvider.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Remove unnecessary line
* Convert ChatHistoryMemoryProvider to use Dynamic collections.
* Sealing options and scope classes.
* Add sample, add scope to logs and improve scope validation
* Move ChatHistoryMemoryProvider to MAAI project.
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix the ordering of chained resolvers in JsonSerializerOptions
We want the resolvers from AIJsonUtilities to be used before the ones from the source generator, in case the source generator emits its own copy in that assembly for the M.E.AI types.
* Apply suggestions from code review
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update dotnet/src/Microsoft.Agents.AI/AgentJsonUtilities.cs
* Update dotnet/src/Microsoft.Agents.AI.Mem0/Mem0JsonUtilities.cs
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
* Remove unused using directive in Mem0JsonUtilities
Removed unused using directive for Microsoft.Extensions.AI.
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: westey <164392973+westey-m@users.noreply.github.com>
* fix: Workflow Validation
* adds orphan validation to workflow builder
* adds tests for workflow validation
* expands on the underlying reasoning why type validation is not supported
* fixup: CodeGen template
* [BREAKING] refactor: Normalize WorkflowBuilder APIs
* "partitioner" => "assigner"
* normalize ordering so sources always to the left of targets for edges
* normalize parameter ordering so sources and targets are always first arguments
* remove `params` (users should use collection expressions instead)
* refactor: Align name with Python
In #1551 we added a mechanism to open a Streaming workflow run without providing any input. This caused unintuitive behaviour when passing a string as input without providing a runId, resulting in the input being misinterpreted as the runId and the workflow not executing.
As well, the name caused confusion about why the Workflow was not running when using the input-less StreamAsync (since the Workflow cannot run without any messages to drive its execution).
* Removed automatic agent cleanup in AzureAIAgentClient
* Revert "Removed automatic agent cleanup in AzureAIAgentClient"
This reverts commit 89846c7212.
* Exposed boolean flag to control deletion behavior
* Update sample
* Initial plan
* Infrastructure setup
* Plan for minimal client
* Plan update
* Basic agentic chat
* cleanup
* Cleanups
* More cleanups
* Cleanups
* More cleanups
* Test plan
* Sample
* Fix streaming and error handling
* Fix notifications
* Cleanups
* cleanup sample
* Additional tests
* Additional tests
* Run dotnet format
* Remove unnecessary files
* Mark packages as non packable
* Fix build
* Address feedback
* Fix build
* Fix remaining warnings
* Feedback
* Feedback and cleanup
* Cleanup
* Cleanups
* Cleanups
* Cleanups
* Retrieve existing messages from the store to send them along the way and update the sample client
* Run dotnet format
* Add ADR for AG-UI
* Switch to use the SG and use a convention for run ids
* Cleanup MapAGUI API
* Fix formatting
* Fix solution
* Fix solution
* feat: Add ChatKit integration with a new frontend application
- Created a new frontend application using React and Vite for the ChatKit integration.
- Added essential files including package.json, vite.config.ts, and Tailwind CSS configuration.
- Implemented core components: App, Home, ChatKitPanel, ThemeToggle, and hooks for color scheme management.
- Established SQLite-based store implementation for ChatKit data persistence in store.py.
- Integrated theme toggling functionality for light and dark modes.
- Set up ESLint and TypeScript configurations for better development experience.
* git ignore
* fix mypy
* add mising file
* minimal frontend for chatkit sample
* update ignore files
* version
* set python version lowerbound on chatkit
* update project settings for chatkit
* update setup
* update setup
* update setup
* update setup
* weather widget
* add select city widget sample
* remove widget helper
* update chatkit to include file attachments and cover more thread item types
* update readme with mermaid diagram
* update diagram
* update instructions
* update chatkit dependency
* fix converter imports
* move to demos/
* move to demos/ -- rename references
* support multiple session instead of using global variable in sample
* support chunk streaming
* fix tests
* Update python/samples/demos/chatkit-integration/store.py
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
* use local host
---------
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
* Adding Sample for writer-critic workflow implemented using Worfklow, custom executors, agents, switch, custom states, different entry points for the executors.
* Update dotnet/samples/GettingStarted/Workflows/_Foundational/08_WriterCriticWorkflow/Program.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update dotnet/samples/GettingStarted/Workflows/_Foundational/08_WriterCriticWorkflow/Program.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* using now structured output, with streaming for UX responsiveness.
* improved comments and order, so comments directly precede what they're describing
* fixing issue with internal class that the analyzer doesn't recognize that CriticDecision is instantiated, just indirectly via JSON deserialization
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
* Port store for adding text to a vector store to AF
* Fix typo.
* Change TextSearchStore to sample, and add sample to use it and do rag with a custom schema
* Add more tests and fix broken ones
* Fix merge issue
* Fix sample after merge.
* Convert TextSearchStore to use Dynamic mode to be AOT compatible.
* Add some more clarification on when to use assistant messages in rag searches.
* Adding sample demonstrating hosted MCP with Responses
* Add mcp readme.md to slnx
* Update FoundryAgent sample to use MCP types from abstraction and to show how to do approval
* Fix param name after package update.
* Fix environment variable name for consistency
* Apply suggestion from @Copilot
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Add workflow as an agent with observability sample
* Address comment
* Fix formatting
* enable sensitive data
* enable sensitive data for sub agents
* adjust aggregator handlers
* initial version of anthropic connector
* updated implementation and added tests
* fix type and readme
* mypy fix and int tests enabled
* add integration test setup
* updated based on comments
* improved function result handling
* added extra unordered test
* updated from review
* fix tool choice handling
* same fix for chat client
* refactor: Unify ExecutorIsh and ExecutorRegistration => ExecutorBinding
* Switch to more modern Record type-tree for Sum Types
* Unify APIs for getting ExecutorBinding
* Fix an issue where workflows consisting entirely of cross-run shareable executors which are not instance-resettable do not properly clear state when running non-concurrently.
* feat: Simplify function-to-executor pattern
* refactor: Normalize API naming
* Propagate cancellation token down the stack
* Added unit tests to cover workflow cancellation scenarios
* Updated tests based on feedback to simplify assert.
* Create custom AsyncEnumrable to gracefully handle cancellation for Channel reader. Tailor cancellation tests to declarative scenarios.
* Update comment and naming for readability.
* Fixing minor stylistic recommendation.
---------
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
* ensure agent thread is part of checkpoint
* Update python/packages/core/agent_framework/_workflows/_agent_executor.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* remove data copying for server side thread.
* refine warning check
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix issue where AIContextProvider messages were not added to MessageStores
* Fix typos
* Update XML docs to reduce ambiguity.
* Update AIContext XML docs
* Fix merge issue
* Prototype: Add request_info API and @response_handler
* Add original_request as a parameter to the response handler
* Prototype: request interception in sub workflows
* Prototype: request interception in sub workflows 2
* WIP: Make checkpointing work
* checkpointing with sub workflow
* Fix function executor
* Allow sub-workflow to output directly
* Remove ReqeustInfoExecutor and related classes; Debugging checkpoint_with_human_in_the_loop
* Fix Handoff and sample
* fix pending requests in checkpoint
* Fix unit tests
* Fix formatting
* Resolve comments
* Address comment
* Add checkpoint tests
* Add tests
* misc
* fix mypy
* fix mypy
* Use request type as part of the key
* Log warning if there is not response handler for a request
* Update Internal edge group comments
* REcord message type in executor processing span
* Update sample
* Improve tests
* Add Rag AIContext Provider
* Fix issues
* Improve options naming based on PR feedback.
* Move Rag Provider to Data namespace
* Add Raw Representation to RagSearchResult
* Renaming RagProvider to TextSearchProvider
* Porting Mem0Provider to AF from SK
* Switch integration tests to manual
* Address issues
* Move Mem0Provider to separate project.
* Move integration tests to new project
* Address PR comments.
* Intro group chat and refactor magentic. Fix as_agent()
* Cleanup and improvements
* Add as_agent docstring clarification
* Standardize orchestration messages to use agent-style inputs.
* Simplify group chat constructs
* Further cleanup
* Add sk to af group chat migration sample. Update README.
* Improvements and simplifications
* consolidating shared orchestration logic
* Further clean up
* Add group chat sample
* Improve typing
* Fix test imports
* Fix readme links
* Cleanup per PR Feedback
Concurrent run support was recently added to workflows, but Orchestrations did not fully update to support it. A few executors were missing Cross-Run Shareable annotations, and the ConcurrentEnd executor needed to be factory-instantiated.
This also ports the fix for #1613 from #1637, to avoid waiting on that PR.
Checkpointing is used by the WorkflowHostAgent to be able to support resume from a provided thread. When a CheckpointManager is not specified, we use the InMemoryCheckpointManager and serialize its state into the thread's Serialize()ed JsonElement.
At some point InMemoryCheckpointManager became not serializable, breaking this behaviour. This change restores serializability, and adds a test.
* Improve conformance of OpenAI Responses API serving
* Update dotnet/src/Microsoft.Agents.AI.Hosting.OpenAI/Responses/AgentRunResponseExtensions.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Update dotnet/src/Microsoft.Agents.AI.Hosting.OpenAI/Responses/AgentRunResponseExtensions.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Sort packages
* Relax adherence where acceptable
* nit
* PromptCacheKey is not obsolete
* format
---------
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Add Handoff orchestration pattern support
* PR feedback
* Use AOAI client in samples
* Adjust to tool
* Handoff to sub-agent via ai function
* PR feedback
* More cleanup
* Improvements
* PR feedback cleanup
* Add handoff migration sample.
* Remove type ignore
* fix markdown link formatting
* Remove readme link for non-existent sample
* use extension methods from A2A package for converting between MEAI and A2A model classes.
* Update dotnet/tests/Microsoft.Agents.AI.A2A.UnitTests/A2AAgentTests.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* remove unused using
---------
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Fix handoff function naming
We don't need the agent's name or a guid in the handoff name... we can just use simple numbering. There's a possibility that someone built an agent with a built-in function tool named "handoff_to_x"; if that turns out to be an issue, we could add back some longer bit of randomness.
* Update dotnet/src/Microsoft.Agents.AI.Workflows/Specialized/HandoffAgentExecutor.cs
---------
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
Remove input type checking in favour of explicit `.DescribeProtocolAsync()` flow. Also removes `.AsAgentAsync()` as the validation happens at workflow run time. This makes it easier to use Workflows with DI without resorting to async-over-sync.
* Add support for getting and creating Assistant and Foundry agents with ChatClientAgentOptions
* Fix options cloning and agent creation
* Fix inconsistency
* Add support for mapping more tools and integration tests for ensuring CreateAIAgent works with those tools.
* Add support for additional openai tools with tests.
* Remove special casing for function tools, since it's either not supported yet, or requires a lot of code duplication.
* Removed unused using.
* Fix broken unit tests
* Change integration test to reduce flakiness.
* Add copilot instructions for c#
* Add some further improvements based on copilot suggestions.
* Update .github/copilot-instructions.md
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update C# sample code guidelines
Added a comment guideline for sample code in C#.
* Fix casing of Program.cs in instructions
* Update guidelines for coding standards and testing
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* unit test for using create agent option by constructor
* remove this for prevent duplicate when ChatClientAgentOption and ChatOption has same Instruction
* update unit test for ChatClientAgentOptions
* ensure function aproval is parsed correctly
* udpate ui, add deployment guide button, other debug panel fixes
* feat(devui): Implement lazy loading architecture with enhanced security and state management
Major architectural improvements to DevUI for better performance, security, and developer experience:
Performance & Architecture:
- Implement lazy loading for entity discovery - entities loaded on-demand instead of at startup
- Add hot reload capability for development workflow via new reload endpoint
- Reduce startup time and memory footprint by deferring module imports
Security Enhancements:
- Remove remote entity loading capabilities (POST /v1/entities/add, DELETE endpoints)
- DevUI now strictly local development tool - no remote code execution
- Add explicit security documentation and best practices in README
Frontend Improvements:
- Migrate to Zustand for centralized state management (replacing prop drilling)
- Add lightweight zero-dependency markdown renderer with code block copy support
- Improve gallery UX with setup instructions modal instead of direct URL loading
- Enhanced message UI with copy functionality and better token usage display
Testing & Quality:
- Expand test coverage for lazy loading, type detection, and cache invalidation
- Add comprehensive tests for new behaviors (+231 lines of test code)
- Improve type safety and documentation throughout
Breaking Changes:
- Remote entity loading via URLs is no longer supported
- Entities must be loaded from local filesystem only
* update ui issues, uupdate test descripion
* refactor: remove unused internals
* feat: Execution Mode for sharing a workflow among concurrent runs
* feat: Update WorkflowHostAgent to support concurrent execution
* Also update AsAgent APIs to support injecting a CheckpointManager and an IWorkflowExecutionEnvironment
* fix: Make Read logic consistent in DeclarativeWorkflowContext
WorkflowHostAgent was initially implemented before Checkpointing was available. This meant that in order to support resuming, the WorkflowHostAgent needed to keep the runs around, which broke it when stricter rules about concurrent sharing of workflows during execution were introduced.
This change updates the hosting logic to release the underlying StreamingRun when the RunStreamingAsync or RunAsync are invoked, in favour of keeping the checkpointing information in the WorkflowThread to enable resumption.
* Python: Fix AI Search Tool Sample and improve AI Search Exceptions
* Python: Fix AI Search Tool Test
---------
Co-authored-by: Dmytro Struk <13853051+dmytrostruk@users.noreply.github.com>
* Python: set role=tool when processing approval responses
* Python: set role=tool when processing approval responses
* Fix approval mode with OpenAIChatClient and threads: add approval requests to assistant message, fix deduplication/rejection call_id, filter approval content, add tests and example
* update test tools after change
* Rename _collect_approval_todos to _collect_approval_responses and filter empty call_ids
* sanitize agent name
* simplify
* Update dotnet/src/Microsoft.Agents.AI/AgentExtensions.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update dotnet/src/Microsoft.Agents.AI/AgentExtensions.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Update dotnet/src/Microsoft.Agents.AI/AgentExtensions.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* Update dotnet/tests/Microsoft.Agents.AI.UnitTests/AgentExtensionsTests.cs
Co-authored-by: Stephen Toub <stoub@microsoft.com>
* change regex to flag underscores as well so their sequence can be replaced with a single one.
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Stephen Toub <stoub@microsoft.com>
- Added ChatMessage[] handler to ConfigureRoutes to support array dispatch
- Workflow runtime can dispatch messages as either List<ChatMessage> or ChatMessage[]
- Added 10 comprehensive unit tests validating message routing behavior
- Tests ensure functionality and protect against future refactoring (issue #782)
- Updated code comments to reflect exact-type-matching requirement
Fixes issue where raw WorkflowBuilder with AIAgent nodes failed to receive
initial messages due to missing array type handler.
* add workflow edge data properties to the workflow.definition tag.
* use workflow info classes for workflow.definition tag value
* add unit test
* fix test
* fix formatting issue
* remove flaky unit test
* remove unused package dependency
* Python: DevUI - Internal Refactor, Conversations API support, and performance improvements
Comprehensive refactor of DevUI package including samples relocation,
frontend reorganization, OpenAI Conversations API support, and critical
performance and code quality improvements.
Key Changes:
Architecture & Organization
- Moved DevUI samples to python/samples/getting_started/devui/
- Consolidated with other framework samples for better discoverability
- Added .env.example files and comprehensive README
- Restructured frontend components into feature-based folders (agent, workflow, gallery, layout)
- Created new OpenAI-compliant message renderers (devui should render oai responses types primarily)
New Features
- Added _conversations.py (467 lines) - Full conversation storage abstraction, replaces the /threads endpoint to better match oai conversations api
- Implements OpenAI Conversations API for thread management, Supports in-memory and extensible storage backends
API Simplification
- Use 'model' field as entity_id (agent/workflow name) instead of extra_body
- Use standard OpenAI 'conversation' field for conversation context.
Performance & Quality Improvements
- Improved context management in MessageMapper with bounded memory (~500KB max)
- Implemented hybrid LRU + cleanup approach to prevent unbounded memory growth
- General QOL improvement - Eliminated ~150 lines of dead/duplicate code, Consolidated helper functions into _utils.py, Extracted magic numbers to module-level constants, Optimized conversation item lookups with index-based approach
Testing
- Added test_conversations.py (13 tests)
- Added test_performance_fixes.py (9 tests)
- Updated existing tests for code consolidation
- 53 tests passing
Impact: 76 files changed: +4,106 insertions, -2,373 deletions
All linting and formatting checks passing. No breaking changes - backward compatible.
Migration: Samples moved to python/samples/getting_started/devui/
* readme lint fixes
* initial support for function approval and minor ui fixes
* Update readme.md sample to show sample parameter values and make azure sdk pre-release
* Add clarifying comment about params.
* Switch sample to use the OpenAI SDK
* Add mapping for application media type in OpenAI responses client
* Enhance multimodal input samples: Add PDF testing functionality and fix image sample
* Standardize filename handling and add multimodal samples
- Standardized filename extraction logic between chat and responses clients
- Both clients now omit filename when not provided (no default fallback)
- Added Azure Responses API multimodal sample with PDF support
- Cleaned up Azure Chat sample to focus on supported features only
- Fixed test comment placement for better code documentation
- Updated README with clear API capability differences
* Enhance multimodal input samples with image and PDF handling
- Refactor image and PDF handling in `azure_chat_multimodal.py` and `openai_chat_multimodal.py` to use new utility functions.
- Add `load_sample_pdf` and `create_sample_image` functions for better test asset management.
- Remove redundant code for creating sample images and PDFs.
- Introduce a sample PDF file in `sample_assets` for testing purposes.
* Fix formatting in OpenAI chat client
---------
Co-authored-by: Dmytro Struk <13853051+dmytrostruk@users.noreply.github.com>
* refactor AgentExecutor, add output_response flag for switching on or off workflow output for each agent.
* introduce add_agent
* make default agent's streaming to false
* address comments
* fix test
* add is_streaming to RunnerContext and WorkflowContext
* fix add_agent return
* fix tests
* address comments
* resolve conflict
* update to address comments
* fix
* improve structured output for chat client agent
* add comment to the result property
* remove code duplication and add tests
* refactor the CreateAIAgent extension methods to return specific types, so consumers can avoid unnecessary downcasting.
* fix type and remove unused using.
* add ChatClientAgentRunResponse and move AgentRunResponse to the abstractions package to reuse later.
* seal ChatClientAgentRunResponse
* update xml comment
* remove funcitons from sample
* rename agent for streaming
* Fix bug where ChatClientAgent throws when providing a ChatMessageStore with a service that requries service storage
* Update dotnet/tests/Microsoft.Agents.AI.UnitTests/ChatClient/ChatClientAgentTests.cs
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* added a sample on integration of Azure OpenAI Responses Client with hosted Model Context Protocol (MCP)
* added additional comments mentioning that the Microsoft Learn MCP server can be replaced by any other desired MCP server
* corrected the MCP server type to local
* added a newline at the end
* Update python/samples/getting_started/agents/azure_openai/azure_responses_client_with_local_mcp.py
Co-authored-by: Dmytro Struk <13853051+dmytrostruk@users.noreply.github.com>
---------
Co-authored-by: Dmytro Struk <13853051+dmytrostruk@users.noreply.github.com>
* Updates to async run loop.
* fix: Workflow Onwership can be release by nonowner
* fix: Incorrect handling of blockOnPending in StreamingRun
Depending on whether we are running in streaming on non-streaming mode, we may be using the StreamingRun in different ways. Unfortunately, the only place we can really know what is the actual state of execution is in the RunEventStream implementations.
This resulted in blocking where blocking was unneeded and occasionally not-blocking when blocking was needed.
The fix is to move the logic of handling this blocking into RunEventStream implementations.
* fix: Fix cleanup on error and end run
This ensures we clean up the background resources correctly.
* fix: Ensure we let the run loop proceed when shutting down
* fix: Add timeout for Input Waiting
* fix: Make the samples properly clean up `Run`s and `StreamingRun`s
* fix: Simplify Declarative Workflow Run disposal pattern
* Also fixes missing .Disposal() in Integration tests
---------
Co-authored-by: Ben Thomas <ben.thomas@microsoft.com>
Capitalize instruction and relax assertion to accept any non-empty response instead of requiring exact text match. Fixes flaky test failures with reasoning models.
* Add dev containers
* Add workspace folder and cs dev extension
* Try other workspace folder format
* Add default solution.
* Move default solution to settings.json
* Fix repo open
* Remove duplicate python codespace and rename folder
* Add recommended C# extensions and a default build task
* Add vscode icons extension by default
* Add python setup and customizations, plus ai studio for all
* Add bash command
* Try running devsetup from workspace folder
* Remove echo and cd
* Change workspace mount
* Change dotnet workspace name
* Revert workspacemount addition
* Try workspace mount to root
* remove trailing slash
* Try workspacefolder with var
* Revert to original approach
* Modify dev containers to work in main repo.
* Remove trailing comma
* Apply suggestion from @Copilot
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Add optional name and description fields to workflows in both Python and .NET implementations, matching the existing agent API pattern.
Python changes:
- Add name/description parameters to WorkflowBuilder.__init__
- Add name/description attributes to Workflow class
- Include name/description in to_dict() serialization
- Add WORKFLOW_NAME and WORKFLOW_DESCRIPTION OTEL attributes
- Add tests in test_serialization.py and test_workflow_observability.py
.NET changes:
- Add Name and Description properties to Workflow and Workflow<T>
- Add WithName() and WithDescription() fluent methods to WorkflowBuilder
- Add WorkflowName and WorkflowDescription OTEL tags
- Add test in WorkflowBuilderSmokeTests.cs
This enables applications like DevUI to display human-readable workflow names (e.g., 'Data Processing Pipeline') instead of auto-generated UUIDs (e.g., 'Workflow 50fdd917').
Fixes: #1181
* support for local function approval
* small fix
* fix mypy
* added bigger test scenario's for function calling and approvals
* updated lock
* updated return message for rejection
* fix test
* updated function result content handling
Should be `Agent2Agent Protocol` not `Agent-to-Agent` unless talking about general agent to agent communication
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
* feat: Add support for Workflow-as-Executor
* Fixes routing of 'object' compile-typed variables to properly take in type information
* Fixes a concurrency issue in StepTracer
* fix: Make Subworkflow ExternalRequests work properly
* fix: Threading and Concurrency fixes; prep for OffThread Mode
* refactor: Remove dead code around OffStreamRunEventStream
Currently not used, and will be replaced with a rewrite when brought back, so having it in the change is not valuable.
* ci: Work around issues with dotnet-format not properly analyzing the source
* fix: Fix the logic of AsyncCoordinator and AsyncBarrier
* Prevent individual wait cancellations from canceling the entire barrier
* Propagate information about whether the wait was completed or cancelled, and whether any waiters were present when released
* fix: Remove superfluous acces to .Keys in InProcStepTracer
* refactor: Clean up AsyncCoordinator's use of AsyncBarrier
* Ignore null, timeout, status codes
* Only run when a markdown file is changed and run nightly
* Add markdown change
* Add run on link checker change and arbitrary change
* updated docstrings of all _files
* fix mypy
* fixed codeblocks in workflows and some other files
---------
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
* Move the AsAIAgent extension methods to the correct class
* Fix format issue
* Disable unit test, see issue #1109
---------
Co-authored-by: Mark Wallace <markwallace@microsoft.com>
* Re-enable ImplicitUsings in samples and clean up NoWarns
* Fix dotnet format
* More dotnet format
* More dotnet format
---------
Co-authored-by: Chris <66376200+crickman@users.noreply.github.com>
We shouldn't be shipping any more dependencies on this package, as it's becoming legacy, replaced by System.Linq.AsyncEnumerable.
We'll eventually want to replace it with System.Linq.AsyncEnumerable in all tests/samples, too, but that's hard to do until SK updates to use S.L.AsyncEnumerable once its 10.0.0 version is released. I did remove the package reference from tests/samples where it's not needed.
All python code resides under the `python/` directory.
All C# code resides under the `dotnet/` directory.
The purpose of the code is to provide a framework for building AI agents.
When contributing to this repository, please follow these guidelines:
## C# Code Guidelines
Here are some general guidelines that apply to all code.
- The top of all *.cs files should have a copyright notice: `// Copyright (c) Microsoft. All rights reserved.`
- All public methods and classes should have XML documentation comments.
### C# Sample Code Guidelines
Sample code is located in the `dotnet/samples` directory.
When adding a new sample, follow these steps:
- The sample should be a standalone .net project in one of the subdirectories of the samples directory.
- The directory name should be the same as the project name.
- The directory should contain a README.md file that explains what the sample does and how to run it.
- The README.md file should follow the same format as other samples.
- The csproj file should match the directory name.
- The csproj file should be configured in the same way as other samples.
- The project should preferably contain a single Program.cs file that contains all the sample code.
- The sample should be added to the solution file in the samples directory.
- The sample should be tested to ensure it works as expected.
- A reference to the new samples should be added to the README.md file in the parent directory of the new sample.
The sample code should follow these guidelines:
- Configuration settings should be read from environment variables, e.g. `var endpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT") ?? throw new InvalidOperationException("AZURE_OPENAI_ENDPOINT is not set.");`.
- Environment variables should use upper snake_case naming convention.
- Secrets should not be hardcoded in the code or committed to the repository.
- The code should be well-documented with comments explaining the purpose of each step.
- The code should be simple and to the point, avoiding unnecessary complexity.
- Prefer inline literals over constants for values that are not reused. For example, use `new ChatClientAgent(chatClient, instructions: "You are a helpful assistant.")` instead of defining a constant for "instructions".
- Ensure that all private classes are sealed
- Use the Async suffix on the name of all async methods that return a Task or ValueTask.
- Prefer defining variables using types rather than var, to help users understand the types involved.
- Follow the patterns in the samples in the same directories where new samples are being added.
- The structure of the sample should be as follows:
- The top of the Program.cs should have a copyright notice: `// Copyright (c) Microsoft. All rights reserved.`
- Then add a comment describing what the sample is demonstrating.
- Then add the necessary using statements.
- Then add the main code logic.
- Finally, add any helper methods or classes at the bottom of the file.
### C# Unit Test Guidelines
Unit tests are located in the `dotnet/tests` directory in projects with a `.UnitTests.csproj` suffix.
Unit tests should follow these guidelines:
- Use `this.` for accessing class members
- Add Arrange, Act and Assert comments for each test
- Ensure that all private classes, that are not subclassed, are sealed
- Use the Async suffix on the name of all async methods
- Use the Moq library for mocking objects where possible
- Validate that each test actually tests the target behavior, e.g. we should not have tests that creates a mock, calls the mock and then verifies that the mock was called, without the target code being involved. We also shouldn't have tests that test language features, e.g. something that the compiler would catch anyway.
- Avoid adding excessive comments to tests. Instead favour clear easy to understand code.
- Follow the patterns in the unit tests in the same project or classes to which new tests are being added
- Each pull request that modifies code should add just one bulleted entry to the `CHANGELOG.md` file containing a change title (usually the PR title) and a link to the PR itself.
- New PRs should be added to the top of the `CHANGELOG.md` file under a "## [Unreleased]" heading.
- If the PR is the first since the last release, the existing "## [Unreleased]" heading should be replaced with a "## v[X.Y.Z]" heading and the PRs since the last release should be added to the new "## [Unreleased]" heading.
- The style of new `CHANGELOG.md` entries should match the style of the other entries in the file.
- If the PR introduces a breaking change, the changelog entry should be prefixed with "[BREAKING]".
Below are some ways that you can get involved in the Agent Framework Community.
## Engage on GitHub
- [Discussions](https://github.com/microsoft/agent-framework/discussions): Ask questions, provide feedback and ideas to what you'd like to see from the Agent Framework.
- [Issues](https://github.com/microsoft/agent-framework/issues) - If you find a bug, unexpected behavior or have a feature request, please open an issue.
- [Pull Requests](https://github.com/microsoft/agent-framework/pulls) - We welcome contributions! Please see our [Contributing Guide](https://github.com/microsoft/agent-framework/blob/main/CONTRIBUTING.md)
We do our best to respond to each submission.
## Public Community Office Hours
We regularly have Community Office Hours that are open to the **public** to join.
Add Agent Framework events to your calendar. We are running two community calls to accommodate different time zones for Q&A Office Hours:
- **Americas & EMEA timezone:** Every Wednesday at 8:00 AM Pacific Time/17:00 CET. Adjusted for daylight savings. Join here: [AF-AG-SK-Americas-Europe-OfficeHours](https://aka.ms/sk-officehours).
- **Asia Pacific timezone:** The second Wednesday of every month at 4:00 PM Pacific Time Wednesday. In much of Asia this occurs on Thursday local time. Adjusted for daylight savings. Join here: [AF-AG-SK-APAC-OfficeHours](https://aka.ms/sk-apac-officehours).
If you are unable to make it live, all meetings will be recorded and posted online.
Welcome to Microsoft's comprehensive multi-language framework for building, orchestrating, and deploying AI agents with support for both .NET and Python implementations. This framework provides everything from simple chat agents to complex multi-agent workflows with graph-based orchestration.
@@ -19,20 +23,37 @@ Welcome to Microsoft's comprehensive multi-language framework for building, orch
# This will install all sub-packages, see `python/packages` for individual packages.
# It may take a minute on first install on Windows.
```
- **[Quick Start Guide](https://learn.microsoft.com/agent-framework/tutorials/quick-start)** - Simple getting started instructions
.NET
```bash
dotnet add package Microsoft.Agents.AI
```
### 📚 Documentation
- **[Overview](https://learn.microsoft.com/agent-framework/overview/agent-framework-overview)** - High level overview of the framework
- **[Quick Start](https://learn.microsoft.com/agent-framework/tutorials/quick-start)** - Get started with a simple agent
- **[Tutorials](https://learn.microsoft.com/agent-framework/tutorials/overview)** - Step by step tutorials
- **[User Guide](https://learn.microsoft.com/en-us/agent-framework/user-guide/overview)** - In-depth user guide for building agents and workflows
- **[Migration from Semantic Kernel](https://learn.microsoft.com/en-us/agent-framework/migration-guide/from-semantic-kernel)** - Guide to migrate from Semantic Kernel
- **[Migration from AutoGen](https://learn.microsoft.com/en-us/agent-framework/migration-guide/from-autogen)** - Guide to migrate from AutoGen
Still have questions? Join our [weekly office hours](./COMMUNITY.md#public-community-office-hours) or ask questions in our [Discord channel](https://discord.gg/b5zjErwbQM) to get help from the team and other users.
### ✨ **Highlights**
- **Graph-based Workflows**: Connect agents and deterministic functions using data flows with streaming, checkpointing, human-in-the-loop, and time-travel capabilities
If you use the Microsoft Agent Framework to build applications that operate with third-party servers or agents, you do so at your own risk. We recommend reviewing all data being shared with third-party servers or agents and being cognizant of third-party practices for retention and location of data. It is your responsibility to manage whether your data will flow outside of your organization's Azure compliance and geographic boundaries and any related implications.
For help and questions about using this project, please create a GitHub issue.
AI Support team will support Microsoft Agent Framework issues for customers under a **Unified support agreement when the issue arises from usage of Azure AI services** (Foundry Models, Foundry Agents etc.) in conjunction with the SDK. Conversely, if customer has any other / non unified support agreement and/or Agent Framework SDK is used in a way **not involving an Azure service**, it is treated as a purely open-source tool – Microsoft’s support organization will not handle it, and users should use GitHub or forums for assistance
For Copilot Studio SDK implementation issues, customers should use GitHub Issues for assistance, as outlined above. Conversely, for prerequisites managed within the Copilot Studio portal, customers can rely on the standard Microsoft Copilot Studio support channels.
## Microsoft Support Policy
Support for this **PROJECT or PRODUCT** is limited to the resources listed above.
Microsoft Agent Framework is a comprehensive multi-language (C#/.NET and Python) framework for building, orchestrating, and deploying AI agents and multi-agent workflows. The system takes user instructions and conversation inputs and produces intelligent responses through AI agents that can integrate with various LLM providers (OpenAI, Azure OpenAI, Azure AI Foundry). It provides both simple chat agents and complex multi-agent workflows with graph-based orchestration.
@@ -40,9 +42,9 @@ Microsoft Agent Framework relies on existing LLMs. Using the framework retains c
- **API Dependencies**: Requires proper configuration of LLM provider keys and endpoints
- **Orchestration Features**: Advanced orchestration patterns like GroupChat, Sequential, and Concurrent orchestrations are "coming soon" for Python implementation
- **Orchestration Features**: Advanced orchestration patterns including GroupChat, Sequential, and Concurrent workflows are now available in both Python and .NET implementations. See the respective language documentation for examples.
- **Privacy and Data Protection**: The framework allows for human participation in conversations between agents. It is important to ensure that user data and conversations are protected and that developers use appropriate measures to safeguard privacy.
- **Accountability and Transparency**: The framework involves multiple agents conversing and collaborating, it is important to establish clear accountability and transparency mechanisms. Users should be able to understand and trace the decision-making process of the agents involved in order to ensure accountability and address any potential issues or biases.
- **Security & unintended consequences**: The use of multi-agent conversations and automation in complex tasks may have unintended consequences. Especially, allowing agents to make changes in external environments through tool calls or function execution could pose significant risks. Developers should carefully consider the potential risks and ensure that appropriate safeguards are in place to prevent harm or negative outcomes, including keeping a human in the loop for decision making.
@@ -95,7 +97,7 @@ Microsoft Agent Framework relies on existing LLMs. Using the framework retains c
The framework supports multiple external service types:
- **Native Functions**: Custom Python/C# functions that agents can invoke
- **A2A Integration**: Agent-to-agent communication and coordination
- **A2A (Agent2Agent)Integration**: Agent-to-agent communication and coordination
- **Model Context Protocol (MCP)**: External tools and data sources through MCP servers
@@ -108,7 +110,7 @@ Microsoft Agent Framework is an open-source framework that allows integration wi
**Data Access by Service Type**:
- **Native Functions**: Custom functions you develop have access to whatever data you explicitly pass to them as parameters
- **A2A (Agent-to-Agent)**: External agents can access conversation history, messages, and any data you configure to share through the communication interface
- **A2A (Agent2Agent)**: External agents can access conversation history, messages, and any data you configure to share through the communication interface
- **Model Context Protocol (MCP) Servers**: External MCP servers can access data according to the specific MCP server implementation and your configuration
- **External Tools**: Third-party tools and APIs have access to data you explicitly send to them through function calls
@@ -136,4 +138,4 @@ Microsoft Agent Framework is an open-source framework that allows integration wi
- Implement proper error handling for tool failures
- Use strong typing and compatibility validation
- Monitor external service health and implement fallback strategies
- Regular repository updates during preview period for bug fixes
- Regular repository updates during preview period for bug fixes
This folder contains sample agent definitions than be ran using the declarative agent support, for python see the [declarative agent python sample folder](../python/samples/getting_started/declarative/).
instructions:You are a helpful assistant. You answer questions is the language specified by the user. You return your answers in a JSON format. You must include Chat as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Assistants as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Chat as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Responses as the type in your response.
instructions:You are a helpful assistant. You answer questions is the language specified by the user. You return your answers in a JSON format. You must include Chat as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Assistants as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Chat as the type in your response.
instructions:You are a helpful assistant. You answer questions in the language specified by the user. You return your answers in a JSON format. You must include Responses as the type in your response.
To allow moving the docs to mslearn later, we are using language pivots as supported with mslearn markdown files.
This means that to make the docs easier to understand for users, we have a [PowerShell script](./generate-language-specific-docs.ps1) that generates language specific versions of the docs in separate folders.
The script strips out any pivots that are for a different language to the target.
Therefore, write your docs in the [docs-templates](./docs-templates/) folder and then
generate the language-specific versions by just running the PowerShell script.
```powershell
.\generate-language-specific-docs.ps1
```
## Using pivots
To have language-specific content, use the `::: zone pivot` syntax in your markdown file.
Note that when using a pivot you always have to have a section for both languages (csharp and python).
@@ -499,7 +499,7 @@ We need to decide what AIContent types, each agent response type will be mapped
| AWS (Strands) | **Approach 2** Supports a special invocation method called [structured_output](https://strandsagents.com/latest/api-reference/agent/#strands.agent.agent.Agent.structured_output) |
| LangGraph | **Approach 1** Supports [configuring an agent](https://langchain-ai.github.io/langgraph/agents/agents/?h=structured#6-configure-structured-output) at agent construction time, and a [structured response](https://langchain-ai.github.io/langgraph/agents/run_agents/#output-format) can be retrieved as a special property on the agent response |
| Agno | **Approach 1** Supports [configuring an agent](https://docs.agno.com/examples/getting-started/structured-output) at agent construction time |
| A2A | **Informal Approach 2** Doesn't formally support schema negotiation, but [hints can be provided via metadata](https://a2aproject.github.io/A2A/v0.2.5/specification/#97-structured-data-exchange-requesting-and-providing-json) at invocation time |
| A2A | **Informal Approach 2** Doesn't formally support schema negotiation, but [hints can be provided via metadata](https://a2a-protocol.org/latest/specification/#97-structured-data-exchange-requesting-and-providing-json) at invocation time |
| Protocol Activity | Supports returning [Complex types](https://github.com/microsoft/Agents/blob/main/specs/activity/protocol-activity.md#complex-types) but no support for requesting a type |
### Response Reason Support
@@ -511,5 +511,5 @@ We need to decide what AIContent types, each agent response type will be mapped
| AWS (Strands) | Exposes a [stop_reason](https://strandsagents.com/latest/api-reference/types/#strands.types.event_loop.StopReason) property on the [AgentResult](https://strandsagents.com/latest/api-reference/agent/#strands.agent.agent_result.AgentResult) class with options that are tied closely to LLM operations. |
| LangGraph | No equivalent present, output contains only [messages](https://langchain-ai.github.io/langgraph/agents/run_agents/#output-format) |
| A2A | No equivalent present, response only contains a [message](https://a2aproject.github.io/A2A/v0.2.5/specification/#64-message-object) or [task](https://a2aproject.github.io/A2A/v0.2.5/specification/#61-task-object). |
| A2A | No equivalent present, response only contains a [message](https://a2a-protocol.org/latest/specification/#64-message-object) or [task](https://a2a-protocol.org/latest/specification/#61-task-object). |
@@ -22,7 +22,6 @@ This document aims to provide options and capture the decision on how to model t
See various features that would need to be supported via this type of mechanism, plus how various other frameworks support this:
- Also see [dotnet issue 6492](https://github.com/dotnet/extensions/issues/6492), which discusses the need for a similar pattern in the context of MCP approvals.
- Also see [the openai RunToolApprovalItem](https://openai.github.io/openai-agents-js/openai/agents/classes/runtoolapprovalitem/).
- Also see [the openai human-in-the-loop guide](https://openai.github.io/openai-agents-js/guides/human-in-the-loop/#approval-requests).
- Also see [the openai MCP guide](https://openai.github.io/openai-agents-js/guides/mcp/#optional-approval-flow).
- Also see [MCP Approval Requests from OpenAI](https://platform.openai.com/docs/guides/tools-remote-mcp#approvals).
@@ -47,9 +47,9 @@ This section provides an analysis of how other major AI agent frameworks handle
| Haystack | Python | N (Pipeline-based interception) | N/A (Pipeline Components/Routers) | Relies on modular pipelines for implicit interception but lacks explicit middleware/filters; custom components can read/write data flow via routing/transformations, but this is compositional rather than hook-based interception. [Details](#haystack) |
| OpenAI Swarm | Python | N | N/A | No explicit middleware/filters; interception requires custom wrappers or manual handling (e.g., function decorators, client subclassing), lacking native framework support for built-in components to accept such modifications. [Details](#openai-swarm) |
| Atomic Agents | Python | N | N/A (Composable Components) | No explicit middleware/filters; modularity allows composable units but no dedicated interception hooks or callbacks for custom reading/modification mid-execution. [Details](#atomic-agents) |
| Smolagents (Hugging Face)| Python | N | N/A | No explicit support; focuses on simple agent building without interception mechanisms or hooks for reading/modifying execution. [Details](#smolagents) |
| Phidata (Agno) | Python | N | N/A | No explicit middleware/filters; agents use tools/memory but no interception hooks for custom reading/modification of calls. [Details](#phidata) |
| PromptFlow (Microsoft) | Python | N (Tracing only) | Tracing | Supports tracing for LLM interactions, acting as callbacks for debugging/iteration; tracing is read-only for observability/telemetry without options to modify context or intercept calls beyond logging. [Details](#promptflow) |
| Smolagents (Hugging Face)| Python | N | N/A | No explicit support; focuses on simple agent building without interception mechanisms or hooks for reading/modifying execution. [Details](#smolagents-hugging-face) |
| Phidata (Agno) | Python | N | N/A | No explicit middleware/filters; agents use tools/memory but no interception hooks for custom reading/modification of calls. [Details](#phidata-agno) |
| PromptFlow (Microsoft) | Python | N (Tracing only) | Tracing | Supports tracing for LLM interactions, acting as callbacks for debugging/iteration; tracing is read-only for observability/telemetry without options to modify context or intercept calls beyond logging. [Details](#promptflow-microsoft) |
| n8n | JS/TS | Y (read/write) | Callbacks (inherited from LangChain) | AI Agent node uses LangChain under the hood, inheriting callbacks for observability; supports reading/modifying metadata or interrupting flow as in LangChain. [Details](#n8n) |
The .NET Agent Framework needed a standardized way to enable communication between AI agents and user-facing applications with support for streaming, real-time updates, and bidirectional communication. Without AG-UI protocol support, .NET agents could not interoperate with the growing ecosystem of AG-UI-compatible frontends and agent frameworks (LangGraph, CrewAI, Pydantic AI, etc.), limiting the framework's adoption and utility.
The AG-UI (Agent-User Interaction) protocol is an open, lightweight, event-based protocol that addresses key challenges in agentic applications including streaming support for long-running agents, event-driven architecture for nondeterministic behavior, and protocol interoperability that complements MCP (tool/context) and A2A (agent-to-agent) protocols.
## Decision Drivers
- Need for streaming communication between agents and client applications
- Requirement for protocol interoperability with other AI frameworks
- Support for long-running, multi-turn conversation sessions
- Real-time UI updates for nondeterministic agent behavior
- Standardized approach to agent-to-UI communication
- Framework abstraction to protect consumers from protocol changes
## Considered Options
1.**Implement AG-UI event types as public API surface** - Expose AG-UI event models directly to consumers
2.**Use custom AIContent types for lifecycle events** - Create new content types (RunStartedContent, RunFinishedContent, RunErrorContent)
3.**Current approach** - Internal event types with framework-native abstractions
## Decision Outcome
Chosen option: "Current approach with internal event types and framework-native abstractions", because it:
- Protects consumers from protocol changes by keeping AG-UI events internal
- Maintains framework abstractions through conversion at boundaries
- Uses existing framework types (AgentRunResponseUpdate, ChatMessage) for public API
- Text message events: `TextMessageStart`, `TextMessageContent`, `TextMessageEnd`
- Thread and run ID management via `ConversationId` and `ResponseId`
### Key Design Decisions
1.**Event Models as Internal Types** - AG-UI event types are internal with conversion via extension methods; public API uses the existing types in Microsoft.Extensions.AI as those are the abstractions people are familiar with
2.**No Custom Content Types** - Run lifecycle communicated through existing `ChatResponseUpdate` properties (`ConversationId`, `ResponseId`) and standard `ErrorContent` type
3.**Agent Factory Pattern** - `MapAGUIAgent` uses factory function `(messages) => AIAgent` to allow request-specific agent configuration supporting multi-tenancy
4.**Bidirectional Conversion Architecture** - Symmetric conversion logic in shared namespace compiled into both packages for server (`AgentRunResponseUpdate` → AG-UI events) and client (AG-UI events → `AgentRunResponseUpdate`)
5.**Thread Management** - `AGUIAgentThread` stores only `ThreadId` with thread ID communicated via `ConversationId`; applications manage persistence for parity with other implementations and to be compliant with the protocol. Future extensions will support having the server manage the conversation.
6.**Custom JSON Converter** - Uses custom polymorphic deserialization via `BaseEventJsonConverter` instead of built-in System.Text.Json support to handle AG-UI protocol's flexible discriminator positioning
### Consequences
**Positive:**
- .NET developers can consume AG-UI servers from any framework
- .NET agents accessible from any AG-UI-compatible client
- Standardized streaming communication patterns
- Protected from protocol changes through internal implementation
- Symmetric conversion logic between client and server
* this means developers can use `pip install agent-framework[google]` to get AF with all Google connectors and dependencies, as well as manually installing the subpackage with `pip install agent-framework-google`.
* this means developers can use `pip install agent-framework[google] --pre` to get AF with all Google connectors and dependencies, as well as manually installing the subpackage with `pip install agent-framework-google --pre`.
### Sample getting started code
```python
@@ -175,7 +175,7 @@ Sub-packages are comprised of two parts, the code itself and the dependencies, t
- Subpackage naming should also follow this, so in principle a package name is `<vendor/folder>-<feature/brand>`, so `google-gemini`, `azure-purview`, `microsoft-copilotstudio`, etc. For smaller vendors, where it's less likely to have a multitude of connectors, we can skip the feature/brand part, so `mem0`, `redis`, etc.
- For Microsoft services we will have two vendor folders, `azure` and `microsoft`, where `azure` contains all Azure services, while `microsoft` contains other Microsoft services, such as Copilot Studio Agents.
This setup was discussed at length and the decision is captured in [ADR-0007](../decisions/0007-python-subpackages.md).
This setup was discussed at length and the decision is captured in [ADR-0008](../decisions/0008-python-subpackages.md).
#### Evolving the package structure
For each of the advanced components, we have two reason why we may split them into a folder, with an `__init__.py` and optionally a `_files.py`:
The durable agents automatically maintain conversation history and state for each session. Without automatic cleanup, this state can accumulate indefinitely, consuming storage resources and increasing costs. The Time-To-Live (TTL) feature provides automatic cleanup of idle agent sessions, ensuring that sessions are automatically deleted after a period of inactivity.
## What is TTL?
Time-To-Live (TTL) is a configurable duration that determines how long an agent session state will be retained after its last interaction. When an agent session is idle (no messages sent to it) for longer than the TTL period, the session state is automatically deleted. Each new interaction with an agent resets the TTL timer, extending the session's lifetime.
## Benefits
- **Automatic cleanup**: No manual intervention required to clean up idle agent sessions
- **Cost optimization**: Reduces storage costs by automatically removing unused session state
- **Resource management**: Prevents unbounded growth of agent session state in storage
- **Configurable**: Set TTL globally or per-agent type to match your application's needs
## Configuration
TTL can be configured at two levels:
1.**Global default TTL**: Applies to all agent sessions unless overridden
2.**Per-agent type TTL**: Overrides the global default for specific agent types
Additionally, you can configure a **minimum deletion delay** that controls how frequently deletion operations are scheduled. The default value is 5 minutes, and the maximum allowed value is also 5 minutes.
> [!NOTE]
> Reducing the minimum deletion delay below 5 minutes can be useful for testing or for ensuring rapid cleanup of short-lived agent sessions. However, this can also increase the load on the system and should be used with caution.
### Default values
- **Default TTL**: 14 days
- **Minimum TTL deletion delay**: 5 minutes (maximum allowed value, subject to change in future releases)
### Configuration examples
#### .NET
```csharp
// Configure global default TTL and minimum signal delay
services.ConfigureDurableAgents(
options=>
{
// Set global default TTL to 7 days
options.DefaultTimeToLive=TimeSpan.FromDays(7);
// Add agents (will use global default TTL)
options.AddAIAgent(myAgent);
});
// Configure per-agent TTL
services.ConfigureDurableAgents(
options=>
{
options.DefaultTimeToLive=TimeSpan.FromDays(14);// Global default
The following sections describe how TTL works in detail.
### Expiration tracking
Each agent session maintains an expiration timestamp in its internally managed state that is updated whenever the session processes a message:
1. When a message is sent to an agent session, the expiration time is set to `current time + TTL`
2. The runtime schedules a delete operation for the expiration time (subject to minimum delay constraints)
3. When the delete operation runs, if the current time is past the expiration time, the session state is deleted. Otherwise, the delete operation is rescheduled for the next expiration time.
### State deletion
When an agent session expires, its entire state is deleted, including:
- Conversation history
- Any custom state data
- Expiration timestamps
After deletion, if a message is sent to the same agent session, a new session is created with a fresh conversation history.
## Behavior examples
The following examples illustrate how TTL works in different scenarios.
### Example 1: Agent session expires after TTL
1. Agent configured with 30-day TTL
2. User sends message at Day 0 → agent session created, expiration set to Day 30
3. No further messages sent
4. At Day 30 → Agent session is deleted
5. User sends message at Day 31 → New agent session created with fresh conversation history
### Example 2: TTL reset on interaction
1. Agent configured with 30-day TTL
2. User sends message at Day 0 → agent session created, expiration set to Day 30
3. User sends message at Day 15 → Expiration reset to Day 45
4. User sends message at Day 40 → Expiration reset to Day 70
5. Agent session remains active as long as there are regular interactions
## Logging
The TTL feature includes comprehensive logging to track state changes:
- **Expiration time updated**: Logged when TTL expiration time is set or updated
- **Deletion scheduled**: Logged when a deletion check signal is scheduled
- **Deletion check**: Logged when a deletion check operation runs
- **Session expired**: Logged when an agent session is deleted due to expiration
- **TTL rescheduled**: Logged when a deletion signal is rescheduled
These logs help monitor TTL behavior and troubleshoot any issues.
## Best practices
1.**Choose appropriate TTL values**: Balance between storage costs and user experience. Too short TTLs may delete active sessions, while too long TTLs may accumulate unnecessary state.
2.**Use per-agent TTLs**: Different agents may have different usage patterns. Configure TTLs per-agent based on expected session lifetimes.
3.**Monitor expiration logs**: Review logs to understand TTL behavior and adjust configuration as needed.
4.**Test with short TTLs**: During development, use short TTLs (e.g., minutes) to verify TTL behavior without waiting for long periods.
## Limitations
- TTL is based on wall-clock time, not activity time. The expiration timer starts from the last message timestamp.
- Deletion checks are durably scheduled operations and may have slight delays depending on system load.
- Once an agent session is deleted, its conversation history cannot be recovered.
- TTL deletion requires at least one worker to be available to process the deletion operation message.
@@ -103,12 +103,12 @@ dotnet run --urls "http://localhost:5002;https://localhost:5012" --agentId "<Log
### Testing the Agents using the Rest Client
This sample contains a [.http file](https://learn.microsoft.com/aspnet/core/test/http-files?view=aspnetcore-9.0) which can be used to test the agent.
This sample contains a [.http file](https://learn.microsoft.com/aspnet/core/test/http-files?view=aspnetcore-10.0) which can be used to test the agent.
1. In Visual Studio open [./A2AServer/A2AServer.http](./A2AServer/A2AServer.http)
1. There are two sent requests for each agent, e.g., for the invoice agent:
1. Query agent card for the invoice agent
`GET {{hostInvoice}}/.well-known/agent.json`
`GET {{hostInvoice}}/.well-known/agent-card.json`
1. Send a message to the invoice agent
```
POST {{hostInvoice}}
@@ -121,6 +121,7 @@ This sample contains a [.http file](https://learn.microsoft.com/aspnet/core/test
"params": {
"id": "12345",
"message": {
"kind": "message",
"role": "user",
"messageId": "msg_1",
"parts": [
@@ -144,7 +145,7 @@ Sample output from the request to send a message to the agent via A2A protocol:
### Testing the Agents using the A2A Inspector
The A2A Inspector is a web-based tool designed to help developers inspect, debug, and validate servers that implement the Google A2A (Agent-to-Agent) protocol. It provides a user-friendly interface to interact with an A2A agent, view communication, and ensure specification compliance.
The A2A Inspector is a web-based tool designed to help developers inspect, debug, and validate servers that implement the Google A2A (Agent2Agent) protocol. It provides a user-friendly interface to interact with an A2A agent, view communication, and ensure specification compliance.
For more information go [here](https://github.com/a2aproject/a2a-inspector).
@@ -223,7 +224,7 @@ Agent: The transaction details for **TICKET-XYZ987** are as follows:
- **Hats:** 200 units at $15.00 each
- **Glasses:** 300 units at $5.00 each
To proceed with the dispute regarding the quantity of t-shirts delivered, please specify the exact quantity issue – how many t-shirts were actually received compared to the ordered amount.
To proceed with the dispute regarding the quantity of t-shirts delivered, please specify the exact quantity issue � how many t-shirts were actually received compared to the ordered amount.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.