AutoGen

AutoGen 0.7.5 introduces thinking mode for Anthropic, reasoning effort for OpenAI, and linear memory for Redis. It also defaults to Docker-based code execution for improved security and includes numerous fixes for streaming and graph state management.

Version 0.7.4 focuses on bug fixes for Redis integration, specifically regarding deserialization and streaming, alongside documentation updates for the agent-as-tool feature.

Release 0.7.3 introduces support for GPT-5 model info, improves Pydantic model typing for anyOf/oneOf, and fixes serialization issues in RedisStore and OpenAIAgent tool schemas.

Release 0.7.2 introduces Redis memory enhancements for JSON/Markdown, adds execution approval functions, and migrates MagenticOne to Docker-based code execution by default while cleaning up OpenAIAgent methods.

This release introduces RedisMemory, supports nested Teams in GroupChats, and enables all built-in tools for OpenAIAgent. It also includes significant upgrades for MCP and GraphRAG integrations alongside various stability fixes.

This release improves GraphFlow state persistence and termination logic, adds tool override capabilities to Workbench implementations, and expands model support for Claude and Qwen2.5-VL.

This release introduces streaming tool support, inner tool calling loops for AssistantAgent, and OpenTelemetry GenAI tracing. It also adds a Mem0 memory extension and several improvements to workflow handling and model compatibility.

This release focuses on bug fixes for GraphFlow validation and cycle detection, while enhancing ToolCallSummaryMessage with detailed function call results.

This release introduces concurrent agent execution in GraphFlow and a new OpenAIAgent. It also features experimental callable edge conditions and various improvements to model clients and code executors.

This release unifies the Azure AI Search Tool API, introduces model context support for SelectorGroupChat speaker selection, and improves OTEL tracing and Agent Runtime registration.

This release introduces GraphFlow for graph-based agent orchestration and concurrent execution, adds Bedrock support for Anthropic models, and includes several fixes for Docker executors and serialization.

This release introduces the Workbench API and McpWorkbench for stateful tool management, adds a FunctionalTermination condition, and provides new integration examples for FastAPI and MCP servers.

This release introduces Agent and Team tools for nested agent workflows, an Azure AI Agent adapter, and experimental Canvas Memory. It also enhances CodeExecutorAgent with self-debugging capabilities and improves SelectorGroupChat compatibility with streaming models.

This release introduces code generation capabilities to CodeExecutorAgent, improves AssistantAgent serialization, and adds shared session support for MCP tools. It also includes bug fixes for Azure AI Search and handoff context management.

This release introduces support for Gemini 2.5 Pro Preview, improves Task-Centric Memory accessibility, and provides several bug fixes for ChromaDB, Azure AI Search, and Docker code execution.

This release introduces structured output support for agents and model clients, refactors AgentChat message types into a class hierarchy for better extensibility, and adds a new Azure AI Search tool.

This patch release fixes a bug in the MCP Server Tool regarding null argument handling and introduces default timeouts and improved error messaging.

This release introduces major UI and observability enhancements to AutoGen Studio, including component validation, Anthropic support, token streaming, and experimental multi-user authentication.

This patch release focuses on security hardening by preventing API key exposure during serialization and fixes bugs in FileSurfer, SKChatCompletionAdapter, and reflection logic.

Release v0.4.9 introduces native Anthropic and LlamaCpp support, experimental task-centric memory, and a breaking change to the serialized state schema to improve state portability.

This patch release fixes a token limit issue in Azure AI streaming and introduces a close() method to model clients to resolve resource leaks and transport warnings.

This patch release fixes a critical bug in the SKChatCompletionAdapter that prevented the use of tools.

This release introduces the Ollama Chat Completion Client and adds support for model 'thought' processes and custom metadata in messages. It also includes breaking changes to FunctionExecutionResult requirements and makes agent exceptions fatal to improve debugging.

This release enforces strict validation for ModelInfo fields and introduces 'strict' mode for tools to support structured outputs. It also enhances the Docker code executor and provides various bug fixes for agent instantiation and tool execution.

This release introduces built-in support for MCP and HTTP tools, enhances MagenticOne and SelectorGroupChat for smaller/text-only models, and simplifies Gemini model integration.

AutoGen Studio v0.4.1 introduces a unified declarative configuration system allowing Python-defined agents to be exported to JSON, alongside new real-time testing tools and specialized research agent teams.

This release introduces token streaming for AgentChat, support for R1-style reasoning 'thought' blocks, and the ability to create FunctionTools from partial functions.

This release introduces full serialization support for AgentChat configurations and states, enabling persistent and portable agent sessions. It also adds a new Azure AI inference client for broader model support and a rich UI for the Magentic One CLI.

This release introduces significant features including model client caching, GraphRAG integration, Semantic Kernel adapters, and a new agent memory interface. It also expands declarative configuration support and brings back the Jupyter code executor.

Version 0.4.2 updates the async input strategy specifically to remove an accidentally included GPL-licensed dependency.

This release fixes critical console input and stop reason bugs, introduces subclassable BaseComponent for better serialization, and disables inaccurate console statistics by default.

This release marks the first stable version of AutoGen v0.4, introducing a new API architecture, support for OpenAI's o1 model, and breaking changes to Azure authentication and intervention handlers.