Skip to content
Agents tracked: 258 Downloads (7d): 219M up 6.1% GitHub stars: 5.5M VS Code installs: 148M Releases (7d): 327 Agent status: 1 with issues Updated Oct 7, 2026
AgentGid

AutoGen changelog: what's new each month

Every stable AutoGen release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.

September 2025

1 release: 0.7.5

September brought improvements to streaming support, memory capabilities, and thinking mode features across multiple AI providers. Several stability fixes addressed edge cases in message handling and reasoning content.

Highlights

  • Added thinking mode support for Anthropic client
  • Implemented linear memory support in RedisMemory
  • Fixed streaming Bedrock responses with tool usage
  • Resolved spurious closing tags in streaming reasoning content
  • Improved GraphFlow cycle detection and recursion cleanup

New

  • Thinking mode support for Anthropic client
  • Linear memory option in RedisMemory

Improved

  • Message ID correlation between streaming chunks and final messages
  • GraphFlow cycle detection now properly cleans up recursion state
  • Extra args handling to disable thinking mode

Fixed

  • Loading streaming Bedrock responses with tool usage and empty arguments
  • Spurious closing think tags caused by empty string reasoning content in streaming
  • Extra args not working to disable thinking mode
  • Message ID correlation issues between streaming chunks

August 2025

3 releases: 0.7.2 → 0.7.4

August brought refinements to agent execution and tooling capabilities, with improvements to Docker integration, code execution controls, and storage reliability.

Highlights

  • Docker is now the default code executor for MagenticOne teams.
  • Added approval function option to CodeExecutorAgent for controlling code execution.
  • Fixed critical Redis serialization and deserialization issues.
  • Expanded Pydantic model typing support for complex schema definitions.
  • OpenAI function tool schema corrected for proper integration.

New

  • Parallel tool call support option in OpenAI model client configuration.
  • Model info added for GPT-5.
  • Approval function capability for CodeExecutorAgent to gate code execution.

Improved

  • Redis storage now properly serializes complex objects.
  • Pydantic model capability extended to handle anyOf and oneOf item typing.
  • OpenAIAssistantAgent documentation updated.
  • Documentation warnings added for AgentTool and TeamTool parallel call limitations.

Fixed

  • Redis deserialization errors resolved.
  • OpenAIAgent function tool schema corrected.
  • MCP example in readme fixed.
  • Documentation typo in NoOpTracerProvider corrected.

July 2025

3 releases: 0.6.2 → 0.7.1

July brought significant enhancements to streaming capabilities, tool handling, and memory management across AutoGen. The month focused on improving agent workflows, tracing infrastructure, and expanding integration options.

Highlights

  • Introduced streaming support for tools including AgentTool and TeamTool with new toolchoice parameter.
  • Added OpenTelemetry GenAI tracing for agents and tools to improve observability.
  • Expanded memory options with Mem0 and Redis implementations.
  • Fixed GraphFlow to support multiple task executions without requiring manual resets.

New

  • Streaming tool support with AgentTool and TeamTool streaming capabilities.
  • toolchoice parameter for ChatCompletionClient create and createstream methods.
  • OpenTelemetry GenAI tracing for Agent and Tool operations.
  • Mem0 memory implementation.
  • Redis memory extension class.
  • Tool name and description override functionality in Workbench implementations.
  • Support for Teams as participants in GroupChat.
  • includenameinmessage parameter to make name field optional in OpenAI messages.

Improved

  • GraphFlow now supports multiple task execution without explicit reset.
  • Activation groups added for workflows with multiple cycles.
  • MCP Workbench expanded to support more MCP Client features.
  • GraphRAG upgraded to v2.3+.
  • Function calling support improved for Llama3.3.
  • Runtime tracing can now be disabled via environment variable.

Fixed

  • Output task messages handling.
  • GraphFlowManager termination to prevent StopAgent from polluting conversation context.
  • Function calling support for Llama3.3.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. AutoGen: pricing, features and alternatives · All changelogs