Change8

python-v0.5.6

📦 autogenView on GitHub →
6 features🐛 7 fixes🔧 7 symbols

Summary

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.

Migration Steps

  1. If using GraphFlow, note that it is currently an experimental API and subject to change in future releases.

✨ New Features

  • Introduced GraphFlow, a new team class for customized workflows using directed graphs, supporting concurrent agents and fan-out/fan-in patterns.
  • Added DiGraphBuilder for constructing agent execution graphs.
  • Added support for Bing grounding citation URLs in Azure AI Agent.
  • Added Bedrock chat completion support for Anthropic models.
  • New multi-agent PostgreSQL data management sample.
  • Added more MCP workbench examples to the MCP API documentation.

🐛 Bug Fixes

  • Fixed DockerCommandLineCodeExecutor to be multi-event loop aware.
  • Fixed GraphFlow serialization and deserialization issues.
  • Fixed MultiModalMessage error when using Gemini with the OpenAI SDK.
  • Fixed McpWorkbench errors regarding properties and graceful shutdown.
  • Resolved workbench and tools conflict during AssistantAgent deserialization.
  • Fixed streaming chunks to ensure they are immediately flushed to the console.
  • Fixed vertical alignment of icons and corrected heading levels in documentation.

🔧 Affected Symbols

autogen_agentchat.teams.GraphFlowautogen_agentchat.teams.DiGraphBuilderautogen_agentchat.agents.AssistantAgentautogen_ext.models.openai.OpenAIChatCompletionClientDockerCommandLineCodeExecutorMultiModalMessageMcpWorkbench