●Change8

python-v0.4.9

Breaking Changes
📦 autogenView on GitHub →
⚠ 1 breaking✨ 8 features🐛 3 fixes🔧 9 symbols

Summary

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.

⚠️ Breaking Changes

  • Serialized state schema change: The team state now uses the agent name as the key instead of the agent ID, and the team_id field has been removed. States saved with the old format may be incompatible with future versions.

Migration Steps

  1. Update autogen-ext with specific extras: pip install -U "autogen-ext[anthropic]" or pip install -U "autogen-ext[llama-cpp]".
  2. Use the provided migration scripts at https://github.com/ekzhu/autogen-migration/ to convert old serialized states to the new name-based key format.

✨ New Features

  • Native support for Anthropic models via AnthropicChatCompletionClient.
  • Native support for LlamaCpp models via LlamaCppChatCompletionClient, supporting local GGUF files and Hugging Face models.
  • Experimental Task-Centric Memory module (Teachability) for long-term learning and self-improvement.
  • Experimental 'Gitty' sample application for automated GitHub issue replies.
  • Added LLMStreamStartEvent and LLMStreamEndEvent for improved tracing.
  • Added ToolCallEvent logging for all builtin tools.
  • PowerShell support for LocalCommandLineCodeExecutor.
  • Significant accessibility improvements to the documentation website, including high contrast mode and screen reader enhancements.

🐛 Bug Fixes

  • Fixed save_state to no longer require the team to be stopped.
  • Fixed span structure for tracing.
  • Fixed high contrast mode focus and keyboard copy events on the documentation site.

🔧 Affected Symbols

AnthropicChatCompletionClientLlamaCppChatCompletionClientMemoryControllerTeachabilityLocalCommandLineCodeExecutorLLMStreamStartEventLLMStreamEndEventToolCallEventAssistantAgent