●Change8

python-v0.4.7

Breaking Changes
📦 autogenView on GitHub →
⚠ 1 breaking✨ 4 features🐛 4 fixes🔧 10 symbols

Summary

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.

⚠️ Breaking Changes

  • ModelInfo's required fields (vision, function_calling, json_output, family) are now strictly enforced. You must include all required fields when passing model_info to model clients or a validation error will occur.

Migration Steps

  1. Update all OpenAIChatCompletionClient (and other model client) instantiations to include a complete model_info dictionary with 'vision', 'function_calling', 'json_output', and 'family' keys.
  2. If running on Windows, review the updated event loop policy documentation to ensure subprocesses for web surfer or local executors function correctly.

✨ New Features

  • DockerCommandLineCodeExecutor now supports additional volume mounts and exposed host ports.
  • Added remove and get subscription APIs for Python GrpcWorkerAgentRuntime.
  • Added 'strict' mode support to BaseTool, ToolSchema, and FunctionTool for compatibility with structured output modes.
  • Made CodeExecutor components serializable.

🐛 Bug Fixes

  • Fixed tool call execution for scenarios where models produce empty tool call IDs.
  • Fixed a ValueError in register_factory when the expected class is not provided.
  • Added model info validation and improved error messaging.
  • Added warning and documentation for Windows event loop policy to prevent subprocess issues in web surfer and local executor.

🔧 Affected Symbols

ModelInfoOpenAIChatCompletionClientDockerCommandLineCodeExecutorGrpcWorkerAgentRuntimeBaseToolToolSchemaFunctionToolAgentInstantiationContextregister_factoryCodeExecutor