[Release] openai/openai-agents-python v0.13.1: v0.13.1

OpenAI Agents Python v0.13.1 introduces support for the any-llm adapter, enabling flexible multi-LLM routing with responses-compatible integration. This release includes critical bug fixes for realtime response handling, function tool conversion, and MCP metadata preservation. The update enhances stability and extends model provider compatibility while improving example robustness.

Key Points

• •New any-llm adapter support enables routing to multiple LLM providers with responses-compatible architecture
• •Fixed realtime response handling to properly wait for response.done before initiating follow-up responses
• •Preserved static MCP metadata during function tool conversion to maintain tool integrity
• •Improved error handling for cancelled single function tools by treating them as tool failures
• •Optionized initialized notification tolerance for more flexible configuration
• •Removed duplicate CompactionItem from RunItem union type to fix type definition issues
• •Hardened example auto-runs against PATH and port conflicts for more reliable testing
• •Updated documentation with 0.13 changelog and translated document pages
• •Multi-LLM provider examples available in examples/model_providers directory
• •Enhanced stability for production agent deployments with critical bug fixes

Workflow Diagram

Concepts