pydantic-ai
https://github.com/pydantic/pydantic-ai
Python
GenAI Agent Framework, the Pydantic way
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported6 Subscribers
Add a CodeTriage badge to pydantic-ai
Help out
- Issues
- `AzureProvider` gives non-OpenAI Foundry models (Kimi, gpt-oss, Qwen, GLM, Phi) the OpenAI profile
- Vercel AI Gateway: reasoning signatures in `reasoning_details` are not round-tripped, so thinking continuity is lost across tool calls
- Add an `anthropic_inference_geo` setting for Anthropic data residency
- Add `openai_safety_identifier`, the replacement for the `user` field that `openai_user` sends
- Gemini on Vertex: stream tool-call arguments with `stream_function_call_arguments`
- Kimi K3 gets no Moonshot profile on Together, Fireworks, Vercel, Ollama and Cerebras, so `thinking` is dropped
- Popular open-weight families (Nemotron, Inkling, MiMo, Step, Hy3, Ling) have no model profile, so `thinking` is dropped off OpenRouter
- `OpenRouterModel` silently drops `CachePoint` on `openai/gpt-5.6*` routes instead of sending `prompt_cache_breakpoint`
- `OpenRouterModel` moves tool-result images into a separate user message, although every OpenRouter upstream tested reads them in place
- Cached-token usage isn't mapped for Cerebras and Hugging Face
- Docs
- Python not yet supported