pydantic-ai

Why we included this project

Pydantic AI is the agent framework the Pydantic team built after they hit the limits of existing LLM libraries while working on Logfire. It gives Python developers a single typed Agent interface over many model providers, so moving from OpenAI to Anthropic or Ollama doesn't mean rewriting your application logic. Responses are validated with Pydantic as they stream in, which catches malformed output early instead of letting it surface as runtime bugs. The framework also handles durable execution for long-running and human-in-the-loop workflows, graph-based control flow, and MCP tools and web search. If your team already relies on Pydantic for data validation, the design will feel familiar, and the Logfire and OpenTelemetry hooks make it easy to trace and evaluate agent behavior in production.