Pydantic AI

From the team behind the Pydantic validation library, Pydantic AI is a Python agent framework that aims to bring a FastAPI-like developer experience to generative AI applications. Agents are fully type annotated, so IDEs and static type checkers catch mistakes early, and model responses are validated into structured Pydantic models, including streamed output validated as it arrives. It abstracts over a wide range of providers, supporting OpenAI, Anthropic, Gemini, DeepSeek, and dozens of others plus local models through Ollama, and adds tool registration via decorators, a type-safe dependency injection system, Model Context Protocol integration, and composable built-ins such as web search. Production concerns are addressed through durable execution that preserves agent progress across failures and restarts, an evals framework for systematic testing, and native tracing and cost tracking through Pydantic Logfire. The framework is open source under the MIT license and is used by Python teams building agents, chatbots, and structured extraction pipelines.