Pydantic AI
frameworkTrial
Pydantic AI is the agent framework of the Pydantic team: agents are typed Python objects, tool signatures and structured outputs are validated by Pydantic - type-safe and well testable, with a leaner surface than the LangChain ecosystem.
- V2 stable since June 2026: "capabilities" as the core primitive - a composable unit bundling an agent's tools, hooks, instructions and model settings (announcement).
- The built-in harness ships memory systems, guardrails, context management and CodeMode (the model writes Python that orchestrates tools in one execution instead of many individual tool calls).
- Model-agnostic; natural fit for teams already using Pydantic in their stack.