pydantic-ai
#33 en Frameworks d'agentsPydantic · 2× · vu le 05 août 2026
23
Momentum
Pydantic AI is an Open Source agent framework for Python developed by the Pydantic team (Pydantic Services Inc.) that applies Pydantic's validation philosophy to building GenAI agents. It offers a model-agnostic interface to virtually all major LLM providers, typed tools, structured outputs, MCP support, as well as tight integration with the Pydantic Logfire observability platform. The framework is completely free and available under the MIT license; commercial offerings exist only for the optional Logfire observability product.
Historique du momentum
19.05.17.08.
Fonctionnalités
| Autonomy Level | Agent-based: agent loop with tool calls, multi-agent patterns (delegation, hand-offs, graph-based control flow), supports autonomous 'deep agents' and human-in-the-loop tools |
| Channels | Model-agnostic use with OpenAI, Anthropic, Gemini, DeepSeek, Grok, Cohere, Mistral, Perplexity, and more, plus cloud platforms like Azure AI Foundry, Amazon Bedrock, Google Cloud, Ollama, Groq, OpenRouter |
| License | MIT License (open source), Copyright Pydantic Services Inc. |
| Modalities | Multimodal input: text, image, audio, video, and documents (via ImageUrl, AudioUrl, VideoUrl, DocumentUrl, BinaryContent), depending on model support |
| Platform | Python library/framework (Python 3.9+ or 3.10+ depending on version), installed via pip; documentation at ai.pydantic.dev / pydantic.dev/docs/ai |
| Price | Free, open source (MIT license); no paid tiers for the framework itself. Optional observability (Pydantic Logfire) team plans start at $49/month, free personal tier available. |
| Release Date | First released on PyPI: May 20, 2024 |
| Tool/MCP Integrations | Full MCP client and server support (MCPToolset, MCP capability) via FastMCP Client; supports HTTP, SSE, stdio, in-process FastMCP servers, and multi-server configs |