Build type-safe AI agents with validated outputs in Python

PydanticAI is a Python agent framework from the Pydantic team that brings type-safety and validation guarantees to LLM-based applications. It enables developers to build AI agents that return validated, structured outputs using Pydantic models instead of raw strings, with support for tool use, dependency injection, streaming responses, and multi-turn conversations.

Use PydanticAI when building production AI agents that call tools and return structured data, when you need validated and typed LLM outputs, when you want to write unit tests for agent logic without hitting a real LLM, or when switching between LLM providers without rewriting agent code. Use it when working with concepts like Agent, @agent.tool, RunContext, ModelRetry, or result_type.

Do NOT use this skill when you need simple, unstructured text generation without validation requirements, or when type-safety and testability are not priorities for your LLM application.

You provide a system prompt, an LLM model identifier (e.g., 'anthropic:claude-sonnet-4-6'), optional Pydantic models for structured outputs, tool functions decorated with @agent.tool, and dependency objects for injection. The framework returns fully typed and validated results as Pydantic model instances, usage statistics (tokens, requests), and streaming text chunks for real-time responses.

PydanticAI supports OpenAI and Azure OpenAI, Anthropic Claude, Google Gemini, Groq, Mistral, Google Vertex AI, and Ollama. Install provider-specific extras:

pip install pydantic-ai

# Install extras for specific providers
pip install 'pydantic-ai[openai]'       # OpenAI / Azure OpenAI
pip install 'pydantic-ai[anthropic]'    # Anthropic Claude
pip install 'pydantic-ai[gemini]'       # Google Gemini
pip install 'pydantic-ai[groq]'         # Groq
pip install 'pydantic-ai[vertexai]'     # Google Vertex AI

The framework includes TestModel and FunctionModel for unit testing without real LLM calls, and supports httpx for HTTP client integration in tools.

PydanticAI is designed for Python developers building production AI agents, chatbots, or LLM pipelines where correctness and testability matter. It serves backend engineers who need to inject services like databases and HTTP clients into agent logic, QA engineers writing unit tests for LLM-powered features, and teams that require provider flexibility without code rewrites. The framework is ideal for developers already familiar with Pydantic's validation patterns who want the same guarantees in their AI applications.

Minimal agent example:

from pydantic_ai import Agent

# Simple agent - returns a plain string
agent = Agent(
    'anthropic:claude-sonnet-4-6',
    system_prompt='You are a helpful assistant. Be concise.',
)

result = agent.run_sync('What is the capital of Japan?')
print(result.data)  # "Tokyo"
print(result.usage())  # Usage(requests=1, request_tokens=..., response_tokens=...)