pydantic-ai

What it is

Pydantic AI is an AI Agent Framework implemented in Python, positioned as a GenAI agent framework aligned with Pydantic tooling. It supports multiple providers and capabilities integration, including tools for thinking and web search, and features for dependency injection and structured outputs.

How it works

The repository showcases an agent model that can be instantiated with a model identifier, optional instructions, and capabilities. It supports dynamic instruction handling, tool registration, and dependency injection via RunContext. Examples demonstrate using capabilities such as Thinking and WebSearch, and registering tools with context-aware dependencies.

Getting started

• The README shows how to instantiate an Agent with model and instructions, then run synchronously:

from pydantic_ai import Agent

agent = Agent(
    'anthropic:claude-sonnet-4-6',
    instructions='Be concise, reply with one sentence.',
)

result = agent.run_sync('Where does "hello world" come from?')
print(result.output)

• A more advanced example demonstrates dependency injection and tools:

from pydantic import BaseModel, Field
from pydantic_ai import Agent, RunContext

# ... define deps and output types ...

support_agent = Agent(
    'openai:gpt-5.2',
    deps_type=SupportDependencies,
    output_type=SupportOutput,
    instructions=(
        'You are a support agent in our bank...'
    ),
)

@support_agent.instructions
async def add_customer_name(ctx: RunContext[SupportDependencies]) -> str:
    ...

@support_agent.tool
async def customer_balance(ctx: RunContext[SupportDependencies], include_pending: bool) -> float:
    ...

• Getting started includes a link to install instructions in the docs:

To try Pydantic AI for yourself, install it and follow the instructions in the examples.

Recent releases

• v2.24.0 (2026-08-04) with bug fix: Fix Google provider dropping an explicit timeout=0 by @pydanty[bot] in https
• v2.23.0 (2026-08-03) with feature: Add Bedrock support for ModelSettings.extra_headers by @kkauy in https://gith
• v2.22.0 (2026-07-31) with feature: Allow MCPToolset clients to skip optional MCP tasks via prefer_tasks by @ad
• v2.21.0 (2026-07-29) with feature: feat(usage): add per_request_input_tokens_limit to UsageLimits by @Oxygen56
• v2.20.0 (2026-07-28) with feature: Add Claude Opus 5 (claude-opus-5) support by @dsfaccini in https://github.com

Traction

• Stars: 19055

Behind the repo

• Repository owner: pydantic/pydantic-ai (organization: pydantic)
• License: MIT

Caveats

• Created: 2024-06-21
• Last push: 2026-08-04
• Open issues: 617
• Forks: 2480
• Language: Python
• Topics: agent-framework, genai, llm, pydantic, python
• License: MIT