Skip to main content
Page stub. Full content lives in the README — Pydantic AI.

​
TL;DR

from hexgate.adapters.pydantic_ai import wrap_pydantic_agent
from hexgate.runtime import User

agent = wrap_pydantic_agent(agent=pydantic_agent, api_key="sk-...")
result = await agent.run(
    "What is the weather in Tokyo?",
    user=User(user_id="alice", role="member"),
)
Returns a HexgatePydanticAgent proxy backed by a clone whose tools are gated by a freshly built PolicyEnforcer. Non-allow decisions raise ModelRetry so pydantic-ai surfaces the policy outcome back to the model as a tool-result message.