PydanticAI + ODEI: Type-Safe AI Agents with Constitutional Validation

#python #ai #agents

PydanticAI + ODEI: Type-Safe Agent Actions with Constitutional Validation

PydanticAI brings type safety to AI agents. ODEI brings constitutional governance. Together, they make the most reliable agent stack in 2026.

Why Type Safety + Constitutional Validation

PydanticAI validates that agent outputs match your schema. ODEI validates that agent actions are constitutionally safe. Different layers, both essential.

Layer	What It Prevents
Pydantic schemas	Wrong data types, missing fields
ODEI guardrails	Unauthorized actions, duplicates, hallucinated refs

Integration

from pydantic import BaseModel
from pydantic_ai import Agent
import requests

class TransferResult(BaseModel):
    success: bool
    tx_hash: str
    amount: float

ODEI_TOKEN = "your-token"

def constitutional_check(action: str, severity: str = "medium") -> bool:
    r = requests.post(
        "https://api.odei.ai/api/v2/guardrail/check",
        headers={"Authorization": f"Bearer {ODEI_TOKEN}"},
        json={"action": action, "severity": severity}
    ).json()
    return r["verdict"] == "APPROVED"

agent = Agent(
    "openai:gpt-4",
    result_type=TransferResult,  # Pydantic type validation
)

@agent.tool_plain
def execute_transfer(amount: float, wallet: str) -> TransferResult:
    # Constitutional check BEFORE execution
    if not constitutional_check(f"transfer {amount} USDC to {wallet}", "high"):
        return TransferResult(success=False, tx_hash="", amount=0)

    # Pydantic validated execution
    return do_transfer(amount, wallet)