Installation

pip install veto pydantic-ai

Quick start

from veto import Veto
from pydantic_ai import Agent
from veto.integrations.pydanticai import wrap_pydanticai_tool

veto = await Veto.init()
agent = Agent("openai:gpt-4o")

async def transfer_funds_handler(amount: float, to_account: str) -> str:
    return f"Transferred ${amount} to {to_account}"

guarded_transfer_funds = wrap_pydanticai_tool(
    veto,
    "transfer_funds",
    transfer_funds_handler,
)

@agent.tool_plain
async def transfer_funds(amount: float, to_account: str) -> str:
    """Transfer money between accounts."""
    return await guarded_transfer_funds(amount=amount, to_account=to_account)

Decorator API

from veto import Veto
from pydantic_ai import Agent
from veto.integrations.pydanticai import create_veto_tool_decorator

veto = await Veto.init()
agent = Agent("openai:gpt-4o")

@agent.tool_plain
@create_veto_tool_decorator(veto, "get_balance")
async def get_balance(account_id: str) -> str:
    return f"Balance for {account_id}: $5,000"

The decorator inspects the function signature, builds an args dict from call kwargs, runs veto.guard(...), and then executes the original function only when allowed.

How it works

For wrapped handlers/decorators, Veto runs:

Build args dict from the function signature + call arguments
await veto.guard(tool_name, args_dict)
If denied, raise ToolCallDeniedError
If allowed, execute the original async handler

Example output

transfer_funds(200) -> Transferred $200 to ACC-001          # allowed
transfer_funds(5000) -> DENIED: Tool call denied:           # blocked by policy
  transfer_funds - Amount 5000 exceeds limit of 1000
get_balance('ACC-001') -> Balance for ACC-001: $5,000       # allowed

total_calls:    3
allowed_calls:  2
denied_calls:   1

Notes

pydantic-ai remains an optional dependency in the SDK.
These helpers wrap plain async Python functions, so they can be used in or out of PydanticAI as long as the function call contract is the same.

PydanticAI Integration