AffixIO authorization guard¶

AffixIO ADK adds a fail-closed authorization check around Python functions exposed as Google ADK tools. An explicit allow decision is required before the function runs. deny, review, malformed decisions, and authorization failures do not execute the tool.

Use cases¶

Tool authorization: Check whether an agent may call a sensitive function before execution.
Review boundaries: Stop review decisions until a separate approval process completes.
Decision evidence: Keep the returned decision and audit evidence with the host application’s run record.

Prerequisites¶

Python 3.10 or later
An AffixIO API key
An ADK agent project

Installation¶

pip install "affixio-adk[google]"

Use with an agent¶

import os
from google.adk.agents import LlmAgent
from affixio_adk import AffixIOAuthorizer

authorizer = AffixIOAuthorizer(
    os.environ["AFFIXIO_API_KEY"],
    agent_id="my-agent",
)

@authorizer.tool
def lookup_order(order_id: str) -> dict[str, str]:
    return {"order_id": order_id, "status": "ready"}

root_agent = LlmAgent(
    name="verified_agent",
    model="gemini-2.5-flash",
    instruction="Use lookup_order only for order questions.",
    tools=[lookup_order],
)

The wrapper is an ordinary Python callable, so it can be registered wherever ADK accepts a function tool. It does not send private prompts or operator identity to AffixIO. Applications should still apply their own input validation and secrets management.

Decision behaviour¶

Decision	Result
`allow`	The tool may execute.
`deny`	The tool is blocked.
`review`	The tool is blocked pending separate approval.
API timeout or error	The tool is blocked by default.