One decorator to make any Python function auditable

I've been building AI agents for a while, and the one thing that always bugged me was this: how do you prove what your code actually did?

Logs are fine. But logs can be edited. They're not proof of anything.

So I built a decorator that adds cryptographic signatures to any Python function call. One line. No refactoring.

Before

Here's a normal function that summarizes a document using an LLM:

def summarize(doc: str) -> str:
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": f"Summarize: {doc}"}]
    )
    return response.choices[0].message.content

Works great. But there's no record that it ran, what it received, or what it returned. If a regulator asks "what did your AI agent do on March 15th?" - you've got nothing.

After

import asqav

asqav.init()  # uses ASQAV_API_KEY env var

@asqav.sign
def summarize(doc: str) -> str:
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": f"Summarize: {doc}"}]
    )
    return response.choices[0].message.content

That's it. One decorator.

Every time summarize() runs, asqav captures the function name, arguments, return value, and timestamp - then signs it cryptographically. The signature gets stored with a unique ID you can look up later.

Verifying a signature

Anyone with the signature ID can verify it. No API key needed:

result = asqav.verify_signature("sig_abc123")

if result.verified:
    print(f"Valid - signed by {result.agent_name}")

This is a public endpoint. Your auditor, your compliance team, or a regulator can verify signatures without access to your infrastructure.

Custom action types

By default, the decorator tags everything as function:call. You can be more specific:

@asqav.sign(action_type="llm:summarize")
def summarize(doc: str) -> str:
    ...

@asqav.sign(action_type="deploy:prod")
def deploy_model(model_id: str) -> None:
    ...

This makes it easier to filter and search your audit trail later.

Works with async too

If your function is a coroutine, the decorator handles it automatically:

@asqav.sign
async def summarize(doc: str) -> str:
    response = await client.chat.completions.create(...)
    return response.choices[0].message.content

No separate async_sign decorator. It just works.

Install

pip install asqav

Get an API key at asqav.com and set it:

export ASQAV_API_KEY="sk_..."

Then add @asqav.sign to whatever functions you want auditable. That's the whole setup.

Why bother?

The EU AI Act requires audit trails for high-risk AI systems. DORA wants operational resilience records for financial services. ISO 42001 needs evidence of AI governance controls.

You could build all of this yourself. Or you could add a decorator and move on with your day.

The SDK is open source: github.com/jagmarques/asqav-sdk