Attest

Prove what your AI agent actually did.

Agents send email, update CRMs, create tickets, share documents, move money. Every framework makes it easy to give an agent tools. None can answer the three questions a company must answer before it lets agents act:

1. Should this action happen at all? — policy
2. Did a human approve the ones that matter? — gate
3. Did it actually happen, and can we prove it? — verification + evidence

Today the answer is "the API returned 200", which is not proof. Attest is the layer that turns it into proof.

Decide   → policy engine: act | ask | refuse, with a risk tier and reasons
Gate     → human confirmation where the team already works: Slack, web inbox, webhook
Verify   → read-back from the system of record: did it really happen?
Attest   → tamper-evident, hash-chained ledger; exportable for audit

One line, any framework, any app. Attest never executes the action — your tool does. That is why it is universal on day one. Verification depth is layered and honest: every ledger entry says exactly how much was checked.

import attest

@attest.action(system="gmail", verb="send", target="to")
def send_email(to, subject, body): ...

Quickstart → GitHub

What it is not

Not an agent framework. Not a connector platform. Not a guardrail / content filter. Not tracing. Those run the agent, execute the action, judge the text, or show what the model said. Attest proves what the world did.