As AI agents start to act (not just generate text), we need a consistent way to record what they intended, what policy governed them, and what evidence shows what actually happened.

*APAAI* defines a minimal HTTP/JSON spec for this loop: *Action → Policy → Evidence*

It’s model-agnostic, open-source (Apache-2.0), and already shipping SDKs in *TypeScript* and *Python*.

Example use case: An agent proposes to send an email → requires approval → submits evidence of delivery — all recorded in a standardized, auditable format.

Docs & spec: https://apaaiprotocol.org TypeScript SDK: https://www.npmjs.com/package/apaai-ts-sdk Python SDK: https://pypi.org/project/apaai GitHub: https://github.com/apaAI-labs

We’re opening the *RFC process* and would love feedback from the HN community — especially anyone building *agents*, *AI infra*, or working on *governance and observability* for autonomous systems.

Thoughts on the data model? Should “accountability as code” become a first-class design pattern for agentic AI?