Cloudflare Agent Memory enters public beta today. It solves a real problem: agents that die between sessions lose context, state, and continuity. Durable Objects backed by KV gives you persistent agent state that survives container boundaries.

But persistence without trust is a different problem.

An agent with write access to Cloudflare Agent Memory can store anything — credentials, session tokens, scraped data, intermediate reasoning chains. If you're accepting memory writes from an external agent, or letting an external agent read from your memory store, you're making a trust decision. The question is whether you're making it deliberately or by default.

This post shows how to use AgentLair's memory-trust endpoint to make that decision deliberately.

What Memory-Scoped AATs Are

AgentLair issues Agent Authentication Tokens (AATs): short-lived JWTs that cryptographically attest what an agent is permitted to do.…