Hey Product Hunt 👋 I built Charter after noticing something odd about how we demo AI agents. We show them booking flights and closing tickets. We never show what happens at 03:04 when one decides to move real money and there's no human in the loop — because there is no loop. Air Canada argued in court that its chatbot was "a separate legal entity responsible for its own actions." The tribunal disagreed and they paid. So that question is settled: it's your liability. What isn't settled is whether you can prove what your agent was allowed to do. Prompt guardrails are instructions to the thing you don't trust. Logs are written after the money has gone. Charter sits in between: the agent has to ask, and it's told "allowed" only after the ledger entry has committed — so it can't act unrecorded. The part I'm proudest of is the verifier. It re-implements the canonicalisation and hashing independently and shares zero code with the gate, so it can't inherit a bug from the thing it's auditing. Change one byte in an old entry and it names the exact sequence number. One honest gap: an agent that never calls Charter isn't governed by it. Fixing that means Charter holding the payment credential itself and refusing to sign. That's next, and it isn't built. Would genuinely value pushback — especially from anyone who's shipped agents with write access to production.

Charter

A policy gate your AI agent must ask before it spends

A policy gate your AI agent must ask before it spends