ActPass Documentation
ActPass sits between your AI agents and the tools they call. Every risky action — a refund, a deploy, an email — gets a deterministic allow / deny / needs-approval decision before it executes, and a tamper-evident audit record after.
How it works
Your agent (or our proxy in front of it) calls POST /api/v1/actions/preflightbefore executing a tool. ActPass verifies the agent's Action Passport, checks whether the tool's definition has drifted since it was approved, evaluates your policy, and returns one decision with a typed reason code. Allowed actions proceed; denied ones don't; ambiguous ones wait for a human. Everything is recorded in a hash-chained evidence ledger.
The decision is made by a deterministic rules engine — never by a model— and the system fails closed: if ActPass is unreachable or a signature doesn't verify, the action is blocked, not waved through.