Authorization for AI agents is the problem of granting an autonomous program exactly the authority it needs — and being able to prove, limit, and revoke it.
Human authorization assumes a person behind each session, sessions that are short, and actions a human initiates. Agents violate all three: they run unattended, chain thousands of actions, delegate to other agents, and decide what to do from model output that can be manipulated. Coarse credentials (API keys) and role checks (RBAC) can't express "this task may spend $20 reading these repos for the next hour."
The fix is to authorize the action, not just the caller. A capability token — Liceat calls it a warrant — encodes precisely what may be done. It's scoped, spend-capped, rate-limited, time-boxed, holder-bound, revocable, and verifiable offline. Delegation only narrows it. Every use is recorded in a tamper-evident log.
Liceat is permission infrastructure for AI agents. You issue a warrant — a scoped, spend-capped, revocable token — that grants an agent narrow authority; the agent presents it at the point of action; Liceat enforces it (scope, spend, rate, holder-binding, revocation) and records every decision in a tamper-evident log. It is open source with a zero-dependency SDK for TypeScript and Python.
No. OAuth authorizes a client to act on a user's behalf but issues broad, stateful access tokens with no built-in spend caps, offline attenuation, or holder-bound sub-delegation. See OAuth for AI agents.
No. A warrant is a signed, self-describing token; the SDK verifies it fully offline against the issuer's public key. Use the gateway when you also want centralized audit, stateful spend/rate limits, and instant revocation.
Free to self-host. 10,000 free decisions on the hosted gateway — no card.
Get an API key Read the docs