The runtime that decides what your agents can do.

Every action an autonomous agent attempts passes through one surface. Intent is compiled. Conformance is scored. Authority is decided. A cryptographic token is issued — or refused. Every step is preserved as evidence.

Authority is decided once, then carried.

A single authority decision flows through every downstream connector. The connector doesn’t re-decide — it verifies the token presented to it. No tokens, no execution. Decisions are made once; verification happens at the gate.

Four named systems, one operator surface.

Each component is independent, documented, and observable. You can adopt them separately — Intent Layer alone for classification, Authority Engine alone for decision-making — or operate the full stack as one runtime.

Failure modes are designed in, not bolted on.

When the policy is unclear, the runtime fails closed. When LIM confidence drops below the operator-set threshold, the decision escalates. When a token is presented past its TTL, the connector refuses it. When the audit chain head is unreachable, no new decisions are issued. Each failure is observable and named.

No silent allow paths. No “default open” fallbacks. No undocumented bypass. The contract is: no token, no action.

Where the runtime gates today.

Every agent action, reviewed before it runs.

The operator console is where intents land, get scored against policy, and are approved, escalated, or denied — every decision written to an immutable audit chain you can verify.

See it run in slow time.

The how-it-works page is a long-scroll ceremony in which one Authority Token is constructed in front of you, signed in your browser, and handed to you to keep. ~90 seconds of slow time to witness what the engine is designed to do in the blink of an eye.