Alpha Program Open

Behavioural Authorizations
for Autonomous Software

The runtime layer that evaluates what your agents actually do, before they do it.

A new security stack

Isolation bounds where an agent runs. Identity bounds what it can reach. Neither reads the action itself, the content, the recipient, the timing. Behavioural authorization does, at runtime, before the tool call leaves the host.

01Isolationwhere the agent runs
02Identitywhat it can reach
03Behaviourwhat it actually does
AI Agent
non-deterministic by design
01
Isolation
where the agent runs
02
Identity
what it can reach
03
Behaviour
what it actually does
The scenario

A support agent is about to send a customer their weekly status. Isolation passes. Identity passes. And the body quotes a project belonging to another client. Only behaviour reads the action itself, before it leaves the host.

Deploy with any
agent harness

Native SDKs, one-line drops into the harnesses your teams already use, and a global API reachable from anywhere. Pick the surface that fits your stack and keep shipping.

SDK

Python & TypeScript SDKs

Wrap your agent runtime in a few lines. Evaluate every tool call against your policies before it leaves the host, no architecture changes required.

Harness

Claude Code & OpenClaw

One-line integration into the harnesses your teams already run. Policy enforcement switches on the first session, no migration required.

API

Global /check API

Reachable from any runtime, anywhere. Send a tool call, get an allow or deny with obligations attached. No sidecar, no daemon.

We're onboarding
the first teams

Limited spots. Every application is reviewed by our team.

Request early access

No commitment required. We'll reach out directly.