AWARE documentation
The security control plane for autonomous AI agents.
AWARE gives every agent a non-human identity, evaluates every tool call against policy before it runs, revokes credentials when authority must end, and records each decision in a hash chain that a third party can independently recompute.
Version
These docs describe
@goodciso/aware 2.12.3, Apache-2.0. The install command below installs that version.
Where to start
- What is AWARE? — the model, in plain terms.
- Quick start — a running instance and a first evaluated tool call.
- Architecture — the components and the request path.
- Execution assurance — what AWARE guarantees, and what it does not.
Two things worth knowing before you start
1. AWARE authorises; it does not verify execution.
The policy engine decides whether an action is permitted. Nothing in the current design observes the target afterwards to confirm the action happened as authorised. See Execution assurance.
2. Revocation is per-agent.
Revoking an agent invalidates that agent’s credential. Cascade to derived agents is not implemented. See Revocation.