Glossary

Our words, and the market’s.

We use plain words on this site. Analysts, standards and security reviews use others. This page maps one to the other.

Terms

Nine words.

The first column is our word.

What it isThe market term
BadgeWhat an agent may do for this job, for whom, and until when. Issued from your directory.Agent identity with a scoped, time-limited credential
The doorThe check every request passes before it reaches an agent or a tool. Steps outside the badge are refused here.Policy enforcement point
NarrowingAt every handoff the badge gets smaller or stays the same. It never grows, including across organisations.Constrained delegation
The recordEvery decision, refusals included, written as it happens. Nobody can quietly edit it.Audit trail, multi-party endorsed
Sent to a humanA step over a limit waits for a person. The answer is written down like any other decision.Human-in-the-loop approval
RevokeEnd an agent’s badges. From then on it is refused at the door, and the revocation is on the record.Kill switch
SessionOne governed exchange between two parties, under one badge. Each handoff starts a new, narrower one.Delegation chain link; just-in-time access grant
Bring your own agentConnect an agent you already run, from any vendor, and govern it without changing its code.Third-party agent governance
Sovereign shapeAres in your building, disconnected if needed, on one small machine at 20–25 W.Sovereign, air-gapped deployment

Why two sets of words

Our words describe what you see happen: a badge is checked at a door and the result goes in the record. The market terms are how the same ideas appear in standards, analyst reports and security questionnaires. When a page here uses one, it pairs it with the other once.

See the permission model for how the badge is issued, narrowed, expired and revoked.

Next step

Watch the words happen.

Run a scenario in the sandbox and see each badge, door and record in one run.