Agent trust layer
Four independent gates, and one place they arrive
Four independent checks sounds like bureaucracy until you notice they fail differently: one is about volume, one about the kind of change, one about how much this workspace has earned, and one about the action itself. What would be exhausting is four separate queues, and there is only one.
14-day evaluation · no card · start before you connect anything
Every request passes through, in order
- 01Agent identity
- 02Permissions and policy
- 03Risk and approvalsYou are here
- 04Action gateway
- 05Tenant isolation
- 06Prompt-injection defence
- 07Credential protection
- 08Audit and decision ledger
- 09Kill switch
Uncertainty at any step resolves to a refusal or to asking a person — never to proceeding.
How a request is decided
Where this sits in the loop
Approve is your step. You decide. It follows Plan and hands on to Execute.
- 01Each gate asks a different questionIs this send large enough to need a person? Is this kind of change permitted at all here? Has this workspace granted this level of autonomy? And is this specific action, with these specific facts, allowed right now?
- 02Uncertainty resolves to asking, never to proceedingWhere a check cannot be completed, the result is a refusal or a request for a person — never a quiet approval. The direction of failure is chosen deliberately: a delay costs you time, the other way costs you an action you did not want.
- 03One Inbox, with the consequence statedSends waiting for approval, changes agents proposed, plans to review and connections that need fixing all arrive in one list. Each says what it is, why it is there and what happens if you approve it — before you press anything.
- 04Turning something down teaches itWhen you reject a proposal you say why, and that reason is read by the agent before it proposes anything similar again. A rejection is information, not just a stop.
Four gates, one queue
Example
Send approval
Is this large enough to need a person?
Action policy
Is this kind of change permitted here at all?
Autonomy ladder
Has this workspace granted this much?
Trust layer
Is this action, with these facts, allowed now?
Anything waiting on a person arrives in the Inbox, with what happens if you approve it.
Four independent checks, because they fail differently. One queue, because four would be unusable.
What this guarantees
An empty queue is never dressed up as good news
Nothing waiting means nothing waiting. It is not presented as an achievement, because whether that is good depends entirely on why.
Autonomy is a ladder you climb deliberately
Observe, recommend, propose, act within limits. Moving up is your decision; moving down can happen automatically when predictions have been poor.
Only the right roles can decide
Approval is restricted to owners and admins, enforced on the server, and who approved what is recorded permanently.
What it cannot defend against
Said plainly. A control whose edges are hidden is worse than no control at all.
Approvals are not routed to individuals
The queue is shared by role. There is no assignment to a specific person and no multi-step approval chain.
No scheduled approval windows
Rules can restrict what may happen and when work may run, but there is no concept of an approval that automatically expires if nobody looks at it in time.
Every control, and what is not built
The security page lists each control as something you can check in the product, and lists separately what is not in place — including that Xyranik holds no certification and has had no external audit.