Agent trust layer
One switch, and it fails in the safe direction
A stop button is only worth having if you know exactly what it stops and what happens when the system cannot tell whether it has been pressed. Both answers are here, and the second is the one most products get wrong.
14-day evaluation · no card · start before you connect anything
Every request passes through, in order
- 01Agent identity
- 02Permissions and policy
- 03Risk and approvals
- 04Action gateway
- 05Tenant isolation
- 06Prompt-injection defence
- 07Credential protection
- 08Audit and decision ledger
- 09Kill switchYou are here
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.
- 01It refuses from that momentNew runs do not start, changes waiting to be carried out are not, and any action an agent asks for is refused. A run already in progress stops at its next step rather than being torn out mid-operation.
- 02Unreadable means stoppedIf the switch itself cannot be read, agents are treated as stopped rather than permitted. That direction costs you a delay; the other costs you an action you did not want.
- 03Narrower switches for narrower problemsBeyond the single stop, individual capabilities can be paused — outbound messages, agent runs, or a specific supplier — for everyone or for one workspace, each with a reason that is recorded.
- 04Nothing is lost by stoppingWhat a switch holds back waits and continues when it is turned back on. Pausing is not the same as cancelling, and it does not discard queued work.
What one switch stops
Example
- New agent runsDo not start
- Changes waiting to be carried outNot carried out
- Anything an agent asks forRefused
- A run already in progressStops at its next step
- Your own peopleUnaffected — they work as normal
If the switch itself cannot be read, agents are treated as stopped. That direction costs you a delay; the other costs you an action you did not want.
What this guarantees
Every change carries a reason
A switch cannot be changed without one, and the reason is written into the record — and, where it affects one workspace, into that workspace’s own audit log too.
You can see what a switch is holding
The number of jobs currently waiting on a paused capability is shown, so the effect of a pause is visible rather than guessed at.
Customers are told when it affects them
Where a pause changes what somebody sees, the workspace is shown a plain explanation rather than left with something that silently does not work.
What it cannot defend against
Said plainly. A control whose edges are hidden is worse than no control at all.
A step in progress finishes
Stopping happens at the next step boundary. An operation already sent to a provider cannot be recalled, and nothing here pretends otherwise.
It takes a moment to reach every server
Switch states are held briefly for performance, so a change takes up to a few seconds to be seen everywhere. That is stated in the interface rather than hidden.
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.