FAQ

[← Back to Docs](/docs)

No. n8n is optional.

The HTTP API is the default integration.

AI Safety Gate can be used from any app or service that can call HTTP endpoints.

Yes.

Call AI Safety Gate right before irreversible actions.

The supported model is:

• call POST https://aisafegate.com/api/validate in your server-side app or service before executing the action
• enforce the returned decision (PASS/WARN/BLOCK)
• for WARN, poll GET https://aisafegate.com/api/decisions/:id/approval?approval_token=... and only proceed on { "approved": true }

Yes.

Integrate AI Safety Gate in your server-side app or service (or serverless functions) so that:

• AI output is validated before it is published or used to trigger side-effects.

Do not put API keys in client-side code.

Any language/runtime that can make HTTP requests.

Yes. You can validate output from any model.

No. Use AI Safety Gate from server-side code only.

Yes.

Fail closed.

• If POST https://aisafegate.com/api/validate cannot be completed safely, do not execute the downstream action.
• For WARN decisions, approval polling returns { "approved": false } on errors; clients must not treat errors as approval.

No, not if you follow the documented integration contract.

• WARN decisions must not execute until approval is observed.
• Approval polling returns { "approved": false } when approval cannot be confirmed.
• Only { "approved": true } allows execution.

Execution stops.

Your app must treat timeouts as “not approved” and must not execute.

Yes, but only within documented limits.

Safe customizations include:

• changing the upstream logic and prompts
• adjusting poll delay and maximum wait time
• changing what happens after PASS

You must not change:

• decision_id propagation
• approval_token propagation
• approval polling URL/query parameter format
• fail-closed STOP nodes