Inspect a request

Inspect text before model delivery and prepare a protection policy.

Before you start

You need deployment-operator access to the security console and synthetic sample text. This inspector checks one text; connecting a production application to the gateway is a separate task.

Click path

  1. Open the console as the deployment operator.
  2. Open Инспектор текста (/playground).
  3. Click Чистый or enter a harmless question in Текст для инспекции.
  4. Select request in Точка инспекции. Use response for model output or tool_output for tool results.
  5. Optionally enter application context. Click Инспектировать or press ⌘Enter.
  6. Inspect classification, protection checks and the policy decision. Test synthetic sensitive examples separately.

Inspection scope. In the checked staging version this screen calls administrative batch inspection without a tenant parameter. The top tenant label and application-context field do not prove that a specific customer policy was applied. Check such a policy through /v1/inspect using that tenant’s inference key and compare decision.policy_id / rule_id.

Secure inspector: actual allow response for harmless text
Secure inspector: actual allow response for harmless text
2026-10-15 · Actual staging; normal operator sign-in, synthetic PUBLIC text, pipeline response and request ID returned by the gateway. This inspector does not prove tenant-specific policy scope.
  1. Text to inspect
  2. Inspection point
  3. Application context
  4. Inspect

Verify and recover

The result must correspond to the exact text and inspection point. A connection failure is an incomplete inspection, not a safe verdict. Restore sign-in for 401; check role and organization for 403.