Skip to main content

Check AI usage reporting

Documentation for Tutela v3.9.0.

Send one harmless AI request and check whether its usage reaches Tutela. Use this procedure after connecting a provider or when AI Spend appears empty despite recent activity.

Prerequisites​

  • A configured AI Chat provider or an API client that reports usage through Tutela.
  • Permission to use that connection and view AI Spend.
  • A non-sensitive test prompt. A policy simulation does not make a provider request or establish token usage.

Check one request​

  1. Note the current time and timezone. In your configured AI Chat or API client, send:

    Documentation usage check: reply with the word ready.
  2. Record whether the request completed, was blocked, or failed. Keep the request ID if available. A blocked request may not incur provider usage; use a permitted request when checking provider token reporting.

  3. Open AI Spend > Usage telemetry. Check the tracking state, latest event time, source and ingestion lag. Compare the event time with your test.

  4. Select a usage-reporting period that includes the request. Check the relevant provider and model breakdown. Operational telemetry covers the last 24 hours; it does not replace a missing selected-period request count.

  5. If the request produced usage without an estimated cost, inspect its pricing state. Confirm the model's price rather than assuming the request was free.

Interpret the state​

StateCheck next
Current usage evidenceCompare the source and event time with your request. Recent evidence from another source does not verify this connection.
No recent routed eventsConfirm that the client used the Tutela gateway and that the request reached the provider. Check Activity and the time filter.
Collection disabled or unconfiguredAsk the administrator to review this source's configuration before sending repeated tests.
Telemetry unavailableRetry the failed source and inspect Service health. Do not use an empty result as evidence of zero usage.
Usage present, price missingCheck the pricing entry for the reported provider and model.

If Activity contains the request but usage remains missing, record the request ID, source and times and follow AI usage and cost problems. Avoid repeating paid requests just to make a counter move.

Validate​

  • The test used the intended provider through Tutela.
  • The latest usage evidence and source were compared with the test.
  • Missing, delayed and unpriced usage were recorded separately.
  • The selected-period totals were checked without substituting the 24-hour operational count.

Next steps​