Skip to main content
Usage shows aggregate model traffic. Use it to understand which models are being used, compare latency and failures, and check token and cost coverage before investigating individual requests.

​
Choose the reporting scope

  1. Select a project in the sidebar, or All projects for an organization overview.
  2. Open Usage and choose the reporting period.
  3. Choose personal or project usage where your account permits both views.
  4. Review volume, token totals, model/provider breakdowns, latency, errors, and attributed spend.
Personal usage follows the user recorded on each project API key. It cannot distinguish people sharing a key. An administrator can limit a member to personal analytics without changing their operational role; see Organization and access.

​
Read the evidence

Usage metadata arrives asynchronously after requests complete. A recent request may appear after a refresh. The reporting history you can query follows your subscription’s analytics retention; changing the selected period does not restore expired history. Token counts and attributed costs depend on the evidence returned by providers. An unknown amount is not free inference. Use Pricing and cost attribution to interpret coverage, provider-reported costs, estimates, and partial accounting. Applicable personal budget cards show the allowance, attributed usage, remaining amount, and reset information. Administrators create and manage those policies in Cost control.

​
External usage monitors

A trusted desktop monitor or server-side client can read a project API key’s retained seven-day request, token, and attributed-spend totals, together with the budgets that apply to that key. Use Get gateway key usage in the API reference with the same Aixy key used by the workload. The reporting request does not run inference, reserve budget, or update the key’s inference activity timestamp. An exhausted hard budget does not prevent reading its balance. The key grants access to its own totals and applicable budget aggregates, including shared organization, project, or team balances. It does not grant access to another key’s usage, other users’ identities, provider credentials, prompts, or responses. This remains an inference-capable secret: only configure it in applications you trust. Disabling, revoking, or expiring the key also removes reporting access; valid rotation overlap works until its expiry. Budget cycles are independent of the seven-day reporting window. Several overlapping limits can apply, so keep their amounts separate. Hard-budget availability includes outstanding reservations; these are not confirmed provider charges. Recorded spend can lag enforcement or cover only retained history. Null usage or balances mean unavailable, not zero, and missing budgets mean no configured limits were returned. Refresh at most once a minute and do not cache responses in shared storage.

​
Move from a trend to an action

  • Open Activity to investigate individual requests and routing attempts.
  • Configure Monitors for sustained errors, latency, or budget utilization.
  • Use Telemetry exports to send request metadata and metrics to your own observability system.