Skip to content

Usage and requests

Every call made with a valid key is logged: client, key prefix, bookmark, status, problem code, duration and the inference it made. Features and outputs are never logged; they stay in the inference. The log is kept for 90 days by default.

The API overview: calls, error rate, p95 latency and this month's calls against the limit, a calls-over-time chart, busiest clients and recent failures.The API overview: calls, error rate, p95 latency and this month's calls against the limit, a calls-over-time chart, busiest clients and recent failures.

API shows the organization’s usage for the last 24 hours, 7 days or 30 days:

Tile Shows
Calls All logged calls in the period.
Error rate The share refused (4xx) or failed (5xx).
p95 latency Of the calls that succeeded.
This month Succeeded calls against the organization’s monthly limit, and when it resets.

Below them, Calls over time stacks succeeded, refused and failed calls per hour or day (UTC buckets labelled in your local time), Busiest clients lists the top five, and Recent failures the last five refused or failed calls with their status and code. Each client’s Overview tab shows the same panel for that client alone.

The Requests log with four calls: three succeeded with an Inference link each, one refused with 422 features_rejected.The Requests log with four calls: three succeeded with an Inference link each, one refused with 422 features_rejected.

API › Requests, or a client’s Requests tab, lists calls newest first. Filter Succeeded, Errors, Refused or Failed, and search by request id, client, bookmark or code. Each entry shows the client, the status and code, the time, the bookmark, the key prefix, the duration and the request id; a call that ran links to its Inference.

Calls without a valid key, and calls whose body is not JSON, are refused before they are logged.

Limit Default Counted
Calls per client 60 a minute Every logged call of the client, per clock minute.
Calls per organization 10,000 a month Succeeded calls of all its clients, per calendar month in UTC. A superadmin can set another limit for an organization.
Concurrent runs 2 on the instance API runs in progress at once, separate from public-page runs.

Past a limit, calls receive 429 or 503 with Retry-After; see the error codes. Operators change the defaults through configuration.