Usage summary
Adds up your team’s API usage over a time window: requests, credits,
errors and results. Pass group_by to split the totals by day,
product, endpoint, charge component, API key, client surface, client
platform or status class. Leave it out to get one total row.
Every row is scoped to your team. No parameter reads another team’s usage.
With the default bucket=1d and a window made of whole UTC days, the
summary is read from daily totals. Those know the day, product,
endpoint, API key, client surface and status class, so some shapes
are rejected with a 400 that tells you to retry with bucket=1h:
group_by=componentcombines only withdayandproduct, and filters only byproducts.- The
statusanderror_typefilters needbucket=1h.
A window that starts or ends mid-day, bucket=1h, or any grouping that
includes client_platform accepts every combination.
This endpoint is free: the response carries no X-Credits-Used
header. It shares a limit of 60 requests per minute with the other
/account/usage/* endpoints.
Authorizations
API key passed as a Bearer token in the Authorization header.
Headers
API version to use. This endpoint currently requires 2025-11-01. Requests without the header, or with any other value, return 400.
2025-11-01 "2025-11-01"
Query Parameters
Dimensions to split the totals by. Repeat the parameter or send a
comma-separated list, for example group_by=day,component. A
dimension can appear once. Omit it for a single total row.
Grouping by day returns the time column as bucket_start.
Grouping by component returns component and label, and
swaps errors and results for quantity.
client_platform is the tool or runtime that made the call, such
as claude-code, python or curl. New values can appear at any
time, so do not hard-code the list. It works with any bucket.
day, product, endpoint, component, api_key_id, client_surface, client_platform, status_class Size of each time bucket when you group by day. 1d returns a date in bucket_start, 1h an ISO 8601 timestamp.
1d, 1h Start of the window, inclusive. An ISO 8601 timestamp such as
2026-09-23T14:00:00Z, or a date such as 2026-09-21, which reads as
midnight UTC. Defaults to 7 days before end. Cannot be more than 12
months ago.
End of the window, exclusive. Same formats as start. Defaults to the coming midnight UTC, so the default window is the last 7 whole days including today.
Keep only requests made with these API key ids (api_key_id on an event). Repeat the parameter or send a comma-separated list, up to 50.
50x >= 1Keep only these endpoints, as they appear in endpoint, for example /person/search. Repeat the parameter or send a comma-separated list, up to 50.
50Keep only these products, as they appear in product, for example person_search. Repeat the parameter or send a comma-separated list, up to 50.
50Keep only these HTTP status codes, for example status=400,404. Up to 20. On GET /account/usage/summary it needs bucket=1h, or a window that is not whole days.
20100 <= x <= 599Keep only one status class.
2xx, 3xx, 4xx, 5xx Keep only requests from these surfaces, as they appear in client_surface, for example api or mcp. Repeat the parameter or send a comma-separated list, up to 10.
10Keep only requests that failed with this error_type, for example invalid_request. On GET /account/usage/summary it needs bucket=1h, or a window that is not whole days.
Response
One row per combination of the grouped dimensions. Rows are sorted by bucket_start when you group by day, then by credits, highest first.
Usage totals, one row per combination of the grouped dimensions.
The rows. A single row when group_by is omitted.

