Skip to main content
GET
Returns counts and rates for your organization only. The response never contains question or answer text. See Analytics for how to read the numbers. Choose the window in one of two ways: days for a trailing window ending now, or an explicit start and end pair. Sending both forms returns 400. With neither, you get the last 30 days. The window is half-open, [start, end), and may span at most 366 days. Filtering by assistant_id narrows every count to that assistant. An ID that belongs to another organization, or to no assistant, is not an error: it matches nothing and every count is 0. Requires the owner, admin or editor role. Members and dataset operators receive 403.
integer
default:"30"
Length of a trailing window ending now, from 1 to 366. ASCII digits only. Cannot be combined with start or end.
string
Start of an explicit window, as an ISO 8601 timestamp with a UTC offset (for example 2026-09-01T00:00:00Z). A timestamp without an offset is refused. Requires end.
string
End of an explicit window, exclusive, as an ISO 8601 timestamp with a UTC offset. Must be after start, and no more than 366 days after it. Requires start.
string
Limit the counts to one assistant (UUID). A malformed ID returns 400.

Response

object
required
The window the counts cover.
integer
required
Conversations started in the window.
integer
required
Questions asked in the window.
integer
required
Answers started in the window that have finished, whether complete, interrupted or failed. Answers still in progress are not counted.
integer
required
Finished answers that carried at least one citation.
integer
required
Finished answers where the assistant declined to answer because its sources did not cover the question. See How answers work.
integer
required
Answers that failed before producing a usable result.
number | null
required
grounded_answers / answers, from 0 to 1. null when answers is 0, meaning not applicable rather than zero.
number | null
required
abstained_answers / answers, or null when answers is 0.
number | null
required
failed_answers / answers, or null when answers is 0.
object[]
required
Answers per model, highest count first, at most 50 entries.
object[]
required
Questions by the kind of caller that asked, highest count first.
Last modified on September 26, 2026