Usage and credits
Read consumption per user, per thread, and per organization, and see how much runway is left.
Usage is metered in credits, with a flat cost per action rather than raw token billing.
Endpoints
GET /api/usage/me
GET /api/usage/threads
GET /api/usage/thread/{id}
GET /api/usage/org
GET /api/usage/runwayrunway is the useful one for dashboards: it projects how long the current
balance lasts at the recent burn rate.
The credits header
Metered responses carry your remaining balance:
X-Credits-Remaining: 412.75Read it rather than polling /api/usage/me: it is on the response you already
made.
Running out
A request that would overdraw returns 402 with a payload describing the
shortfall, rather than failing with a generic error. Handle it as a prompt to
top up, not as an outage.
Costs differ by action
A chat turn, a config generation, and a web search are not priced the same. If you are budgeting, meter the action, not the request count.