Usage, limits and the log
Who pays, usage as records, limits per business, key and customer, and the log of every call.
Who pays
A business you brought (made for your connection, or a test business) counts on your plan; a project its owner already had keeps its own. Your calls inside a business are always counted on your project, by connection and key.
Usage as records
GET /usage/records answers one meter between two days, cut by day, connection, key or customer. api_calls is the calls made with your keys, those inside your businesses included.
curl -sS "https://api.rivier.ai/usage/records?meter=api_calls&from=2026-12-01&to=2026-12-07&by=connection" \
-H "Authorization: Bearer $RIVIER_API_KEY"Your own limits
PUT /usage/allocations bounds a meter for one connected business, one key or one customer. You hear usage.threshold when one nears its end and usage.limit_reached when it is full, each naming the scope.
curl -sS -X PUT https://api.rivier.ai/usage/allocations \
-H "Authorization: Bearer $RIVIER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"meter": "bookings", "scopeKind": "connection", "scopeId": "$A", "limit": 500}'The log of calls
Every call made with a key is a row in GET /usage/requests for 30 days: the operation, status, time taken, key, connection, idempotency key and error code, and no bodies. Every answer carries x-rivier-request-id; quote it when you ask about a call.
curl -sS "https://api.rivier.ai/usage/requests?status=4xx&connection=$A" \
-H "Authorization: Bearer $RIVIER_API_KEY"Limits and refusals
Too many calls answer 429 with retry-after in seconds and the class that was full. An allowance that is used up answers quota_exceeded with a sentence you can show; it is not worth retrying before the period resets.