API reference
Rate limits & retries
Streaming, throughput limits, quota, and automatic failover.
Streaming
Streaming is supported: set "stream": true on /v1/chat/completions, /v1/messages and /v1/responses to receive Server-Sent Events. Non-streaming requests return one complete JSON body.
Throughput & quota
Throughput is bounded by your prepaid balance and per-key quota, plus an optional per-key RPM limit; exceeding the RPM limit returns 429 with a Retry-After header, while running out of balance or quota returns 402.
Failover
When a supplier upstream fails, Kumo automatically fails over to the next available channel for the same model; if every channel is exhausted you get a contract-shaped 502 (or 503 no_available_channel when all channels are disabled).