Limits and quotas

routeur.ai does not apply a fixed per-key requests-per-minute or tokens-per-minute cap. Throughput is governed by your plan allowance at the organisation level, plus gateway backpressure when the service is saturated.

Plan allowance

Each plan includes a pooled monthly routed-request allowance, scaled by seats. Usage beyond the allowance is billed as overage rather than refused — you are not cut off at the boundary.

Traffic is only refused once an organisation passes the blocking guardrail well above its allowance:

429allowance_exceeded

The monthly routed-request allowance and its overage guardrail have both been reached. Add seats, upgrade the plan, or contact support to raise the cap.

402billing_inactive

The organisation's subscription is cancelled. A past_due subscription keeps serving traffic through the dunning window.

Both gates are deliberately eventually consistent — usage is rolled up on a schedule, projected to the gateway, and cached — so they take effect within minutes, not instantly.

Read-only metadata endpoints (GET /v1/models) deliberately keep working when an organisation is capped, so a client can still connect and discover why it is being refused.

Budgets

Budgets are separate from plan allowances: you set them per workspace or project in the dashboard to alert on, or stop, spend against your own thresholds. See the dashboard's Cost section.

Gateway backpressure

Independently of billing, the gateway sheds load when its in-flight request cap is reached:

503application/json

caller response

{
  "error": {
    "code":    "overloaded",
    "message": "server is at capacity",
    "type":    "routeur_error"
  }
}

A 503 overloaded carries a Retry-After header. Retry with jittered backoff.

Upstream rate limits

If the upstream provider rate-limits us for your call, that surfaces as 429 upstream_rate_limited. This reflects the provider's own limits on the credential in use, not a routeur.ai cap.

No rate-limit headers. The gateway does not emit X-Routeur-RateLimit-* headers. Track consumption from the usage dashboard or your traces rather than from response headers.

← Errors

Request lifecycle →