Limits and fair use

llmfaucet is a shared gateway. Limits protect the public service and make capacity available to more developers.

Plan limits

Daily limits reset at midnight UTC. A personal key changes the access tier; it does not create a dedicated provider or reserve tokens.

Request ceilings

max_tokens is plan-specific. Anonymous requests are capped at 1,024 tokens; registered requests at 2,048; Supporter and Pro requests at 4,096. Request bodies are bounded before provider routing.

Rate-limit headers

Successful API responses include:

X-RateLimit-Limit: 20
X-RateLimit-Remaining: 19
X-RateLimit-Reset: 2026-08-25T00:00:00.000Z
X-LLMFaucet-Plan: anonymous

When a limit is reached, the gateway returns an OpenAI-shaped error:

{
  "error": {
    "message": "Daily request limit reached.",
    "type": "rate_limit_error",
    "param": null,
    "code": "daily_limit_exceeded"
  }
}

Honor Retry-After. Do not retry immediately or create parallel streams to work around a limit.

Routing and failover

The router reads a precomputed catalog, filters by capability and health, then uses a bounded fallback chain. A failed upstream can be cooled down for a short period. Failover does not cross vision, tools, JSON, embeddings-family, or other incompatible capability boundaries.

Fair-use rules

Do not automate account creation, rotate identities, bypass provider limits, scrape protected services, resell access, or use the gateway for production- critical workloads. We may rate-limit, suspend, or revoke access when traffic threatens shared capacity or violates provider terms.

Free-provider availability changes without notice. No plan guarantees a model, token quantity, latency, uptime, output quality, or production SLA.