Skip to content
Relaykit
Menu

Status

Is the gateway answering?

This page asks the gateway directly when you open it and shows the answer your browser just got. Further down: each model's uptime over the last 24 hours, counted from real requests.

Gateway backend

Checking

Sending one request to the gateway.

Response time
—
Last checked
—
Endpoint
GET /api/status

The check re-runs every minute while this tab is open. It confirms the gateway is up and reachable from your network. Each model also depends on its upstream provider, so one model can be slow or failing while the gateway itself is fine.

Uptime by model

Last 24 hours, hour by hour

Each bar is one hour: the share of requests to that model that got an answer. It comes from the gateway's own request log, recounted every 5 minutes. A request that failed and then succeeded on a retry counts as answered. Client errors (bad key, no balance, rate limits, bad requests) are not counted.

  • 95% or more answered
  • 75% to 95%
  • Under 75%
  • Fewer than 3 requests
  • No requests

The 24-hour figure appears once a model has 20 or more requests in the window. Loading

  • Claude Opus 5.5
    Loading
  • Claude Sonnet 5
    Loading
  • Claude Haiku 4.5
    Loading
  • GPT-6.1 Sol
    Loading
  • GPT-5.6 Sol
    Loading
  • GPT-5.6 Terra
    Loading
  • Grok 4.6
    Loading
  • Claude Sonnet 5.5
    Loading
  • Grok 4.7
    Loading
  • DeepSeek V4 Flash
    Loading
  • DeepSeek V4 Pro
    Loading

Troubleshooting

Gateway up, requests still failing?

Most failures come from the key, the balance or one upstream model. The status code tells you which. Requests to the API go to api.oneaiapikey.com.

  1. 401

    Key rejected

    The key is missing, mistyped or revoked. Check the Authorization header, or create a new key.

    Open API keys
  2. 403

    Balance, cap or model

    The balance or the key spending cap is used up, or the key is not allowed to use this model. Top up, raise the cap or edit the key.

    Open billing
  3. 429

    Rate limited

    Back off and retry after the number of seconds in the Retry-After header. Per-key limits are set in the console.

    Error reference
  4. 5xx

    One model failing

    If only one model errors, its upstream provider is usually the cause. Try a different model and tell us the time and model ID.

    All models
  5. cut off

    Long answer stopped mid-way

    Proxies can close a non-streaming request that stays silent for about 100 seconds. Send "stream": true for long outputs.

    Streaming setup

Seeing a failure this page does not show? Send the time (UTC), the model ID, the status code and the X-Request-Id response header.

Report a problem