AI gateway health-check guide

A small-spend self-test checklist to decide whether an AI API gateway fits your project. It replaces no guarantee — it just makes the risk visible.

View the status pageCompare candidates
Small amounts
Trial first

Checklist

Six checks before you launch

Run the same script and the same set of models against every candidate so stability and real cost stay comparable.

  1. 1

    Check the site and docs

    Confirm the website, console, docs, and pricing page all load, and read the terms of service, refund rules, model list, and recent announcements.

  2. 2

    Check /v1/models or the model catalog

    If the platform is OpenAI-API compatible, request /v1/models first; for native APIs, go by the platform's model catalog and the models selectable in the console.

  3. 3

    Run a minimal text request

    Send one non-streaming request with a low-cost model and record the HTTP status code, time to first token, total latency, error message, and actual charge.

  4. 4

    Test streaming output

    If your product depends on SSE streaming, test stream=true separately — including mid-stream disconnects, timeouts, and proxy-layer buffering.

  5. 5

    Verify balance and logs

    After the request, check the balance deduction, call logs, token counts, and whether failed requests are billed. Don't ship on the strength of one successful response.

  6. 6

    Set up production safeguards

    Before launch, configure budgets, rate limits, timeouts, retries, fallback models, and backup providers — and never hand a core API key or sensitive data to an unknown platform.

Commands

Minimal check commands

Replace the example domain with the platform's base_url and test with a low-value key. Never paste a real key in a public environment.

Model catalog

curl https://gateway.example.com/v1/models \
  -H "Authorization: Bearer $AI_GATEWAY_API_KEY"

Streaming

curl https://gateway.example.com/v1/chat/completions \
  -H "Authorization: Bearer $AI_GATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1-mini",
    "stream": true,
    "messages": [{"role": "user", "content": "Reply with pong"}]
  }'

Record

Fields worth recording

Checked atbase_urlModel nameHTTP status codeTime to first tokenTotal latencyError messageActual chargeBalance changeLogs queryable

Risk

Risk notice

This site only organizes information about AI API relay services; it makes no guarantee about third-party platforms' stability, price, data security, or availability. Judge the risk yourself before use, and avoid uploading sensitive data, personal privacy, trade secrets, or core API keys to unknown platforms.