AI API Relay Online Checker
Use your own API base URL and key to inspect model discovery, protocol compatibility, and streaming first-byte latency. Keys are not stored in the database.
Test cost
Model requests use a very small output limit but may still create a small charge at the target relay provider.
Key safety
Use a temporary, low-balance key. Do not test with a long-lived key that has broad permissions or substantial credit.
What each check means
| Check | Can show | Cannot prove |
|---|---|---|
| Base URL | HTTPS endpoint shape and reachability | Who operates the upstream or whether it is safe |
| Model discovery | Whether the key can read a model list | Whether a listed model can generate a response |
| Generation request | Protocol completion for the selected model | Model identity or long-term availability |
| Streaming timing | First event and total response time for one request | Capacity during other hours or sustained concurrency |
| Usage fields | Reported input, output, and cache fields when present | Accurate wallet billing without a balance reconciliation |
Recommended workflow: create a temporary low-balance key, select the exact model and protocol you intend to use, run the smallest request, save the result, and revoke the key. Do not paste an administrator key or a key with substantial credit.
Frequently asked questions
Is the API key stored?
The detector sends the temporary key only for the current server-side check and does not write it to the monitoring database. Use a newly created, low-balance key and revoke it after testing.
What can the online checker prove?
It can show whether a base URL and key complete the selected protocol and model request at that moment. It cannot by itself prove model identity, long-term reliability, privacy practices, or correct billing.
Why can model discovery pass while generation fails?
The model-list endpoint and generation gateway may use different permissions, groups, upstreams, or rate limits. A listed model is therefore not counted as verified until a generation request completes.
How much does a check cost?
The generated response is capped at a very small token count, but the relay may still charge for input, output, cache, or a minimum request amount. Check the provider's billing rules first.