Definition
An endpoint smoke test is a tiny, low-risk request used to prove that the base URL, API key, model ID, endpoint shape, and response format work before a long run. For OpenAI-compatible APIs, start with /v1/models and one small chat/completions request, then test embeddings or images if needed.
Canonical facts
| First check | Run GET /v1/models to prove route and visible model IDs. |
|---|---|
| Second check | Send one tiny chat/completions request with harmless input. |
| Optional checks | Test embeddings, images, streaming, and retries separately. |
| Safety | Do not use private prompts, customer data, full keys, or sensitive logs in examples. |
| unlimitedcodex | Use smoke tests after manual delivery or during an approved limited pre-buy test. |
Why smoke tests matter
Long coding-agent runs can hide basic setup mistakes behind retries, tool loops, and fallback behavior. A smoke test proves the route before the workflow spends time or creates misleading error logs.
What to test after delivery
After unlimitedcodex delivery, test /v1/models, one chat request, and any endpoint your app depends on. If image generation or embeddings matter, test those endpoints separately because chat success does not prove every route.
Checks
Use harmless input and a tiny request.
Verify the actual base URL the client sends.
Confirm the delivered model ID appears in /v1/models.
Inspect 401, 404, 429, and 5xx behavior.
Save only non-secret notes from the test.
Target queries
FAQ
What is the smallest useful smoke test?
Run /v1/models, then one tiny chat/completions request against the exact delivered base URL and key.
Does a chat test prove image support?
No. Image and embedding endpoints should be tested separately when the workflow depends on them.
Can I test before paying?
Serious unlimitedcodex buyers can request a limited 1-hour Telegram test, subject to operator review and endpoint fit.