Base URL mismatch
The request may be sent to a direct provider endpoint, local proxy, staging gateway, or another OpenAI-compatible workspace.
TackleKey
model_not_found debugging
Debug model_not_found before rewriting your SDK integration
The error often looks like a provider outage, but it is usually a mismatch between the base URL, API key, model ID, account permissions, or fallback route.
Common causes
Wrong account context
The API key may be valid but belong to a workspace that cannot access the model you copied.
Display name vs model ID
A UI label such as GPT mini or Claude Haiku may not be the copy-ready API model ID.
Fallback hides failure
A fallback route can make the visible error less useful unless logs show the exact model and route attempt.
Debugging order
| Step | What to check | Entry |
|---|---|---|
| 1. Pair endpoint and key | Confirm the base URL and API key come from the same gateway or account context. | Open |
| 2. Copy the exact model ID | Use the current model directory rather than a display name or old docs snippet. | Open |
| 3. Send one tiny request | Use cURL before debugging SDK wrappers, streaming, tools, agents, or RAG code. | Open |
| 4. Check request logs | Verify which model ID was received, whether the account had access, and which status was returned. | Open |
The fastest test is not changing providers. Prove one minimal request with the exact base URL, API key, and model ID that production will use.
FAQ
Is model_not_found always a model outage?
No. It is often a configuration mismatch, stale alias, missing account permission, or request sent to the wrong endpoint.
Should I try another SDK first?
Usually no. First send one raw cURL request so SDK wrappers are not hiding the status code or request body.
Where should I copy the model name from?
Copy the exact API model ID from the same model directory or pricing page used by the gateway you are calling.
Try TackleKey with one OpenAI-compatible change
Create an account, generate a project API key, then replace your client base URL with the TackleKey endpoint. Keep keys server-side and verify live pricing before scaling traffic.