The app tries to tell you exactly what went wrong instead of a generic "request failed" — these
are the real messages it shows, grouped by provider and cause.
This is the most common one — it's a billing issue, not a bad key. Anthropic requires a funded
account before a key can make requests at all.
The key itself was rejected (typo, revoked, wrong account).
Same message, with the provider name swapped in — shown on a 401, an "Incorrect API key" response,
or (NVIDIA only) a 403. NVIDIA reports invalid keys with 403 instead of 401, which the app
specifically accounts for so this message still shows correctly there.
Shown when NVIDIA is selected but no key is configured at all.
On a 429 / rate_limit_exceeded response.
On a 404 / model_not_found — usually a typo'd custom model id typed into the picker's custom-id
field.
Ollama errors surface as Ollama stream error: {status} or Ollama complete error: {status} —
almost always because the requested model was never actually pulled. See
Ollama not responding for that specifically.
Shown when nothing at all is configured — every free-tier fallback (Anthropic → NVIDIA → OpenAI →
Ollama) came up empty.