Models, Providers, and BYOK Issues
When the wrong model runs, BYOK doesn't kick in, or a provider rejects the request.
"Model not found" or "model unavailable"
A model that worked before now errors out. Causes:
- Provider deprecated the model. OpenAI/Anthropic/etc. retire models periodically. The model picker shows a strikethrough on retired models.
- Provider outage. Check status pages — status.openai.com, status.anthropic.com, status.cloud.google.com.
- Your BYOK key lost access. If your provider account had its model access revoked (rare, usually billing-related), the key fails. Check your provider's dashboard.
- Rate limit hit at the provider. Different from Nyquest's rate limit. Provider says "you exceeded N requests/min." Wait or upgrade your provider plan.
Quick fix: pick a different model from the picker. The chassis remembers your last selection per-conversation.
Wallet was charged when I expected BYOK
Your BYOK key didn't activate for this request. Possibilities:
- Wrong provider for the model. A model from Anthropic only uses your Anthropic key, not your OpenAI key. Check the model picker — green "BYOK" badge means it's BYOK-active for that model.
- BYOK toggle off. Settings → Providers → "My Provider Mode" toggle. If off, all requests use platform-hosted regardless of your keys.
- Active provider mismatch. Settings → Providers shows which provider is set as "Active." Only the active one is used. If you have OpenAI + Anthropic keys but only OpenAI is active, Anthropic models go through platform-hosted.
- Key invalid. Try the Test button next to the key. Red X means broken; check provider dashboard or rotate.
To verify which path a request took: Settings → Usage shows per-conversation cost. BYOK requests show $0 platform cost.
"API key invalid" or test fails
The key was added but Nyquest can't authenticate with the provider:
- Wrong key entirely. Double-check you copied the full key (often 50-200+ chars). Re-copy from the provider dashboard.
- Wrong provider type. A key labeled OpenAI in your dashboard might actually be an OpenRouter key. Check the prefix:
sk-(OpenAI),sk-ant-(Anthropic),sk-or-(OpenRouter),gsk_(Groq), etc. - Key revoked. If you rotated at the provider, the old key is dead. Add the new one.
- Region restriction. Some providers restrict keys by region. Check your provider's settings.
- Permissions too narrow. Some providers let you scope keys; the key needs at minimum chat-completions or messages permissions. Check provider key settings and grant full chat scope.
Replies start, then truncate or fail mid-stream
Mid-stream failures usually mean:
- Provider connection dropped. Keep the partial response and check Usage/request status before retrying.
- Token limit hit. The model exceeded its max output tokens. The reply is truncated but kept. Re-prompt asking for a more concise version.
- Content policy block. Some providers block on completion if the output triggers their safety filter. The error message will mention it. Re-phrase the prompt.
- You hit Stop accidentally. Easy to do with rapid scrolling on mobile. The partial reply is preserved.
Switching models mid-conversation has weird results
Each conversation has its full history sent to whichever model is active. Switching models mid-conversation is fine — the new model sees the full history — but the new model's quirks/style might clash with what's already there.
For best results when switching to a much different model (e.g. cheap mini → reasoning model): start a fresh conversation with the new model. The old conversation is preserved.
Where to next
- Picking a Model — comparing models
- BYOK Setup — initial key setup
- Managing Provider Keys — key management