Common Error Messages
Error messages describe a request or connection problem; they do not always mean the provider did no work. Check the saved result and accounting state before generating again.
Before retrying a generation
A connection error does not prove the provider did no work. Use this recovery table before submitting the same operation again:
| Outcome | Next step |
|---|---|
| Rejected before dispatch (invalid input or authentication) | Correct the request and submit it again |
| Rate/capacity rejection | Honor Retry-After and back off |
Answer returned with x-nyquest-settlement: pending | Keep the answer; check Usage rather than generating again |
| Audio/media settlement pending | Check Usage and the result before retrying; contact support if it remains unresolved |
Video submitting, running, or requires_reconciliation | Poll the existing job ID; do not create another job to recover the first |
| Ordinary chat shows Incomplete answer or Answer unavailable | Use Reload saved answer when offered, then check its recorded status and Usage |
| Splicer loses its connection, stops, or has no synthesis | Read the available model responses and use Reload saved run to check the same run's outcome and billing |
| Stream interrupted or provider outcome unknown | Check history, Usage, and job status; a new request may generate and charge again |
Nyquest does not provide a universal client idempotency-key/response-replay guarantee. Confirmed video submission rejection or confirmed render failure can release a hold; an unknown provider outcome needs reconciliation first. Contact support with the request/job ID if it stays unresolved.
Error messages in Nyquest follow a consistent format: a short title, a one-sentence explanation, sometimes a suggested action. This article maps the most common ones to their root causes.
"Incomplete answer" or "Answer unavailable"
Where: below an ordinary-chat response, including when reopening saved history. Means: the response ended before completion was confirmed, or no completed answer was produced. The notice distinguishes causes such as a model output limit, timeout, provider failure, or interrupted connection when that information is available. Received text and images are retained. Fix: use Reload saved answer when offered. This checks the stored result without another model request; it does not resume generation. If it is not available yet, wait briefly and reload again. Check Usage before submitting another request. See Streaming.
Splicer partial answers, failed synthesis, or interrupted connection
Where: the Splicer result and individual model-response tabs. Means: a model may have stopped early, fewer than two complete answers may be available, or synthesis may not have completed. A Partial model response remains readable but does not count toward consensus. A synthesis with partial coverage used only the complete answers from part of the selected group. Fix: read the available responses, then use Reload saved run to check its recorded outcome and billing. Check Run details for the run ID and reason. If details cannot load, keep the displayed text and retry loading the same run. Charged, Still reserved, and Billing describe accounting separately from answer completion. A stopped or failed run can still have charges or a reservation awaiting reconciliation. See Multi-Model Splicer.
"Insufficient balance"
Where: chat input shows it on send, agent runs reject it on start. Means: the wallet cannot cover the request or its required reservation. An unresolved hold can reduce available funds even when earlier work did not produce a complete answer. Fix: check Usage and any outstanding reservations, then add funds through the wallet controls if needed. For Splicer, a smaller model selection or shorter context may reduce the reservation. BYOK availability depends on the request and provider; see Wallet and Funds.
"Rate limit exceeded"
Where: chat input rejects the send. Means: you've sent more requests than the per-minute cap allows. Limits:
- Free tier: 20 chat requests/min, 10 auth requests/min
- Pro tier: 120 chat requests/min, 10 auth requests/min
Fix: wait 60 seconds. If you regularly hit it, fund the wallet (becomes Pro) or talk to a real human.
"Model not found" or "Model unavailable"
Where: model picker shows strikethrough or chat fails on send. Means: the model is retired or your provider revoked access. Fix: pick a different model from the picker. See Models, Providers, and BYOK Issues.
For API callers, check both the JSON model and x-nyquest-model header: the header takes precedence. An unavailable pinned model rejected before provider submission has a no_charge outcome; distinguish that rejection from a request that was already submitted.
"API key invalid"
Where: Settings → API Keys (Test button), or chat fails on a BYOK provider. Means: Nyquest can't authenticate with your provider using the saved key. Fix: verify the key at the provider's dashboard, rotate if needed, update in Nyquest.
"Provider unreachable" or "Provider error 5xx"
Where: chat fails mid-stream or right after send. Means: the upstream model provider had a transient outage. Fix: preserve the response/request ID and check Usage or job status first. Retry only once you have distinguished a rejected request from pending settlement or an unknown provider outcome. See the recovery table above.
"Conversation not found"
Where: clicking a conversation link. Means: the conversation was deleted or belongs to a different account. Fix: none — sign in to the right account, or accept it's gone if deleted.
"Project not found"
Where: clicking a project URL or refresh. Means: the project was deleted or you're on a different account. Fix: check Settings → Projects for your active project list.
"Forbidden" or 403
Where: admin endpoints, or rare auth-related calls. Means: your account doesn't have permission for that action. Fix: if you're trying to access /admin, you need admin status — only granted by request. Otherwise, this usually indicates a session issue: sign out and back in.
"Unauthorized" or 401
Where: any backend call after the JWT expires. Means: your session expired (default: 8 hours after sign-in). Fix: sign out and sign back in.
"CORS error" (in browser console)
Where: browser dev tools. Means: a frontend API call is being blocked by CORS policy. Almost always indicates a deploy mismatch. Fix: hard-refresh the page (Cmd/Ctrl+Shift+R). If persistent, talk to a real human — could be a backend deploy issue.
"Service Unavailable" or 503
Where: any endpoint. Means: the backend is briefly down (deploy in progress, restart, etc.). Fix: inspect the error type and the recovery table above. A capacity rejection can be retried after backoff; pending settlement or unknown completion needs a status check first. If unresolved, contact support.
"Network error" / "Failed to fetch"
Where: browser console or chat input.
Means: your network couldn't reach api.nyquest.ai at all.
Fix:
- Check your internet connection
- Disable VPN if you have one (some block our domain)
- Try a different network
- Check status.nyquest.ai (when deployed)
"Image generation failed"
Where: image-capable model returns no image. Means: the upstream image model errored — usually content policy or provider outage. Fix: rephrase the prompt (avoid trigger words), or try a different image model.
"Memory recall service unavailable"
Where: chat completes successfully but no recall happened. Means: the memory layer was briefly unavailable. The chat itself worked; recall just didn't trigger. Fix: provide any important context in your message and try recall again on a later turn. The current response may lack relevant remembered context; do not assume it used those memories.
"Discord support unavailable"
Where: /help/live-support shows a 503 or "service disabled" message. Means: the Discord bot is offline or env vars are misconfigured. Fix: talk to support via email (the email path documented in Contacting Support). This is rare; usually means a deploy hiccup on the backend.
Where to next
- Sign-in Issues — auth-specific
- Billing Issues — wallet-specific
- Models and Providers — model-specific
- Contacting Support — escalation