Help
AI connection errors and how to fix them
Most AI errors in PlotWrite come from your AI provider, not from PlotWrite, and one of the seven fixes below solves them. PlotWrite (plotwrite.ink) is a web app for LitRPG and progression fantasy web serial authors that runs on your own AI: your ChatGPT or Claude plan through the connector, an API key, OpenRouter or a local model. Each section explains what happened, why, and the exact steps to fix it.
Model not found
What it means
The AI model your project is set to use is no longer available on your provider.
Why it happens
Providers regularly retire older models (Google deprecates Gemini 1.5 family, OpenAI rotates GPT versions, OpenRouter delists models). The model name your account remembered is now invalid.
How to fix
- Open Settings → AI connections, open the three-dot menu on your provider and choose Refresh model list. plotwrite.ink will pull the latest model list directly from your provider.
- Pick a current model for each of the four task slots (Quick tasks / Creative tasks / Deep analysis / Background tasks).
- Save and retry your last action.
- If you leave a slot empty, plotwrite.ink uses a built-in default model for your provider, which can also be retired over time, so picking a current model yourself is the most reliable fix.
API key invalid or expired
What it means
Your provider rejected the API key with a 401 or 403 error.
Why it happens
Either the key was revoked, the project it belongs to was disabled, your billing lapsed, or the key was rotated and not updated here.
How to fix
- Log in to your provider's dashboard and confirm the key is still active.
- If it was rotated, copy the new key and update it in Settings → AI connections (Replace key, in the three-dot menu on your provider).
- If billing was suspended, restore it, then click Test next to your provider in Settings → AI connections.
No credits left
What it means
Your provider returned a 402 / quota / billing error.
Why it happens
BYOK costs are paid directly to your provider. Your prepaid credits or monthly cap have been used up.
How to fix
- Top up your provider account (OpenRouter, Anthropic, OpenAI, Google, Groq, etc.) directly.
- Once funds are added, retry. No need to re-add the key.
- Optional: switch your active provider to one that still has credits.
Rate limited
What it means
Provider returned 429: too many requests in a short window.
Why it happens
Most providers throttle free tiers and lower paid tiers (Google AI Studio free tier is especially aggressive). Many parallel generations triggered the cap.
How to fix
- Wait 30–60 seconds and try again.
- If it keeps happening, upgrade your provider plan or switch to a higher-throughput provider like OpenRouter or Groq.
No models available for this provider
What it means
We couldn't fetch any usable models for the selected provider.
Why it happens
Either the provider's /models endpoint is down, your key has no model access, or the model list never finished syncing.
How to fix
- Open Settings → AI connections, open the three-dot menu on your provider and choose Refresh model list.
- If it still shows zero models, click Test next to your provider. Your key may not have access — check your provider account permissions.
AI provider unreachable
What it means
We could not reach the provider's API at all.
Why it happens
Usually a transient network issue, the provider being down, or (for LM Studio) the local server isn't running.
How to fix
- Try again in a moment.
- For LM Studio: confirm the LM Studio app is running with the server enabled on http://localhost:1234.
- Check the provider's status page if it persists.
Unknown AI error
What it means
The provider returned an error we don't have a specific recipe for yet.
Why it happens
Providers occasionally return new error shapes. The toast description has the raw provider message — that's usually the fastest clue.
How to fix
- Read the toast description for the provider's exact message.
- Try refreshing models and switching to a different model in Settings.
- If the issue persists, contact support@plotwrite.ink with the message text.