Troubleshooting
Most issues relate to model choice or GPU memory. Use the table below to find the fix.
| Message or symptom | Cause | What to do |
|---|---|---|
| "I couldn't generate a usable answer. Please try again." | The model ran out of room (often a reasoning model spending its budget) or the question was too broad. | Ask a shorter, more specific question, or switch to a recommended non-reasoning model such as Gemma 4 Effective 4B or Qwen2.5-Coder 7B. |
| Answers get cut off, stop mid-sentence, or feel incomplete | The recent conversation is very long, a manually set local context window is too small, or the requested answer reaches the reply limit. | Start a new chat, keep the default local context window, or ask a shorter, more focused question. See Answers get cut off (context length). |
| "The configured AI model (…) rejected the request. Verify the model id…" | The model name is likely wrong or not installed. | Check the model name in Settings, or select one from the Browse… list. |
| "…rejected the request. This often means tool calling is enabled on a model that does not support tools." | The chosen model cannot run scripts automatically. | Select a recommended model, or turn off "Allow the assistant to run scripts" in Settings. You can still run scripts manually. |
| Status shows "Model not installed", or "The local AI model … is not installed yet." | A local model was selected (for example by the setup wizard) but has not been downloaded yet. | Open the Settings tab, click Browse…, and download the model — or pick one that is already installed. The status turns green once the model is ready. |
| "Knowledge retrieval is loading — please wait…" | The documentation index is starting up (first use or after a model change). | Wait a few seconds; it turns green when ready. |
| "Ready (no retrieval)" or "Knowledge retrieval is currently unavailable." | The documentation search did not start. After an unexpected shutdown, this can happen if a leftover background process is still holding the assistant's data folder open. | See Knowledge retrieval won't load after an unexpected shutdown below if Volvicon was forced to close or the data folder is "in use." If it still persists, reinstall or contact support. |
| "Your computer may be too slow for the local AI Assistant." | No graphics card was detected and the computer has less than 16 GB of system RAM, so local models would run on the CPU. | This is only a notice — you can continue, but answers may be slow. For a faster experience, select Anthropic or OpenAI in Settings. |
| "Local AI is not available on this computer. Install Local AI Support…" | The local AI engine (Ollama) is not installed or could not start. | Install Local AI Support, or select Anthropic or OpenAI in Settings. A discrete AMD or Intel GPU may also require a supported Ollama version and current graphics driver. |
| Answers are very slow | The model is large for your GPU, or it is running on the CPU. | Switch to a smaller recommended model, or use a cloud provider. |
| The assistant replies in another language | Known with some models. | Switch to a recommended model such as Gemma 4 Effective 4B or Gemma 3 12B. |
| The 3D view becomes sluggish while chatting | A large model is using most of your GPU memory. | Choose a smaller model so the GPU has room for rendering. |
| A generated script fails | Local models make occasional API mistakes. | Tell the assistant the error and ask it to fix the script, or edit it in the Scripting Editor. |
Knowledge retrieval won't load after an unexpected shutdown
To answer questions from Volvicon's built-in documentation, the AI Assistant runs a small background helper. This helper normally starts and stops together with Volvicon.
Volvicon normally clears this automatically when you reopen it. If Volvicon closed unexpectedly — for example it crashed, or it was forced to close — and the status indicator still stays on "Ready (retrieval loading…)" before changing to "Ready (no retrieval)", a leftover background helper may still be holding the assistant's knowledge files open. You may also notice that the AI Assistant's data folder cannot be renamed or deleted because it is "in use by another process" — this is the same problem.
In this state the assistant can still answer general questions, but it can no longer look things up in the Volvicon documentation, so its answers may be less accurate.
How to fix it
The simplest fix is to restart your computer, which clears any leftover background process. Then open Volvicon as usual.
If you prefer not to restart, you can close the leftover process manually:
- Close Volvicon completely.
- Press Ctrl + Shift + Esc to open Task Manager.
- Open the Details tab and look for entries named python.exe.
- For each python.exe, right-click it and choose Open file location.
- If the file location is inside the Volvicon installation folder, return to Task Manager, select that python.exe, and click End task.
- Open Volvicon again. The status indicator should return to "Ready" and knowledge retrieval should load normally.
Only end a python.exe from the Volvicon installation folder. If you are unsure, restart your computer instead — it is always safe and achieves the same result.
Answers get cut off (context length)
To answer, a local model reads your question, the documentation the assistant retrieved, and recent conversation history, then writes its reply — all within a fixed context window (the most it can read and write at once). Volvicon's bundled local AI service sets this window to 32,768 tokens by default. This provides room for a full question, retrieved documentation, and a long answer, but model capabilities and conversation length can still limit a reply.
If an answer is truncated or looks incomplete, one of these usually applies:
- The conversation has become very long. Several long exchanges can leave less room for the next answer. Start a new chat to clear the recent history sent to the model.
- The requested answer is very long. A reply can reach its output limit even when the context window has space. Ask a more focused question or split the task into smaller parts.
- You are running your own local AI service. If you started Ollama yourself instead of letting Volvicon manage it, that service's own settings control the context window, not Volvicon's. Let Volvicon start its bundled service, or adjust your service's configuration and restart it.
This context window applies only to local models. Cloud providers manage their own limits; if a cloud response is incomplete, starting a new chat or narrowing a very long request still helps.
Quick reference
- Open the assistant: title-bar AI Assistant icon, or Ctrl + Alt + A.
- Recommended local model: Gemma 4 Effective 4B (8 GB GPU). For 12 GB and larger GPUs, Gemma 4 12B or Gemma 3 12B.
- Run a script the assistant wrote: say "run this" and approve the dialog.
- Fix a broken script: tell the assistant the error and ask it to rewrite.
- Faster or lighter: choose a smaller model. Higher quality: choose a 12B model.
- Fully private: Local provider with Local retrieval.
- Integrated graphics, a low-memory card, or no graphics card: local AI runs on the CPU and is slow, so a cloud provider (Claude or OpenAI) is usually better.