Skip to main content

5.1 Moss not responding

Typical symptoms

  • The workbench spinner never finishes after you send a message.
  • The composer shows “Fast / thinking model not configured correctly.”
  • Local Ollama reports the service is unreachable.
  • Remote models return key errors, “model not found,” or timeouts.

Check composer banners first

AI Dock surfaces model setup issues proactively:

BannerWhat to do
Fast / thinking points at local Ollama but the service is downOpen 3.12 Local LLMs and start the service
Model missing from local Ollama listPull the matching tag on the Local LLM page, or switch to a full model name from the list
Remote model incompleteOpen Settings → AI engine and fill in model name, Base URL, and API key

Remote-model checks

Go to Settings → AI engine and test the current thinking or fast model:

ErrorCommon causes
Key error or insufficient permissionWrong or expired API key, or missing role
Model not foundWrong model id, or wrong Base URL
Connection timeoutNetwork, proxy, corporate firewall, or overloaded endpoint
Provider errorProvider does not match the platform you configured

When unsure what is wrong, share the vendor’s field guide, the test error (with keys redacted), or a screenshot with keys hidden—do not paste API keys into chat.

Local model checks

If you use Ollama:

  1. Open Local LLM.
  2. Confirm the runtime is installed.
  3. Confirm the service is running.
  4. Confirm the model name you configured exists in the list.
  5. Run the model test.

More detail: 5.13 Local LLM / Ollama issues.

Device offline is not “Moss silent”

When the device is offline, Moss can still plan and answer from docs; board-side commands wait for reconnect or extra confirmation. If you asked for execution, check whether the device chip shows online first.