Troubleshooting
"Couldn't load models. Check that a service is configured."
You'll see this in the model picker if no service is connected yet. Go to Services and connect at least one option — an OpenRouter key is the quickest to set up.
Ollama (Local) shows no models, or fewer than expected
The managed catalog only shows models your instance has enough RAM to run. See Ollama (Local) hardware guidance — increasing the VM's memory allocation will unlock larger models.
GPU acceleration warning
If Ollama reports it's running on CPU only, confirm your hypervisor is actually passing a GPU through to the VM. This doesn't stop local models from working — it just makes them slower.
Ollama connection (local or remote) shows "Unreachable"
- Managed mode: try Stop then Start again in Services.
- External/remote mode: confirm the address is correct, that Ollama is actually running there, and that no firewall is blocking the CrabShack VM from reaching it.
Ollama Cloud test fails
Re-check the API key was pasted in full, and confirm your Ollama Cloud account/subscription is active.
ChatGPT (Codex) connection code expires
Device codes are short-lived. Click Connect again to get a fresh code and link.
A model download won't start or stalls
CrabShack checks free disk space before starting a download. If you're told there isn't enough room, delete a model you no longer use or increase the VM's disk size.