Ollama is not running

Open Settings → Local Models, check the runtime state, and use the service start control. Review logs if startup continues to fail.

“failed to allocate buffer / kv cache”

The selected model or context length may be too large for available RAM/VRAM. OrvixCode can retry with smaller context levels.

Example context fallback
8192 → 6144 → 4096 → 3072 → 2048

CUDA / GPU error

A safe CPU fallback can keep the model usable. You can retry GPU mode later from the local model controls.

Local model is too slow

Try a smaller model, close memory-heavy applications, and confirm that compatible GPU acceleration is active.