The agent feels slow

If responses are taking noticeably longer than expected on the local (Ollama) backend, the most common cause by far is that your GPU isn't actually being used — the model silently fell back to running on your CPU.

Why this happens

Outdated or missing GPU drivers are the usual culprit. Ollama needs a driver that supports GPU acceleration for your specific card; without one, it still works, just far slower, and there's no loud error telling you why — it just feels sluggish.

Fix

1. Update your GPU drivers to the latest version from your manufacturer (NVIDIA, AMD, or Intel).
2. Restart your machine.
3. Try the agent again — a noticeable speed jump confirms it's now using your GPU.

If you don't have a dedicated GPU

CPU-only is genuinely slower for local models — that's a hardware limit, not a bug. The setup wizard's hardware detection should have already steered you toward a smaller model suited to CPU-only use; if it feels wrong for your machine, switching to a smaller model in Settings will help. Alternatively, the Claude API backend (see the setup wizard) moves the work off your machine entirely.