Loading a Model

Supported models

Model	Size	RAM needed	Best for
Gemma 4 E2B Q4_K_M (recommended)	2.9 GB	8 GB	Fast, everyday use
Gemma 4 E4B Q4_K_M	4.6 GB	8+ GB	Better code quality

Models are saved to ~/CyberPaw/models/ by default.

---

Downloading via the UI

1. Open Settings (⚙ top-right)
2. The Models panel lists all available models
3. Models already on disk show a green ✓ Installed badge and a Load button
4. To download a new model, select it and click Download
5. After download completes, click Load — the terminal shows the loading spinner

---

Downloading via script

cd cyberpaw
./scripts/download-model.sh

---

Context window

CyberPaw auto-calculates the context window from available RAM:

budget = total_RAM × 0.75 − model_weight_size
n_ctx  = budget / kv_bytes_per_token  (rounded to nearest power of 2)
max    = 65,536 tokens

You can override this in Settings → Context Window.

---

Switching models

Open Settings, find the new model in the Models panel, and click Load. The terminal shows a loading spinner and prints Model ready. when complete. The new model path is persisted and auto-loaded on next launch.

To clear the saved model path, open Settings and click Clear next to the Active Model field.