Reference/Models & engine
Models & engine
A router model reads your commands and a summary model summarizes documents and the web. Both run in-app on Apple Silicon with MLX.
| Model | Role | Size | Recommended for |
|---|---|---|---|
| Alchemist-nano 0.8B | Command parsing | 0.4 GB | 8 GB Macs |
| Alchemist-mini 2.6B | Command parsing + summaries | 2.9 GB | 16 GB and up |
| Qwen3 1.7B | Document & web summaries (optional) | 1.0 GB | Alongside the 0.8B |
Switch in Settings ▸ Models; new models can be installed from the card. With Memory saving mode on, the summary model unloads after 3 idle minutes and reloads on demand. If an MLX model cannot be used, the reason is shown at the top of Settings ▸ Models.