Start here
Choosing a model
How the app decides what fits, and how to pick between models.
The Models screen lists open chat models from Hugging Face. For each one the app works out how it would run on your computer before you download anything.
What the labels mean
| Label | Meaning |
|---|---|
| Runs well | Fits in your graphics card or memory with room to spare, at a comfortable speed. |
| Runs slowly | It works, but it's tight, partly on the processor, or slower than about 5 tokens a second. |
| Won't fit | Needs more memory or disk than this computer has. The reason is shown. |
Versions
Each model comes in several versions (quantisations): the same model stored more compactly. Smaller versions are faster and need less memory; larger ones are slightly smarter. The app picks the version that fits your computer best; you can choose another on the model's page.
Rules of thumb
- 8 GB memory: 1B to 4B models for quick everyday answers.
- 16 GB: 7B to 12B models. A graphics card makes them several times faster.
- 32 GB or more: larger and mixture-of-experts models, and reading images alongside chat.
- Models marked Uses tools work best with web search and the browser.
- Models marked Reads images can look at photos and screenshots you attach.
Context window
The context window is how much of the conversation the model can see at once. The app sets it automatically (32K, 16K or 8K) from your memory. You can change it in Settings → Model settings.