Download a local model
- Download a local model to run AI on your machine, with visible progress and the option to re-download or update later.
- Browse and pick from a model catalog, with guidance on each model’s size and whether it fits your device’s memory.
- See your device profile (memory, GPU, CPU cores) to judge which models you can run.
- Models published in several parts download and run as a complete model. The progress indicator names the part currently downloading.
1
Open Settings → Model
Open Settings from the account menu at the bottom left of the sidebar, then choose the Model section.
2
Pick a model that fits your Mac
Browse the model catalog: each entry shows its size and whether it fits your device’s memory. Your device profile (memory, GPU, CPU cores) helps you judge what you can run.
3
Download it
Click Apply to start the download and watch the progress. For multi-part models, the indicator shows which part is downloading. You can re-download or update a model later at any time.
Backend (llama.cpp)
In Settings → Model, Backend is the engine that runs a downloaded local model on your Mac. GGUF (llama.cpp) is the default native backend on the desktop app. It loads GGUF model files from the catalog and runs them on your device. When you use a downloaded model with this backend, replies are generated on your machine, not in the cloud.Local engine on first install
On a first-time install, the local engine downloads a complete, known-good copy so fresh installs can run local models instead of rejecting a newly downloaded engine.Context window
The context window is how much of a conversation the model can keep in mind at once: your messages, its reply, and any attached content for that turn. It is measured in tokens. A token is a small piece of text, often a word or part of a word. 10K, 32K, and 128K mean about 10,000, 32,000, and 128,000 tokens. A larger window can hold a longer thread before older turns fall out of view. Larger windows also use more memory, which is why webAI sizes the window to your device tier.How models load
When you use a downloaded local model, webAI sizes the context window to your Mac’s memory and estimates memory for that model’s architecture. The size follows your device tier.- Tier 1 (under 16 GB, typically 8 GB Macs): a model starts at a 10K context and can grow to 32K.
- Tier 2 and Tier 3: a model starts at 32K and can grow to 128K.
- Tier 4 (64 GB and above): a model starts at 128K.
- Vision support loads when you send the first image, instead of blocking the initial load.
- Capability defaults and the fallback model are matched to your machine. If a load fails, webAI drops to a lighter model rather than a heavier one.
Personas
Under Settings → Model, open the Personas list to review the personas on your device. Click the pencil icon next to a persona to edit it. For creating personas, chatting on the Personas tab, and the full persona guide, see Personas.Common questions
More answers are in the FAQ.