The Assistant page shows a toggle between On my device and Cloud provider. Switching to cloud restores whichever cloud provider you last used, so you do not lose your choice.
Built-in engine (offline)
The built-in engine is llama.cpp, bundled with the app. It downloads on first use. Models are downloaded from the Models page. Two settings appear only for this engine:
A bigger context window holds more conversation history and makes screen vision and web search more reliable, at the cost of more memory. The model unloads after idle time to free RAM and VRAM, then reloads on next use.
Ollama and LM Studio
Pick the Local provider and point it at your server’s address. Ollama defaults tohttp://localhost:11434/v1, LM Studio to http://localhost:1234/v1. No API key is needed.
Cloud providers
Every cloud provider is preconfigured. Pick it from the list and enter your key. The full list:
Click Load models to fetch the provider’s model list, or type any model name directly. The picker accepts values that are not in the list.
API keys
Keys are stored in your operating system’s keychain, not a settings file. Each provider keeps its own key, and switching providers does not lose one you already entered. See Privacy.Shared provider
The assistant, AI cleanup, and Generate with Flow share the same provider and model. Picking a provider on the Assistant page changes it for all three, unless AI cleanup has its own model set.Apple Intelligence appears in the provider list for AI
cleanup on Apple silicon Macs, but not for the
assistant.
Assistant master switch
The Assistant page has an on/off switch in its header. Turning it off:- Removes the assistant panel window entirely (it is destroyed, not hidden).
- Disables both assistant shortcuts (the quick ask and the conversation). They do nothing.
- Stops the built-in llama.cpp engine instead of leaving it idle.
- Hides Profiles and Memory from the Assistant page.
Testing the connection
A Test connection button sits at the bottom of the provider group on the Assistant page. It sends a throwaway question and reports whether the model answered. Use it after entering a new key or changing a model to catch a bad key, an empty balance, or a model id the endpoint does not serve, before the first real ask fails silently.Custom models
Add your own GGUF files or link a folder.
Quick ask
How to use the assistant once it is set up.