A local model runs on your own computer, so your text never leaves the machine. This is the option for material that cannot go to a hosted provider.
Ollama
Install and start Ollama on your computer, then enable the Ollama provider in Settings under Models on the desktop app. No key is needed. Click Add Model under a list and Grape lists the models you have pulled: chat models under Chat Models, embedding models under Indexing Models. If it says it could not reach Ollama, Ollama is not running.
LM Studio
Start LM Studio's local server, enable the LM Studio provider, and add your loaded models the same way, from Add Model. Grape talks to LM Studio at its standard local address.
Desktop only
Local providers are a desktop feature, since the model has to run somewhere with the memory to hold it. Phones use Grape AI or a hosted provider instead.
What to expect
A local model is as good as the hardware and the model you picked. Expect slower answers than a hosted provider and quality that varies by model. For summarizing and chatting over notes, mid sized local models are genuinely usable, and the privacy trade is the whole point.
Mixing local and hosted
Providers are switches, not a commitment. Some people run chat locally and keep a hosted model enabled for transcription, which local setups rarely handle well.