Use a local AI model on your Mac
Keyprofile can send rewrites to LM Studio, Ollama, or llama.cpp running on your Mac. Download and manage models in your server app. Choose a model that runs locally to keep your writing on your Mac.
If you need a model, start with Qwen3 4B Instruct (2507), 4-bit. It is a compact option for everyday rewrites.
With LM Studio
- Open LM Studio and load a model. To try the recommended model, search for
Qwen3-4B-Instruct-2507in Discover and download the 4-bit version. - Start the server. Open the Developer view in LM Studio and switch the server on, or run this in Terminal:
lms server start
LM Studio serves on port 1234. The server address to enter in Keyprofile is http://127.0.0.1:1234.
With Ollama
Install Ollama, then fetch a model:
ollama pull qwen3:4b-instruct
That is all. Once Ollama is installed and a model is pulled, it serves automatically on port 11434. The server address to enter in Keyprofile is http://127.0.0.1:11434.
With llama.cpp
Install llama.cpp (for example brew install llama.cpp), download a GGUF model file, and start its server with that model:
llama-server -m ~/models/your-model.gguf --host 127.0.0.1 --port 8080
For the recommended model, download the Qwen3 4B Instruct Q4_K_M GGUF file.
The server address to enter in Keyprofile is http://127.0.0.1:8080. Leave that Terminal window running while you use it; a llama.cpp server runs the one model it was started with, so changing models means restarting it with a different file.
Connect Keyprofile
- Open Keyprofile Settings → Writing and choose Local providers under Where rewrites run.
- Choose your server’s preset or enter its address. If your server requires a key, enter it in Server key.
- Choose Connect to list the models your server offers.
- Pick a model, then try a rewrite on a short selection to confirm it works.
When something does not connect
- The server is not running. Keyprofile can only connect while your server is running on this Mac. Start it (in LM Studio, switch the server on in the Developer view; for llama.cpp, the Terminal window with your
llama-servercommand needs to stay open), then choose Connect again. - The port does not match your server. Check the address shown by your server app. The presets use ports 1234 for LM Studio, 11434 for Ollama, and 8080 for llama.cpp.
- LM Studio is running but has no model loaded. LM Studio can run its server with nothing loaded, and then there is no model for Keyprofile to pick. Load a model in LM Studio first, then connect again.
If the model takes too long or returns an empty rewrite, your text stays unchanged. Try a shorter selection or a different model. If your server app has a reasoning setting, try turning it off for rewrites.
Once connected, follow the guide to rewriting selected text on a Mac to try your first rewrite. For help connecting a model, email hello@keyprofile.app.