Get started

Download and install#

1. Open the downloads section.
2. Download the package for your operating system.
3. Install it and start Assistant.

The Windows and macOS packages are not signed. The operating system shows a warning on the first start. Confirm it to continue.

The app includes a model runner (llama.cpp) for your CPU. On macOS it uses the GPU (Metal). On a computer with an NVIDIA graphics card you can add GPU acceleration later. See Models and engines.

First run#

The setup wizard opens after the first start. It has five steps. Each step saves when you leave it.

1. Introduction.
2. Basics. Set an app-lock PIN (6 digits, optional), your time zone, your home location and the update check.
3. Inference engine. Keep the built-in model runner, or enter the address of a llama.cpp server that you run elsewhere and press Test. Select the memory model. When the built-in runner has no model, the wizard downloads a starter model (a small file, a few hundred MB). An optional TypeSafe key adds Jev, a decision model for group conversations.
4. Starter personas and recipes. Select the ones you want.
5. Done.

Skip setup keeps the defaults. You can change every value later in Settings.

Your first chat#

1. Wait until the status bar shows the engine online. The first request loads the model, which takes a moment.
2. Type a message and press Enter.
3. The reply streams in. Press Esc to stop it.

Open the conversation settings panel (View → Conversation Settings Panel) to select another model or change the prompt. See Chat.

Add a model#

Models are files in the GGUF format. Open Settings → Engines, open the Built-in llama.cpp engine and use the Download list. Or copy a GGUF file into the models folder shown there. See Models and engines.