Getting Started

Getting Started

This page takes you from nothing to a model answering in the chat.

What you need

• Windows 10 or 11, 64-bit.
• A llama.cpp build that includes llama-server.exe. RulZ does not bundle an engine; it runs yours. See Building the Engine, or download a prebuilt Windows release from the llama.cpp releases page, unzip it, and point RulZ at the folder.
• One or more GGUF model files in a folder. The default folder is C:\LLM; you can change it.
• An NVIDIA GPU is strongly recommended. Models also run on the CPU, but much more slowly.

Install

Option 1: the installer

1. Open the latest release and download RulZ-LLM-Suite-Setup-<version>.exe.
2. Run it. It installs for your user account and lets you pick the install folder.

The installer is not code-signed, so Windows SmartScreen shows an "unrecognized app" warning the first time. Choose More info → Run anyway.

Option 2: from source

Install Node.js 22.12 or newer, then in the project folder:

install.bat     # installs dependencies, downloads Electron and builds the app
run.bat         # starts the app

run.bat build rebuilds before starting; run.bat dev starts with hot reload. See Development for more.

First launch

1. Choose your look. A dialog asks for Dark or Light; dark is preselected. You can change it any time with the sun/moon button in the top bar or in Settings → Appearance.
2. Check the engine. Open Settings (Ctrl+5). Under Engine you should see your llama-server version and your GPUs. If it says not available, set the folder that contains llama-server.exe and press Test. The app also looks in C:\coding\llama.cpp\build-cuda\bin\Release and ...\build\bin\Release by itself.
3. Check the models folder. Under Settings → Models, or on the My Models page, make sure the folder is the one that holds your .gguf files. Changing it rescans immediately.

Load a model and chat

1. Press Ctrl+L (or click the model box in the top bar).
2. Pick a model on the left. The defaults are sensible: an 8K context, GPU offload set to Auto, and flash attention on Auto. The panel shows a memory estimate so you can see whether it will fit.
3. Click Load model. A progress bar in the top bar follows the load; the small file icon next to it opens the live server log if something goes wrong.
4. Type a message and press Enter.

Good to know:

• If you pick a model and send a message without loading it first, RulZ loads it for you with its saved settings.
• Turn on Remember settings for this model to keep the settings you chose for next time. See Loading Models for every option.
• By default, loading a second model unloads the first. You can change that in Settings → Models.

Where to go next

• Chat — everything the chat window can do.
• MCP Tools — let the model use tools.
• Local API Server — use your models from other programs.
• Troubleshooting — if something did not work.