Getting started.

1. Download the app

Get the installer for your computer from the download page:

• WindowsWindows 10 or 11, 64-bit. A Setup .exe.
• macOSApple silicon Macs with macOS 14 or newer. A .dmg, or a .zip if you prefer. There is no Intel build yet.
• Linuxx86_64. A .deb for Ubuntu and other Debian-family systems, or a portable AppImage.

Each download lists a SHA-256 checksum. Checking it is optional, but it is the easiest way to be sure the file arrived intact.

2. Install and open it

Preview builds are not signed yet, so your system will warn you the first time you open one. That is expected:

• WindowsIf SmartScreen says “Windows protected your PC”, choose More info, then Run anyway, and follow the installer.
• macOSDrag VARIANT-1 into Applications and open it. If macOS stops it, go to System Settings → Privacy & Security, scroll down, and choose Open Anyway. On macOS 14 you can also Control-click the app and choose Open.
• LinuxInstall the deb with sudo apt install ./VARIANT-1-*-linux-amd64.deb, or make the AppImage runnable with chmod +x and open it. If it won’t start because FUSE 2 is missing, put APPIMAGE_EXTRACT_AND_RUN=1 in front of the command.

Python and the other runtimes the app needs come bundled. Updates are manual for now: quit the app, back up its data folder, and install the new version over the old one.

3. Connect a model

VARIANT-1 doesn’t include a model; you choose one. Open Settings → Providers and pick whichever route suits you:

• An account you already haveSign in with a subscription such as ChatGPT, or use a desktop service you are already signed in to, such as Ollama.
• An API keyPaste a key for OpenAI, Anthropic, Google Gemini, OpenRouter, xAI, DeepSeek, or one of dozens of other built-in providers.
• A model on your own machinePoint VARIANT-1 at a GGUF model file and let it run llama.cpp for you, or connect LM Studio.

Models & connections compares the routes, including who you pay.

4. Try a first task

Start small, with something you can check by hand. Your first task has a ready-made example: four numbers, a total, and a follow-up that builds on the same data.

When you move on to your own files, work on copies and say where new output should go. A fresh folder with just the files you need keeps things tidy.

5. Add extras when you need them

The browser may ask for a short setup the first time a task uses it. Voice, the Telegram and Discord messaging gateway, and plugins are optional and live in Settings. Speech engines and model files are not bundled, so voice needs a speech service of your own.

Building from source instead? Follow the setup guide in the repository.