Installation

One line, no git, no admin, no compiler. Re-run it any time to update.

Windows (PowerShell):

irm https://raw.githubusercontent.com/dadwritestech/LlamaForge/master/install.ps1 | iex

Linux / macOS:

curl -fsSL https://raw.githubusercontent.com/dadwritestech/LlamaForge/master/install.sh | sh

The installer:

1. finds Python 3.10+ (the backend is pure standard library, nothing to pip install). On Windows, if you have none, it drops a private, SHA-256-pinned copy of python.org's embeddable Python;
2. downloads the latest LlamaForge release into %LOCALAPPDATA%\LlamaForge (Windows) or ~/.local/share/llamaforge (Linux/macOS). Updating keeps your config, models and engines;
3. adds Start menu and desktop shortcuts and an Apps & Features uninstaller (Windows), a llamaforge command plus an app-menu entry (Linux), or ~/Applications/LlamaForge.app (macOS);
4. starts LlamaForge and opens the dashboard.

In the dashboard, click Install llama.cpp. It fetches the official llama.cpp release for your GPU (CUDA, Vulkan, Metal or CPU), verifies it and starts the router. Then continue to First Run.

Installer options

Set these environment variables before running the one-liner:

Daily use

Open LlamaForge from the Start menu or app menu, or run llamaforge. It starts the llama.cpp router and the dashboard if they aren't already running, then opens your browser at http://127.0.0.1:8090.

To shut down the dashboard, the router and the models it spawned, run llamaforge stop (Linux/macOS) or stop.ps1 in the install directory (Windows). Only processes LlamaForge started are stopped; any other llama-server you run is left alone. On Windows it also stops a vllm serve it started inside WSL.

Start at login

Nothing starts on its own unless you ask. llamaforge service on (in a git checkout: python3 backend/service.py on "$PWD") writes ~/.config/systemd/user/llamaforge.service on Linux or ~/Library/LaunchAgents/io.github.dadwritestech.llamaforge.plist on macOS. Either one runs run.sh with LLAMAFORGE_NO_BROWSER=1 when you log in, so it starts the same router and dashboard the launcher does, and the dashboard still listens on 127.0.0.1 only. On Linux, systemctl --user start llamaforge starts it right away, and loginctl enable-linger $USER makes it start at boot before anyone logs in. llamaforge service status shows whether it's set up. llamaforge service off removes the file but leaves a running copy alone (stop that with llamaforge stop). llamaforge uninstall turns it off as well.

On Windows, tick Setup > Startup > start LlamaForge when you log in (the same switch works on Linux and macOS). It writes LlamaForge.vbs into your Startup folder (%APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup), which runs run.ps1 hidden with LLAMAFORGE_NO_BROWSER=1 when you log in. No admin rights and no scheduled task. Untick it, or delete that file, to turn it off; a running copy keeps running until stop.ps1. The uninstaller removes it too. From a terminal: python backend\service.py on|off|status .

Uninstall

Windows: Settings → Apps → Installed apps → LlamaForge → Uninstall. Linux/macOS: llamaforge uninstall. Both ask before removing your settings or models, and only remove files the installer put there.

From source (git clone)

Use this if you want to hack on LlamaForge or build llama.cpp from source.

git clone https://github.com/dadwritestech/LlamaForge
cd LlamaForge
powershell -ExecutionPolicy Bypass -File bootstrap.ps1   # Windows
./bootstrap.sh                                           # Linux / macOS

The bootstrap script checks for Python and Git (asking before installing anything), asks whether to use an existing llama.cpp checkout, writes config.json and a starter models.ini, then launches the dashboard. Set LLAMAFORGE_LLAMA_SRC to supply the checkout non-interactively. Afterwards, start it with LlamaForge.vbs (Windows) or ./run.sh, and stop it with stop.ps1 / ./stop.sh.

Building llama.cpp from source needs more tools, all detected on the Setup tab:

• Git, CMake, Ninja;
• a C++ compiler: MSVC on Windows, clang++/g++ on Linux/macOS;
• CUDA, only for NVIDIA builds (macOS uses Metal).

On Windows and macOS the Setup tab can install missing tools after you confirm (winget/choco, Homebrew). On Linux, LlamaForge never runs sudo: it shows the exact apt/dnf/pacman command for you to run.

What the launcher does

The launcher reads config.json and starts the llama.cpp router (llama-server --models-preset <models.ini> --models-max <n> --offline --host <router_host> --port <router_port> --metrics --api-key <key>) and the dashboard backend (<n> is 1, or the slot_cap pool size when multi-model is on), each only if its port isn't already in use, then opens the dashboard. Running it twice is safe. Launch output goes to logs/ in the install directory, so a router that fails to start leaves a log you can read.