ggrun
Terminal launcher that plans and runs local GGUF models via llama.cpp backends.
Tested on Linux
Use it
Linux x86_64
You need first
- A GPU is optional
Download setup.sh for inspection before running
curl -fsSL -o /tmp/setup.sh https://raw.githubusercontent.com/raketenkater/ggrun/main/setup.sh && wc -c /tmp/setup.sh && head -n 100 /tmp/setup.shRun setup.sh installer (built ggrun + CPU backend)
bash /tmp/setup.sh
Check that it works
mkdir -p ~/.local/bin && ln -sfn ~/ggrun/ggrun ~/.local/bin/ggrun && export PATH="$HOME/.local/bin:$PATH" && ggrun versionThe output should contain “ggrun v3.2.13”.
First thing to try
mkdir -p ~/.local/bin && ln -sfn ~/ggrun/ggrun ~/.local/bin/ggrun && export PATH="$HOME/.local/bin:$PATH" && ggrun versionRun ggrun version to confirm the tool is installed and on your PATH.
What it does
- Get model suggestions that fit your hardwareNot shown working
- Run local models through a compatible backendNot tested
- Plan how large MoE models split across GPU and RAMNot tested
- Start a Claude Code workflow with a local modelNot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 0 of 4 features work
- Works offline
macOS and Windows: the project says it runs there; pickfit hasn’t tried it.
Good for
- find a capable model that fits hardware and context requirements
- run a model through a compatible backend with automatic memory placement
- plan large-MoE placement across GPU memory and system RAM
- expose OpenAI-compatible API for chat completions
- launch a local Claude Code workflow with local model endpoint
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
Its main function works
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version v3.2.13
Differences the test machine hit
- README: Open a new terminal after adding ggrun to PATH.
In the test: Symlink ~/ggrun/ggrun to ~/.local/bin/ggrun (or source ~/ggrun/.env.sh); noninteractive setup did not add PATH entry. - README: curl setup.sh | bash one-command install
In the test: Installer auto-downloaded a Go toolchain and built ggrun from source plus a bundled CPU llama-server; fetched binaries cannot be pre-read.