Contributing
GoRunRun Local AI is open source under the Apache License 2.0, and help is welcome, whether that’s an afternoon of testing or a new platform. Read CONTRIBUTING.md and the Code of Conduct first. For anything big, open an issue before you start, so we can agree on the approach.
- Test it on your Mac
- Bring it to Windows and Linux
- Add features
- Docs and bug reports
- Before you open a pull request
Test it on your Mac
It’s built and tested on one Mac: an M5 Max with 64 GB. Every other chip and memory size is untested. Which Mac? lists them.
- Preview what the installer would do on your Mac, or any other, without installing anything:
install.sh --dry-run, or add--machine=M4Max-48GBto simulate a different one. - Install it, use it for a day, and file a test report. Reports that everything works are just as useful as problems.
- Timings help most: how quickly replies start, how long a video takes, and whether voice mode keeps up.
Bring it to Windows and Linux
This is the biggest open project. Today these parts are tied to Apple silicon and macOS:
| Part | On the Mac today | What another platform needs |
|---|---|---|
| Chat models | MLX (mlx-vlm) behind llama-swap |
A backend for NVIDIA, AMD or CPU, such as llama.cpp, with the same model list in config/models.yaml |
| Speech and voice | MLX builds of the speech-to-text and text-to-speech models | Versions of the same models that run on CUDA, ROCm or the CPU |
| Video creation | ltx-2-mlx and an MLX build of Wan 2.2 |
The original PyTorch versions on a GPU with enough video memory |
| Installer | install.sh (Homebrew, macOS checks) |
A Linux script and a Windows installer, with memory tiers based on GPU video memory as well as system memory |
| Background service | A launchd agent |
systemd on Linux, a service or scheduled task on Windows |
| Desktop app | A Swift window around the web app | A cross-platform window, or just the browser |
The web app, the orchestrator, the tools and the tests are ordinary Python and TypeScript, so most of the code carries over. A good first step is a Linux build that runs the chat model through llama.cpp on an NVIDIA GPU, with an entry in the memory table for it.
Add features
Some ideas, if you’re looking for one:
- Editing photos with a prompt, alongside animating them.
- More voices and languages for voice mode.
- New built-in tools, or ready-made MCP server setups (see Extending it).
- Translating the app’s interface.
Check the open issues and feature requests first.
Docs and bug reports
Clearer instructions, a missing FAQ answer, or a bug report with the last lines of the log (Help → Show Logs) all help. Questions: amit@gorunrun.ai.
Before you open a pull request
- Follow Build from source to set up, then run
make test(andmake test-e2efor interface changes). - Keep each pull request focused, and add or update tests for what you change.
- By contributing, you agree that your contribution is licensed under the Apache License 2.0.