For developers
GoRunRun Local AI is open source under the Apache License 2.0. The code lives at github.com/gorunrunai/local-ai.
It’s a small, readable codebase:
- Python (FastAPI) for the backend, orchestration and media handling
- MLX for running models on Apple silicon
- React for the web app
- Swift for the Mac app
Most extensions need only a configuration change or a single file.
- Build from source: clone, set up, run and test.
- Architecture: how a message travels from the box to the answer.
- Extending it: add models, tools, MCP servers, prompts and styles.
- Contributing: test on more Macs, bring it to Windows and Linux, add features.
Principles
- Local first. No telemetry and no hosted AI APIs. The network is used only to download models and by tools the user turns on.
- Loopback by default. Servers listen on
127.0.0.1. Remote access is opt-in and token-protected. - Untrusted data stays data. File contents, web pages and tool output are clearly marked as data, never followed as instructions. Risky tools ask first.
- Measured. Latency and memory claims come from benchmarks in the repo (
make bench,make bench-voice).
Contributions are welcome: see Contributing, and read CONTRIBUTING.md and the Code of Conduct, and report security issues privately as described in SECURITY.md.