Briefly
The unglamorous local-AI work: keeping the runtime recoverable, switchable, and understandable after something breaks.
PiLocalLLM Setup is my local runtime manager for the Pi/OpenWebUI stack. It tracks the scripts and app shell needed to run the stack, switch models, install LaunchAgents, check service health, and rebuild the machine from a clean state.
The repo deliberately excludes the parts that should not be public: generated certificates, private keys, secrets, logs, backups, Python caches, and model weights.
The public view is deliberately cleaned up. I want to show the shape of the system, not local IPs, paths, certificates, or service output.
What I built
- Native macOS manager app calls the same scripts as the command-line workflow.
- LaunchAgent templates cover OpenWebUI, bridge, embeddings, HTTPS proxy, and optional local backends.
- Model-runtime scripts and the Pi /model hook make model switching an operational behavior, not just a saved preference.
- Restore docs and doctor checks make the setup portable across clean-machine rebuilds.
A few notes
- The repo contains bootstrap, doctor, model-download, LaunchAgent install, and Pi model-hook scripts.
- The README documents the exclusion policy for certs, keys, secrets, logs, backups, virtualenvs, and model weights.
- The native manager app lives under apps/PiLocalLLMManager and is backed by the repo scripts.