Airunner — local model runner manager

Manage and launch the GGUF models on this machine. Supports both runners:

• llama.cpp (llama-server) — /home/fred/ai/llama.cpp/build/bin/llama-server
• DwarfStar (ds4-server) — /home/fred/dwarfstar/ds4-server

Zero dependencies: Python 3 stdlib backend + a single-page vanilla JS frontend.

Run

cd /home/fred/ai/airunner
python3 airunner.py
# web UI: http://127.0.0.1:8090

Optional CLI args: --host, --port, --llamacpp-bin, --dwarfstar-bin.

What it does

• Launch models from either runner, with any of the runner's launch options (context, threads, GPU layers, backend, KV cache, SSD streaming, DSpark, …).
• Launch-option table per runner: shows every option with its default, a suggested value, and lets you edit before launching. For both llama-server and ds4-server.
• Parallel models: run multiple models at once — each gets its own process, console, and Stop button.
• Output console: a live textbox per running model showing the runner's stdout/stderr.
• Remembers past setups: save any configuration as a "setup"; apply or delete them later. Setups are persisted in ~/.local/share/airunner/state.json.
• Autostart at machine boot: mark a setup "autostart". Airunner installs a systemd user service (via the UI button) so it comes back on reboot, then auto-launches every autostart setup.
• Restart if it crashes: mark a setup "restart on crash" and a watchdog (default every 5 s) relaunches the model whenever its process dies. Restart count is shown per model.
• Point at binaries & models: the Config panel sets the llama.cpp / DwarfStar binary paths and the model directories scanned for .gguf files.

Autostart at boot

In the UI: open Config → Install autostart service. This writes ~/.config/systemd/user/airunner.service and enables it, so after a reboot the service starts Airunner, which then launches every setup marked Autostart. Use Remove service to disable.

Note: systemd user services start after your desktop login. If you want the models up before login, you'd need a system-level unit (requires root) — not installed here by default.

REST API (same as the UI)

Method	Endpoint	Purpose
GET	/api/state	full state (config, setups, procs, models, options)
GET	/api/models	discovered .gguf models
GET/PUT	/api/config	runner binaries, model dirs, host/port, watchdog interval
GET	/api/launch-options?runner=	option table for llamacpp / dwarfstar
POST	/api/run	start a model ({setup})
POST	/api/stop	stop a model ({id})
POST	/api/save-setup	remember a setup
POST	/api/apply-setup	launch a saved setup ({id})
POST	/api/delete-setup	forget a setup
POST	/api/systemd/install /api/systemd/remove	autostart service
POST	/api/autostart/start-all	launch all autostart setups now

Files

• airunner.py — backend: process manager, watchdog, autostart, REST API
• web/index.html — the single-page UI