Dakodeon lives in your menu bar. Pick a curated model, start the local server, and point any agent at an OpenAI-compatible API — no runtime to bundle, no config to babysit.
A built-in Settings window tracks every curated model. Watch downloads in real time, free up disk in a click, and switch models without touching a terminal.
Every model shows progress, size, and state — pulled straight from the shared Hugging Face cache.
Choose a model and the server stops and relaunches with its parameters — drafts, MTP, and all.
Delete weights you're done with. One confirm and the backing files leave the cache.
Serves llama-server at 127.0.0.1:8080/v1 with a stable local alias for agents like OpenCode.
Uses the system llama.cpp and hf you already have. The app bundle stays tiny.
Each profile ships ready to run — weights, draft model, and tuned flags. No knobs to get wrong.
Requires llama.cpp and hf on your PATH · macOS 14+