Hi Product Hunt! 👋 I'm Erik. I run local models with Ollama on my Mac, and a loaded model can hold 15–20 GB of memory long after I've stopped using it. Killing the process didn't always work either: with brew services, a LaunchAgent brings it right back. So I built ModelNap, a tiny native menu bar app: • One click (or ⌥⌘O) turns Ollama on or off, using the same mechanism it started with • Idle models are unloaded after 5–60 minutes; Ollama stays on and reloads them on the next message • New in 1.3: it also controls Splash (Inco AI), shows its real memory, and makes room by unloading Ollama first • No accounts, no telemetry, nothing sent to the internet. MIT licensed. It's free and open source. llama.cpp support is next on the list, so tell me how you run your local models!

ModelNap

Your local AI, on only when you use it

Your local AI, on only when you use it