I built a lightweight tool that automatically frees up VRAM for gaming and restores your local LLMs afterward

I run local LLMs pretty much around the clock, but I also enjoy gaming on the same PC. Having models sitting in VRAM while gaming isn't exactly ideal, and constantly unloading and reloading them manually got annoying. So I built GamePause . It's a lightweight, open-source Windows tray app that: Automatically detects when you launch a game (Steam, Epic, GOG, etc.) Unloads your LM Studio/Ollama models to free up VRAM (other providers tentatively supported, not tested) Restores your models automatically when you're done gaming Supports per-game rules, manual controls, and crash recovery It works at the model server level, so it doesn't matter whether you're using a local chat interface, coding assistant, or an autonomous agent like Hermes or OpenClaw. It's designed to be a install-and-forget experience. Everything runs locally, with no telemetry or accounts. Written in Rust and MIT licensed. GitHub: github.com/Vkuparin/gamepause-for-local-ai I feel like it's now tested and stable enough to share, but please do report on Github if you run into any issues.