Model Launcher 0.3.0

New

• Benchmark page. Test whether a model can use a computer, with Computer Use. Run the built-in tweet test or write your own benchmarks (+ New benchmark), then check the result and share the report on X or as an issue. Reports are saved in ~/Documents/Computer Use Benchmarks.
• Run anyway. Load a model bigger than your RAM budget or free memory. Other chat models unload, and what doesn't fit on the GPU goes on the CPU.
• External SSD support. Move all your models to a drive in Settings › Storage (~/Models links to it), and turn on "Use disk as extra memory" so big models read their weights from the drive.
• Internet access. Chat models can search the web and read pages. Turn it off in Settings › Internet access.
• Updates itself. Checks GitHub at launch and every 6 hours, and installs new versions once nothing is running (Settings › Updates).
• Linux (preview). A .deb and an AppImage, using the official llama.cpp, whisper.cpp and stable-diffusion.cpp builds (Vulkan or CPU). They're attached to this release a few minutes after it's published.
• Support page, and active downloads now show at the top of Library.

Fixed

• macOS no longer asks for file access again after every update: the app is now signed with a stable identity. It asks one last time after this update.
• Pages stay centered at any window size (the Connect page drifted right).
• The Add a model window kept the wrong styling after the "Add your own model" box was added.

Updating from 0.2.0: 0.2.0 can't update itself, so download the DMG below once, then drag Model Launcher to Applications. From 0.3.0 on, updates install themselves.

Model Launcher 0.2.0

• Several chat models loaded at once on Macs with room (Settings › Chat models at once), unloading the least recently used to stay in your RAM budget.
• Video: Wan 2.1 1.3B, in Studio, the API (/v1/videos/generations) and MCP (generate_video).
• Add a model from anywhere: paste a Hugging Face link, or move in files already on your Mac, with walkthroughs.
• Every card shows its modality, the maker's logo, its source and its license.
• The window can be dragged; calmer, tighter design; RAM budget is labelled as RAM.

Model Launcher 0.1.0

First release. Launch open-source chat, voice and image models on your Mac with memory limits you control, an OpenAI-compatible API and an MCP server.

Open the DMG and drag Model Launcher to Applications. Apple Silicon, macOS 13 or later. The app is ad-hoc signed, so the first time, right-click it and choose Open.