collabosm
A local app that rents a GPU in your own Google Colab account and serves a large Qwen3.8 model on it, showing the cost before it starts and exposing the model to Codex or any OpenAI-compatible client.
collabosm runs Qwen3.8-27B on a rented A100-40G, or the larger Qwen3.8-Flash-Next mixture-of-experts model on an A100-80G, using compute units from the user's own Google Colab plan rather than a service collabosm itself operates; the user pays Google only while the GPU is running, and collabosm is free and open source.
A local desktop app starts and stops the rented GPU with a button, shows the price before billing begins, and lets a user chat with the model directly, including with images, or point another program such as Codex, ModelDock or Open WebUI at its fixed local OpenAI-compatible endpoint. It requires a Google account with Colab compute units and Python 3.12 or newer, and runs on Windows, macOS and Linux.