wy51ai/easycolab

What it does

EasyColab is a macOS and Windows desktop client for driving a Google Colab GPU runtime without keeping a browser notebook open. The macOS app is built with SwiftUI/AppKit; the Windows app uses PySide6/Qt6 and bundles Python, FFmpeg and the Colab bridge. The client wakes a Colab instance, then exposes model and workflow services inside that instance.

The product targets Colab G4 (RTX PRO 6000 Blackwell, 96 GB VRAM), H100, A100 and L4 runtimes. It includes four model-oriented work streams: MiniMax-H3 video generation, a ComfyUI node workspace with persisted custom nodes and models on Google Drive, a Qwen vision-language gateway exposing OpenAI- and Anthropic-compatible endpoints, and Qwen-Image/Z-Image image generation and editing. It also graphs GPU memory, system memory and disk use, shows remaining compute units, and can automatically release an idle instance.

Getting started

The simplest supported path depends on the platform. On macOS, clone the repository, then compile the app locally with the provided script; this avoids a Gatekeeper prompt on an unnotarized bundle:

git clone https://github.com/<your-username>/EasyColab.git
cd EasyColab
bash macos_app/build_macos_app.sh
open EasyColab.app

On Windows, either download the portable EasyColab-Windows-x64.zip from Releases or the Actions build and run EasyColab.exe, or build it from a Python 3.10–3.13 install with windows_app\build_windows_app.bat. Source development uses a virtual environment and python windows_app/easycolab_win.py.

If you only have the free Colab tier or prefer a browser workflow, open Google Colab, upload EasyColab_Server.ipynb, select a G4/A100/L4 runtime and run the cells. The notebook prints a https://*.gradio.live address that you paste into the client's service-address field and connect.

The full Windows and macOS instructions are in the README; a separate English README is also provided.

Requirements and constraints

A paid Google account with Colab Pro, Pro+, or Google AI Ultra and available compute units is needed for the no-browser wake-up path. The macOS build requires Xcode Command Line Tools; the Windows portable package is meant to avoid a local Python install. The source-build path needs Python 3.10–3.13. Model weights are not shipped with the repository: each model is downloaded at runtime from Hugging Face under its own license, and ComfyUI is pulled separately under GPL-3.0. The project is Apache-2.0 licensed.

The README explicitly describes the wake-up functionality as built on google-colab-cli and warns that it should be used on your own paid account, not on free tiers for mining or resale. It also notes this is an unofficial project, not affiliated with Google, MiniMax or Qwen. The macOS release path accepts an unnotarized .app after running xattr -cr; the Windows path advises bypassing SmartScreen via "More info → Run anyway." Those steps reduce OS protection, so verify the release source and the build artifact you obtain before running either binary. OAuth tokens are stored locally; API-key authentication is recommended when the Qwen gateway is reachable externally.

HotRepo's take

EasyColab entered HotRepo's new-repository list for Python on 2026-11-08 with 101 stars and 20 forks, all gained within its first recorded day. The repository has no recorded all-language daily rank and only a single star-curve sample, so there is not enough history to call a trend. What stands out technically is the scope of a brand-new client: separate native UI stacks, four large model workflows, and a Colab wake-up path that depends on an external CLI and paid compute units. That breadth is also the main risk to watch — the prebuilt Windows .exe and unnotarized macOS .app are central to the documented setup, while the repository materials do not establish how those artifacts map to the source or who publishes them.