CLI enabling Free Cloud GPU access in your terminal for learning CUDA
# Install cgpu npm i -g cgpu # First run will launch an interactive setup wizard # Connect to a cloud GPU instance quickly without setup any time after that cgpu connect # Run a command on a cloud GPU instance without a persistent terminal (but mantaining file system state) cgpu run nvidia-smi
Serve Gemini for Free as OpenAI-compatible API
You can start a local server that proxies requests to Google Gemini using the cgpu serve command. This allows you to use Gemini with tools that expect an OpenAI-compatible API.
# Start the server on port 8080 cgpu serve # Specify port and model cgpu serve --port 3000 --default-model gemini-2.0-flash
For an example of using this with the OpenAI client, check out python_example. This requires you have the gemini cli installed.
Vision
The primary goal of this project to facilitate a high quality developer experience for those without GPUs who would like to learn CUDA C++ This means 3 main things:
- Free: Avoid having to pay while learning.
- Highly Available: Run quickly instead of having to wait in a queue so that users can compile quickly and learn faster.
- In User Terminal: Allows developers to use their own devtools/IDEs (Neovim, Cursor, etc) so they can be most productive.
Next Steps
I will continue to add to the CLI as I find more free compute sources and developer experience improvements. To see what I am currently planning to add check out the Issues tab on Github. Feel free to create new Issues for suggestions/problems you run into while learning!