Moonglow connects local Jupyter notebooks to remote GPUs, letting ML researchers scale up experiments instantly. No more converting notebooks to scripts or dealing with SSH: with Moonglow, you can move your notebook from a CPU to H100 in under a minute.
Hey everyone, I'm Trevor, and together with @leilaclark, we are the founders of Moonglow!
Moonglow makes it easy to move your machine learning experiments from local Jupyter notebooks to cloud GPUs.
I used to do ML research at Stanford, while Leila was a software engineer at Jane Street. A problem we've both experienced is how annoying it is to transition from trying things in a Jupyter notebook on your computer to running large scale experiments in a remote GPU.
Right now, the workflow involves picking the right configs on your cloud provider, spinning up a remote GPU, ssh-hing into the machine, setting up the dev environment and pulling your code from Github. All of this is before you've run a single cell in your notebook. If you want to share your work or come back to it later, either you need to keep your GPU running (expensive) or go through this entire process again (slow).
We're building Moonglow for ML researchers and engineers, so that this entire workflow takes one click. Just change switch runtimes and we'll handle all of the DevOps under the hood.
We currently support connecting VSCode / Cursor to Runpod GPUs, with more platforms coming soon. We'd love for you to try it out at moonglow.ai, and we can't wait to hear what you think!
Hey Trevor,
Does it automatically sync local environments with the cloud GPU? Have you considered adding features for collaborative work, like shared notebooks or version control integration?
Congrats on the launch!
@kyrylosilin Hey Kyrylo, we currently provide the same environment (designed for ML, with all the standard packages e.g. numpy, pandas, pytorch etc.) every time, but we're planning to add custom environments that are based on your local environment.
Shared notebooks are definitely in the roadmap too! As for version control, since your file is local, we are not currently planning on adding this, and instead let you use the version control you'd be using already for your local files.
Hey Trevor! This sounds like an incredible solution for the ML community. The pain point of transitioning from local to cloud has definitely been a hurdle for many researchers. It's amazing to see how Moonglow streamlines this entire process, especially the one-click runtime switch. This could save users so much time and frustration.
I love the background you and @leilaclark bring to the table, with your experiences at Stanford and Jane Street, which really makes Moonglow resonate with its target audience. Excited to see you support more platforms in the future!
Can't wait to test it out and provide some feedback. Wishing you both the best of luck with the launch!
Moonglow
Telebugs
Moonglow
Moonglow