Welcome to Cloudy's first ever changelog! Starting today, you can expect weekly updates, every Friday.
Cloudy's features
Since this is our first changelog, we decided to summarize everything we shipped in the past few weeks.
Storage Volumes
You can now create persistent storage volumes that live across sessions.
- Create and delete volumes
- Fork any volume instantly (makes a copy you can experiment with)
- Upload and download files through the browser
- Preview markdown, code, and text files directly
- Make volumes public to share with anyone, or keep them private to your team
GPU Pods
Rent GPU instances on demand. We have H100 (80GB) and H200 (141GB) available.
- Select GPU type, count, and region
- Attach a storage volume to your pod
- SSH into your pod with your own keys
- See live uptime and cost per hour
- Stop pods when you're done
Teams
Work together with your teammates.
- Invite people by email
- Shared volumes and pods across the team
- Manage who's on your team from settings
What's Next
Here's what we're working on:
- Multi-node clusters with InfiniBand for distributed training at 3.2TB/s
- cloudy-cli so you can do everything from the terminal (works with Claude Code too)
- Open source templates for RL, LoRA, inference, and other popular workflows
- One-click benchmarks to test your models on standard benchmarks right after training
- Better pod controls with custom templates, environments, and port configs
- Monitoring dashboard to see if your jobs are running and using the hardware well
- Cluster testing service where we help companies verify their compute matches specs with NCCL and network tests
Preview of multi-node H100 rentals with Infiniband and managed Kubernetes (coming soon):
See you next Friday!