March 15, 2026
A new Storage Node host type, clearer deployment failure reasons, and several stability fixes.
Major features
Section titled “Major features”Storage Node — a new host type
Section titled “Storage Node — a new host type”Providers can now onboard dedicated storage machines through a new setup flow that includes RAID configuration, alongside the existing GPU compute host type.
Improvements
Section titled “Improvements”- Unsupported GPU models are now automatically unlisted, with guidance to contact support if you want a new model supported.
- Pod deployment failures now show the actual failure reason instead of a generic error.
- Connect-type pods (like ComfyUI) now wait until the internal app is actually ready before allowing connection.
- Pods stuck during deletion are now force-removed after a timeout instead of lingering indefinitely.
- Improved multi-GPU pod scheduling with hardware-aware resource placement, plus better dark-mode chart readability and more consistent pod/storage template styling.
- Fixed a bug where a failed backend request could make parts of the console disappear entirely — errors are now handled gracefully everywhere.
- Fixed a billing/transaction rollback bug that could leave state inconsistent after a failed operation.