Changelog

Each entry summarizes the changes since the previous minor release.

0.6.0 (2026-10-15)

Support offline installation on AMD GPU and GPU-less hosts; NVIDIA hosts use online installation.
Add switch congestion counters, affected RNICs, device rankings and recorded congestion events with dashboard alerts.
Provide RHEL-family RPMs through the internal repository.
Add Fabric pages for switches, links and connected devices.
Extend monitoring to Kubernetes inference services, RNICs, AMD GPUs, LLMBoost and MASS.
Update DPU firmware on multiple nodes at once and post dashboard announcements.
Strengthen sign-in, access controls, audit logging and password recovery, and limit each BMC remote console to the user who opened it.
Improve readability in light and dark themes, and the reliability of node registration, updates and hardware readings.
Refuse accidental node downgrades and connections to switches whose SSH host key changed.
Breaking: API routes are renamed and responses simplified. mac start and mac restart wait for the services by default (--no-wait returns at once), and mac images --prune is now mac images prune. Update scripts that use them.
Users must sign in again once after upgrading.

Add a read-only LLMBoost dashboard and MASS storage management.
Add cluster-wide health checks with mac test and streaming SONiC switch telemetry.
Manage MAC versions across the cluster on Updates.
Add PCIe BIOS settings.