Requirements & supported versions

AlchemiStudio Platform is validated on the following SUSE AI environment. These are the versions we test against and support; newer patch releases within the same minor are expected to work and are supported on a best-effort basis until formally re-validated.

Supported version matrix

Cluster prerequisites

• Rancher Prime managing an RKE2 cluster (the SUSE AI Factory baseline).
• Longhorn storage class available (the RKE2 default). AlchemiStudio uses both RWO (databases, caches) and RWX (shared file workspaces via Longhorn share-manager).
• Ingress / load balancer reachable from clients. In cloud IaaS, provision a real LoadBalancer for APISIX; on bare metal use your existing L4/L7 front end.
• Egress (or a mirrored private registry) to pull the AlchemiStudio images and, for model serving, the SUSE Application Collection / NVIDIA (dp.apps.rancher.io) and open-model images.
• A DNS name + TLS for the platform. AlchemiStudio supports bring-your-own certificate, ACME (Let’s Encrypt HTTP-01), or self-signed for lab use.

Optional / model serving

• NVIDIA GPU node + the SUSE NVIDIA integration for NIM serving (validated on a T4).
• vLLM for open-model serving with no external model registry dependency (validated with Qwen2.5).

Resource guidance (starting point)

The full trial stack (app tier + bundled data tier) runs comfortably on a small multi-node RKE2 cluster. Size GPU nodes to the model you intend to serve. Tune per-service replicas and limits for production load.

See Deploy on Rancher / RKE2 to install, or Support & lifecycle for the version-support policy.