We're at KubeCon in Salt Lake City this week, showing up across yesterday’s colocated events and at the main show from today at stand 125. If you haven't caught up with PaletteAI lately, this is a good time to stop by.
As well as a host of new features, we’re sharing innovations with our partners, from AI infrastructure with NVIDIA to model and tool access with Traefik. Keep reading for just a few highlights of what we have in store.
Cluster templates for the fleet you already run #
From the start we built our platform around declarative management: you define the software and configuration you want, and use that definition throughout the lifecycle when you’re deploying and maintaining clusters across cloud, data center and edge environments. That also includes GPU infrastructure and virtual machines through our Virtual Machine Orchestrator (VMO). Cluster profiles are the magic ingredient. They’re how you define the ‘desired state’ of your stack, from the operating system and Kubernetes distribution through to networking, storage and applications. Earlier this year, we introduced cluster templates: a bold extension of the profile concept, bringing in policies for managing a fleet. You can set maintenance windows, for example, so upgrades run at an appropriate time for each cluster's location. Our introduction to cluster templates explains how the pieces fit together.
Initially, templates were only available for new clusters. Now you can attach an existing, running cluster provisioned by Palette to a template and bring it under the same configuration and maintenance policies. You review the changes before applying them; configuration differences may still require node replacement during a maintenance window.
Hadron joins our list of supported OS options #
We've always believed you should be able to choose each layer of your stack from the open source projects that suit your needs, and that includes the foundational operating system.
Earlier this year the Kairos maintainers, supported by Spectro Cloud, announced Hadron: a lightweight Linux distribution with base images under 100 MB. A smaller OS means less software to distribute and maintain, and a smaller attack surface. It's useful at the edge, especially over slow links, but those benefits apply in the data center too.
You can now try Hadron with Palette through our tech preview for connected clusters. The Kairos team will be at the show if you'd like to talk through a deployment.
AI virtual machines in PaletteAI #
PaletteAI gives your developers access to AI models, applications and compute resources on their organization's infrastructure. As of version 1.4, they can also create GPU-enabled virtual machines from the same console, using VMO underneath.
An AI VM gives you a full operating system and persistent storage for work such as running a notebook environment or using software that expects a VM. Administrators can set resource limits and control who can access or share a VM, with usage attributed to the team, project or organization that owns it. The public PaletteAI API reference is available for teams automating their workflows.
More storage choice for VM Launchpad #
For teams starting with VM modernization, our PaletteAI VM Launchpad software appliance now offers Portworx and Piraeus storage options. Piraeus is available in FIPS and non-FIPS builds; Portworx is available in the non-FIPS build.
The power of partnerships #
KubeCon is all about the ecosystem working together... so what better opportunity to make some announcements with our partners?
NVIDIA is introducing NVIDIA Kubernetes Engine (NVKE), a Kubernetes management layer in DSX OS, currently in preview. It automates provisioning and lifecycle management for clusters on NVIDIA GPU infrastructure. We're developing a phased NVKE integration with PaletteAI as part of our work in the NVIDIA DSX ecosystem. The NVIDIA Run:ai team will be in our booth on Wednesday.
Traefik is the default gateway for PaletteAI Model as a Service. With Traefik Hub, you can apply access policies and usage limits to model and MCP tool calls, including in self-hosted and air-gapped environments. Its audit records let you investigate which calls were allowed or refused. Our joint walkthrough shows the integration in operation; you can see it demonstrated at our booth on Tuesday at 4pm and Wednesday at 2pm.
And in case you missed it, today we're introducing AMD Instinct Coder 355, powered by Dell PowerEdge XE9785 servers, AMD Instinct MI355X GPUs and PaletteAI Inference Launchpad. It expands our local AI coding offering for larger development teams and models that need more GPU memory. Learn more here.
See the products, meet the people #
So those are the headlines. At stand 125, we'll do our best to make them feel real. We’ll have demos covering cluster fleet management, local AI inference, VM modernization with KubeVirt and the AI factory stack. Our booth theater will run short sessions with our engineers and partners throughout the show, so come see us, ask questions about what's new, and meet the partners building all this with us.
One last thing: tonight, you’re invited to join us, AMD and AWS in the speakeasy room at Edison House. Book a meeting or register your interest in the afterparty on our event page, and follow us on LinkedIn for the daily booth schedule. We'd love to hear what you're working on.