Skip to main content
Version: v1.2.x

What is PaletteAI?

Deploying AI/ML applications on Kubernetes is complex. Data science teams need GPU-accelerated clusters, specialized storage, and properly configured networking; however, they would rather focus on experiments and models than infrastructure. Meanwhile, platform teams want to provide self-service access while maintaining control over resources, costs, and security.

PaletteAI solves this by providing a Kubernetes-native platform where platform teams manage infrastructure and data science teams deploy applications through a self-service interface - no Kubernetes expertise required.

Who Benefits?​

Data Science Teams​

With PaletteAI, data scientists can deploy AI/ML environments in minutes and get to work without worrying about Kubernetes. Bringing repeatability and consistency to application and model deployments, data scientists can focus on what matters to them instead of being bogged down with complex infrastructure configuration and deployment complexities. Whether initial deployment or ongoing lifecycle management, PaletteAI makes sure applications are kept up to date.

  • No Infrastructure Expertise Required - Click to deploy ClearML, Run:ai, or model servers
  • Consistent Environments - Same configuration across development, staging, and production
  • Self-Service Access - Deploy when needed without waiting for platform teams
  • GPU Management - Automatic allocation and quota enforcement

Platform Engineering Teams​

Platform engineers configure the compute resources and Profile Bundles that power self-service, pairing the right hardware (GPUs, storage, networking) with the security and governance controls their organization requires. This declarative approach makes life easier for platform engineering teams with consistency, repeatability, and all the enterprise-grade controls and governance they need for both training and inference clusters.

  • Reusable Templates - Define infrastructure once in Profile Bundles and deploy everywhere
  • GPU Quotas - Enforce limits at Tenant and Project levels to control costs
  • Multi-Tenancy - Isolated workspaces with RBAC for different teams
  • Policy Enforcement - Ensure clusters meet security and compliance requirements
  • GitOps Compatible - All configurations stored as Kubernetes custom resources

Integration with Palette​

PaletteAI integrates with Palette to automate Kubernetes cluster provisioning with:

  • Infrastructure templates (Cluster Profiles) for OS, Kubernetes, networking, and storage
  • GPU hardware discovery and management
  • Edge host and bare metal orchestration

Ready-to-Use Application Catalog​

The PaletteAI Studio provides a catalog of pre-built Profile Bundles for common use cases, including AI/ML applications (such as Run:ai and ClearML), models, and infrastructure stacks. Each bundle packages the dependencies and configurations needed to deploy an application or model. You can customize an existing bundle or build your own to standardize custom, in-house stacks across teams.

Key Concepts​

Whether you are a platform engineer setting up infrastructure or a data scientist deploying workloads, the following concepts explain how PaletteAI organizes resources and manages AI/ML deployments.

Multi-Tenancy​

PaletteAI uses a hierarchy of scopes to organize teams and control access to resources.

  • Multi-Tenancy - The scope model (Project, Tenant, and system) for isolation, GPU quotas, and resource sharing. Tenants represent organizations or teams; Projects are workspaces within Tenants.
  • Settings - External integrations (Palette API credentials) and namespace-scoped configuration.
  • Roles and Permissions - RBAC roles automatically created for Tenants and Projects.

Compute Resources​

These resources define where your AI/ML applications run. Clusters are provisioned using Palette as the infrastructure platform.

  • Compute - Discovers available machines for cluster provisioning.
  • Compute Config - Default settings for cluster deployment (networking, SSH, node configurations).
  • Compute Pools - Kubernetes clusters where applications run (dedicated or shared modes).

Workload Resources​

These resources define what gets deployed and how workloads are configured.

  • Workloads - Workloads, WorkloadProfiles, and WorkloadDeployments.
  • Definitions - Components, Traits, and Policies that compose workloads.
  • Variables - User-defined configuration values for templating.
  • Environments - Placement policies that control which clusters receive workloads.

Deployments​

These resources package and deploy applications to your infrastructure.

  • Profile Bundles - Reusable packages combining Palette Cluster Profiles and PaletteAI Workload Profiles for consistent, repeatable deployments.
  • App Deployments - Deploy AI/ML applications, such as Run:ai and ClearML, to Compute Pools using Profile Bundles.
  • Model Deployments - Deploy AI/ML models for inference using Model as a Service or custom configurations.

Example Workflow​

A typical deployment involves these concepts working together:

  1. Platform setup - A platform engineer creates a Tenant with Settings for Palette integration.
  2. Project creation - Teams get Projects with GPU quotas and role-based access. Refer to Create and Manage Projects for how to create Projects.
  3. Infrastructure provisioning - Compute discovers available machines; Compute Pools provision Kubernetes clusters.
  4. Application packaging - Profile Bundles package infrastructure and application configurations.
  5. Deployment - Data scientists create App Deployments or Model Deployments using Profile Bundles on Compute Pools.

Next Steps​

Choose your path to get started:

  • Install PaletteAI - Follow the Appliance Installation Guide to create a cluster using the PaletteAI ISO and deploy self-hosted Palette and PaletteAI.
  • Look Under the Hood - Review the Architecture guide to understand how PaletteAI delivers a seamless AI/ML deployment experience.
  • Discover Resources - Browse the Resources for custom resource definitions and more.