DKubeX Documentation#

Version v0.2.2

DKubeX is a Kubernetes-native platform for deploying, managing, and operating AI and ML applications from a single control plane. It layers a self-service application catalog, a governed model plane, and integrated developer tooling over a standard Kubernetes cluster — so teams can run the full model lifecycle, from training to governed inference, on their own infrastructure.

At a glance#

Category

Self-hosted MLOps / LLMOps platform

Foundation

Kubernetes (k3s-compatible), single-cluster

Deployment

On-premises or cloud; firewall-aware, private-registry-based

Control plane

FastAPI backend · React 19 UI · Kopf operator

App model

Custom Application CRD + one-click catalog (Helm under the hood)

Ingress & auth

Traefik + ForwardAuth SSO, per-app RBAC, seat-based licensing

Data layer

PostgreSQL 16 · MinIO (S3-compatible) · Redis · NFS shared storage

Model serving

KubeAI (LLMs/embeddings) · KServe (classic ML) · MLflow registry

AI gateway

SecureLLM — governed, audited, OpenAI-compatible model access

GPU

NVIDIA GPU Operator, per-model resource profiles, scale-to-zero

Observability

ClickStack + OpenTelemetry (GPU / vLLM dashboards)

Interfaces

React web UI · REST API (/api/v1, Swagger/ReDoc) · Application CRD (kubectl)

Applications

Workspace · ModelStudio · MLflow · SecureLLM · RAGFlow · Langflow

Core license

MIT (platform core)

What DKubeX is#

Rather than stitching together separate tools for training, serving, RAG, and developer environments, DKubeX provides a single control plane that:

  • Runs the full model lifecycle — train in a workspace, register versions in MLflow, deploy classic ML through KServe and LLMs/embeddings through KubeAI, then govern every model through SecureLLM.

  • Treats applications as declarative Kubernetes resources — each install is a custom Application resource that an operator reconciles to a running Helm release, with a live status phase and access URL.

  • Enforces one identity and governance layer — a single authenticated control plane provides SSO across every app, per-application RBAC, seat-based licensing, and per-user namespace isolation.

  • Keeps models and data on your infrastructure — models are served cluster-local and fronted by an auditable gateway; the platform is firewall-aware and pulls images from a configurable private registry.

Note: Apps install into a dedicated dkubex-apps namespace; each user also receives an isolated namespace and a per-user persistent volume (user-storage PVC).

Platform architecture#

The user interface is routed through Traefik and authentication into the Kubernetes cluster, where the core control plane, application store, data and storage, observability, and admin tooling run as coordinated subsystems.

DKubeX platform architecture — user interface routed through Traefik and auth into the Kubernetes cluster (core control plane, app store, data and storage, observability, admin tools, and infrastructure).

DKubeX platform architecture at a glance.#

Explore the documentation#

  • Installation — install DKubeX on your Kubernetes cluster with Helm.

  • Applications — user guides for Workspace, ModelStudio, RAGFlow, SecureLLM, and Langflow.

  • Tutorials — end-to-end guides across governing LLM access, LLMs and RAG, MLOps, and coding agents.

Contents