CAIP Offboard v1.5: Managed Kubeflow
Managed Kubeflow is the Connected AI Platform's fully managed, Kubeflow-based machine-learning platform. It supports the full lifecycle of Classic ML use cases — from interactive development in notebooks, through pipelines and experiment tracking, to model serving.
Where this fits in the CAIP evolution
The Kubeflow product has stayed the same across the platform's evolution — what changed is the operating model:
- CAIP: Classic Kubeflow (CAIP1) — Distributed Kubeflow (bring-your-own-cluster, legacy): use cases prototyped on the shared AI-Research cluster, and for production requested their own Orbit (AWS EKS) cluster with CAIP provisioned into it per environment (test/int/e2e/prod). This caused heavy duplication and platform overhead.
- CAIP Offboard v1.5: Managed Kubeflow (this section): CAIP operates a managed cluster where each use case gets an isolated, namespace-based Kubeflow Workspace. No cluster to run or maintain — CAIP handles the full lifecycle and all infrastructure.
For new modular, API-driven services (LLM API, Agents, Workflows, Apps, …) see CAIP Offboard v2: All-New CAIP. To move off the legacy CAIP1 setup, see Platform Evolution & Migration.
Get started
- Onboarding — set up your Managed Kubeflow Workspace.
- Quick Start — Classic ML — scaffold your first project from the CD4ML template.
- Requests & Self-Service — request a workspace, manage users, set up CI Trust or a CDH connection; or use PR-based Self-Service for changes not (yet) covered by a Service Request.
What's inside
| Area | Description |
|---|---|
| Kubeflow | Notebooks, VS Code CDE, Pipelines, KServe, and Kubeflow flavours. |
| MLflow | Experiment tracking and model registry. |
| Classic ML | Data engineering, training & testing, and model serving. |
| End-to-End Deployment | Step-by-step guide from notebook to exposed endpoint. |
| Data Access | Connect to and consume CDH data. |
| CAIP SDK | Python SDK and runtime library for building and running pipelines. |
| Staging | Stage pipelines and models within a single installment. |
| GPU Compute | On-demand GPU nodes for training and inference. |
| Observability | Grafana dashboards and CDLS logging. |
| IAM | Roles, rights, and cluster login. |
| CI Trust | Secure GitHub ↔ workspace automation. |
| Data Visualization | Streamlit apps for data and model monitoring. |
| API Gateway | Kong-based API management and ingress. |
| Customer-Managed Resources | Naming and permissions for customer-managed secrets. |