Connected AI Platform Documentation Hub
Welcome to Connected AI Platform's Documentation Hub. Here you can find all information you need to understand, work, develop, and operate with Connected AI Platform (CAIP).
Ready to accelerate your AI journey? Whether you're building Classic ML models or exploring the latest in Generative AI, CAIP provides everything you need to innovate, collaborate, and scale — securely and efficiently.
What is Connected AI Platform?
The Connected AI Platform (CAIP) is a ready-to-use and scalable off- and onboard AI infrastructure for data science teams to develop, train, deploy, and monitor machine learning models and AI applications. It is the BMW standard solution for AI-based connected services and products.
It is not built in isolation, but integrated with the Connected Drive Cloud Platform ORBIT and uses ORBIT's existing services such as its Monitoring & Logging stack, Docker Registry, and Microservice Runtime. CAIP extends this offering and enables users to integrate with other fully-managed AWS services (e.g. AWS Glue, Amazon EMR).
Platform Evolution
CAIP has evolved through three major offboard (cloud-side) stages, plus a separate onboard (in-vehicle) offering. This documentation is organized around them so you can quickly find what applies to you and see what is already live.
Offboard Platform Evolution
| Stage | What it is | Status | Where to look |
|---|---|---|---|
| CAIP Offboard v1: Classic Kubeflow — Distributed Kubeflow | Bring-your-own-cluster: prototype on the shared AI-Research cluster; for production, request your own Orbit (EKS) cluster with CAIP provisioned into it. High duplication and platform overhead. | 🔴 Retired | Platform Evolution & Migration |
| CAIP Offboard v1.5: Managed Kubeflow | The same Kubeflow product, now delivered on a managed cluster as isolated, namespace-based Kubeflow Workspaces. No cluster to run — CAIP operates all infrastructure. | 🟢 Live | CAIP Offboard v1.5: Managed Kubeflow |
| CAIP Offboard v2: All-New CAIP — Service Offerings | A set of completely self-contained, managed, API-driven, modular offerings (LLM API, Agents, Workflows, Apps, Inference, …). Fully self-service, rapid provisioning. | 🟡 Being built up | CAIP Offboard v2: All-New CAIP |
CAIP Offboard v1 → CAIP Offboard v1.5 changed the operating model only — the Kubeflow offering stayed the same. CAIP Offboard v2: All-New CAIP adds brand-new service offerings on a new, modular stack.
Onboard Platform Evolution
| Stage | What it is | Status | Where to look |
|---|---|---|---|
| CAIP Onboard | Deploys AI models to run directly on the vehicle (in-vehicle / on-device inference) via the CDC platform on IP Next ECU, for non-safety digital-service use cases. Independent of the offboard stages above. | 🟢 Live | CAIP Onboard |
CAIP Onboard is a distinct, onboard (in-vehicle) capability — it runs alongside, not as a replacement for, any of the offboard stages.
Where should I start?
- Running Classic ML on Kubeflow? Start with CAIP Offboard v1.5: Managed Kubeflow → Onboarding.
- Looking for LLMs, agents, workflows, or apps? Explore CAIP Offboard v2: All-New CAIP and check the status table to see what's live.
- Need to run a model directly on the vehicle? See CAIP Onboard.
- On the AI-Research cluster or
code.connected.bmw? You must migrate — see Platform Evolution & Migration. - Not sure which CAIP capability solves your problem? Browse the Cookbook — it maps common customer challenges to CAIP "recipes".
- Need help? Visit Support.
All Doc Sections
Every top-level chapter of this manual, for quick orientation:
| Section | What you'll find there |
|---|---|
| General | Cross-cutting topics that apply to all CAIP versions: administrative prerequisites (APP-ID) and the AI Dataschema data layer. |
| CAIP Offboard v1.5: Managed Kubeflow | The fully managed, Kubeflow-based ML platform — onboarding, service requests, and the full service portfolio (notebooks, pipelines, experiment tracking, model serving). |
| CAIP Offboard v2: All-New CAIP | The modular, self-service, API-driven stack — LLM API, Agents Foundry, Workflows, Apps, Inference, and the underlying platform (Spaces). |
| CAIP Onboard | Deploying AI models to run directly on vehicle hardware (in-vehicle inference) via CDC IP-Next. |
| Platform Evolution & Migration | Reference material for retired offerings — the completed migrations off code.connected.bmw and the AI-Research cluster — plus Traefik ingress guidance. |
| Support | How to raise a production incident, get technical support or consulting via Microsoft Teams, or book a walk-in session. |
| Cookbook | Not sure which CAIP offering fits your problem? Recipes that map common customer challenges to the CAIP capabilities that solve them. |
| FAQ | Answers to frequently asked questions about using CAIP. |
| Roadmap | The current product roadmap for the Connected AI Platform. |
| Release History | A running, customer-facing index of unified CAIP releases by offering family and product area. |