CAIP Offboard v2: All-New CAIP
CAIP2 is the next stage of the Connected AI Platform: a set of completely self-contained, managed, API-driven, and modular service offerings. Everything is self-service with rapid provisioning and no per-use-case platform overhead.
Unlike Managed Kubeflow (the same Kubeflow product delivered under a managed ops model), CAIP2 introduces brand-new service offerings. It is currently being built up — some offerings are already live, others are in active development. The service portfolio below is the complete picture.
Service Portfolio
CAIP2 groups its offerings into a handful of categories — from data access and developer tooling through Classic ML, Generative AI, and the underlying infrastructure. The map below shows the portfolio at a glance:
The sections below break this map down category by category. Each service offering links to its documentation where it exists. For offerings exposed as APIs, the EMEA / CN / US columns show per-region availability and, where live, the endpoint. Some offerings are consumed within the platform and don't expose a dedicated endpoint — these are shown as bullet lists or marked —.
Status: 🟢 Live · 🔵 Planned · 🔴 Not planned yet
Data
Data offerings provide managed access to data sources; they are consumed from within your workloads rather than through a dedicated endpoint.
- 🟢 Data Access to CDH
- 🔵 Data Access to Vehicle Shadow — Planned
Core Tools
Core Tools are the shared building blocks every use case relies on — the developer experience for building and shipping, and the infrastructure it runs on. The service map below shows how they fit together:
DevEx
The dedicated CAIP2 Notebooks offering is being recreated and will be available soon. Until then, use Notebooks as part of Managed Kubeflow.
| Service Offering | EMEA | CN | US |
|---|---|---|---|
| Notebooks | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
| SDK Client library — no endpoint. | — | — | — |
| Portal | 🟢 caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
Infrastructure
| Service Offering | EMEA | CN | US |
|---|---|---|---|
| Distributed AI Compute (Ray Jobs, Batch Execution) | 🟢 workflows.api.caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
| CAIP Environments | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
| CAIP Apps | 🟢 apps.api.caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
Onboard
Deploying models to run onboard the vehicle (on CDC). This is a distinct capability, independent of the CAIP evolution stages — see the dedicated CAIP Onboard section.
- 🔵 CAIP Onboard on CDC — Planned
Classic ML
The Classic ML service map covers the traditional ML lifecycle — a feature store, training, and offboard/onboard inference:
| Service Offering | EMEA | CN | US |
|---|---|---|---|
| Feature Store | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
| Offboard Inference | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
| Onboard Inference (see Onboard) | — | — | — |
Classic ML + GenAI
The dedicated CAIP2 Model Registry is being built and will be available soon. Until then, use the model registry as part of Managed Kubeflow (MLflow).
| Service Offering | EMEA | CN | US |
|---|---|---|---|
| Workflows | 🟢 workflows.api.caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
| Model Registry | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
GenAI
CAIP2's Generative AI offerings span two complementary areas. The Generative AI map covers model access — the LLM API, self-hosted models, and guardrails:
The Agentic AI map covers building and operating agents — the Agent Foundry (Agents API & SDK), agent observability with Langfuse, and the MCP and Agent registries:
| Service Offering | EMEA | CN | US |
|---|---|---|---|
| Self-Hosted LLMs | 🟢 llm.api.caip.bmw.cloud | 🟢 llm.api.caip.bmwchina.cloud | 🔴 Not planned yet |
| LLM API | 🟢 llm.api.caip.bmw.cloud | 🟢 llm.api.caip.bmwchina.cloud | 🔴 Not planned yet |
| Agent Foundry (Agents API, SDK) | 🟢 agents.api.caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
| Agent Observability (Langfuse) | 🟢 langfuse.caip.bmw.cloud | 🔵 Planned | 🔴 Not planned yet |
| MCP Registry | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
| Agent Registry | 🔵 Planned | 🔵 Planned | 🔴 Not planned yet |
Get started
- Onboarding (LLM API) — create a CAIP Space, get an API key, and start consuming the LLM API.
- Service Requests — request a CAIP Space or an LLM API key.
- Web Portal — manage Spaces, users, and API keys.
- Authentication — how to authenticate CAIP APIs (WebEAM User, WebEAM M2M, API Keys).