Skip to main content

Connected AI Platform FAQ

Find answers to the most common questions about using the Connected AI Platform below.

🚀 Getting Started​

How do I get started with the Connected AI Platform?

It depends on what you need. If you want to run Classic ML on Kubeflow (notebooks, pipelines, model serving), start with the Managed Kubeflow onboarding guide. If you're looking for LLMs, agents, workflows, or apps, start with CAIP 2.0 instead:

What's the difference between CAIP 1.0, 1.5, and 2.0? Which one should I use?

CAIP has evolved through three stages:

  • CAIP 1.0 — Distributed Kubeflow (🔴 retired): bring-your-own-cluster, prototyped on the shared AI-Research cluster.
  • CAIP 1.5 — Managed Kubeflow (🟢 live): the same Kubeflow product, now delivered as a fully managed, namespace-based Workspace — no cluster to run yourself.
  • CAIP 2.0 — Service Offerings (🟡 being built up): brand-new, self-contained, API-driven offerings (LLM API, Agents, Workflows, Apps, Inference, …).

If you're doing Classic ML with Kubeflow pipelines/notebooks, use CAIP 1.5. If you need LLMs, agents, or other modular API-driven services, use CAIP 2.0. See the Documentation Hub for the full comparison.

I want to use the Connected AI Platform, but I don't have an APP-ID. What should I do?

Every use case requires its own APP-ID. This is a simple administrative process. Follow our step-by-step guide to create and link your APP-ID:

🧩 Requests & Self-Service​

How can I request services, features, or support for the Connected AI Platform?

Most changes are handled via a Service Request (workspace provisioning, user management, CI Trust, CDH connection, CAIP Space, API keys, …). If your request genuinely isn't covered by a Service Request, use our PR-based Self-Service instead. Start here:

Do I need a CAIP Space to use Managed Kubeflow (CAIP 1.5)?

Yes. A CAIP Space is your umbrella project container on the Connected AI Platform, and a Managed Kubeflow Workspace is provisioned inside it. Make sure you already have a CAIP Space — or request one first — before requesting a new Managed Kubeflow Workspace.

How do I add more users to my CAIP use case?

It depends on which layer you mean:

  • To add users to your CAIP Space (Self-Service Portal access, API keys, etc.), submit a Manage CAIP Space Users Service Request.
  • To add users to your Managed Kubeflow Workspace/UI for a specific namespace, add their email address to your namespace in product.yaml and raise a Pull Request in the products repository.

🧠 Managed Kubeflow (CAIP 1.5)​

How do I log in to my Managed Kubeflow Workspace / AWS account?

Use the orbit-use CLI to authenticate and assume your caip-developer role — the old AWS CLI / BMWAWS CLI login has been deprecated. See:

How do I get GPU compute for my notebook or pipeline?

CAIP provides pre-configured GPU nodes that spin up on demand when you request the right notebook/pipeline configuration (GPU count, vendor, tolerations, and enough CPU/RAM/volume for GPU workloads). See:

How do I enable MLflow (Experiment Tracking & Model Registry) for my workspace?

MLflow is disabled by default to save costs. Enable it by adding mlflow.enabled: true to your namespace in product.yaml and opening a Pull Request:

How do I connect my Managed Kubeflow Workspace to Cloud Data Hub (CDH)?

First set up the initial CDH connection for your product, then link individual datasets to your namespace:

My Streamlit app or a custom IAM role (e.g. a "data-visualization" role) gets an AssumeRole error when accessing CDH data, even though CDH is connected — why?

By default, only your namespace's *-cd4ml, developer, and execution roles are allowed to assume caip-role-managed-cdh. Any other custom role (like a Streamlit data-visualization role) must be explicitly granted sts:AssumeRole on caip-role-managed-cdh via a policy, added by raising a Pull Request in the products repository:

How do I set up CI Trust so GitHub Actions can deploy to my workspace?

CI Trust establishes a secure, credential-free connection between your GitHub repository and your Managed Kubeflow Workspace using OIDC:

How do I restrict or authenticate access to my KServe InferenceService endpoint (API keys, mTLS, M2M)?

It depends on how the endpoint is exposed:

  • For private endpoints reached via the Kong ingress, use WebEAM M2M bearer-token authentication.
  • For endpoints exposed to the public internet via Traefik, protect them with basic-auth credentials configured as a Traefik Middleware.

See:

Can I make my KServe model endpoint accessible from the public internet, not just the BMW intranet?

Yes — this is a supported, documented deployment option and includes setting up authentication credentials via a Traefik Middleware:

My pipeline/notebook can't reach a database or service that lives in another (non-CAIP) AWS account — what should I do?

This is generally not a CAIP/Orbit-side issue — connectivity to resources outside CAIP's own accounts depends on the target environment's network configuration. Before raising a ticket with us, check with the owning team of the target resource whether:

  • Its Security Group / Network ACL / database allowlist permits traffic from your CAIP namespace.
  • If the target uses a private DNS zone, whether VPC peering (or a Transit Gateway) and private hosted zone association have been set up between your CAIP VPC and theirs — DNS resolution failures usually indicate this step is missing.

If you determine the CAIP side needs a change (e.g. an allowlist or peering request), reach out to Kubeflow-support with the specifics.

⚡ CAIP 2.0​

How do I get an API key to use the LLM API (CAIP 2.0)?

Request or create a CAIP Space, then request an API key via the Self-Service Portal or a Service Request. Only UUID-format keys are accepted for the RoW LLM API:

How do I authenticate against CAIP 2.0 APIs from my own service?

CAIP 2.0 supports three methods: API Keys (fastest, best for prototyping), WebEAM User (signing in as yourself with your q-Number), and WebEAM M2M (recommended for productive service-to-service calls). See:

Can I see the cost of my CAIP usage per API key or project?

Not yet — granular cost visibility per API key/project is planned and expected to become available directly in the CAIP Self-Service Portal. Check the Roadmap for the latest timeline, or ask in the Community channel for the current status.

🔄 Platform Evolution & Other Offerings​

I'm still using code.connected.bmw or the AI-Research cluster — what should I do?

Both have been retired — migration is complete. New setups should use bmw.ghe.com (GHE), and workloads should run on Managed Kubeflow. The guides below remain as reference for any stragglers:

My workflow can't resolve a CAIP GitHub Action (e.g. connected-ai/deploy-caip-streamlit-action, read-caip-config-action, teams-notify-action) — "repository not found"

These actions have been migrated from code.connected.bmw to bmw.ghe.com/connected-ai along with the rest of the CAIP repositories. Update your workflow's uses: reference to point to the GHE-hosted action (same name, under the connected-ai org on bmw.ghe.com) and pin it to the version tag you previously used. If a specific action still 404s, ask in the Community channel — it may not have been migrated yet.

After the nginx → Traefik ingress migration, my endpoint is no longer reachable from CI/CD or another external client — what should I do?

This is usually not a CAIP configuration issue but a network firewall/allowlist that hasn't been updated for the new ingress. Raise an Incident requesting the source IP range (e.g. your CI runner's egress IPs) be allowlisted for the migrated endpoint:

What is CAIP Onboard? Is it the same as Managed Kubeflow or CAIP 2.0?

No — CAIP Onboard is a separate, independent capability for deploying AI models to run directly on the vehicle (on-device/in-vehicle inference), rather than in the cloud. It doesn't belong to Managed Kubeflow or the CAIP 2.0 service portfolio. See:

🆘 Support​

How do I get help or report a production incident?

For outages or critical malfunctions, raise a production incident right away. For how-to questions, design guidance, or general support, ask in our Microsoft Teams Community or book a Walk-In session: