Deploying a foundation model in SAP AI Core, step by step

Before you can send a single prompt to a large language model on SAP Business Technology Platform, the model has to be deployed in SAP AI Core. Deployment is what turns a catalog entry, say Claude Sonnet 4.5 or GPT-4o, into a running endpoint with its own URL that your applications can call.

There are two ways to do this: through the AI Core API or through the AI Launchpad UI. This guide walks through the Launchpad approach, which is the fastest path to a working endpoint and needs no scripting.

Prefer the API route? SAP documents it in the Deploy models section of the AI Core Service Guide.

Prerequisites

• An SAP AI Core instance provisioned on SAP BTP
• Access to SAP AI Launchpad with the Generative AI Hub enabled
• Sufficient entitlement / quota for the model you want to deploy

1. Open the Generative AI Hub

Log in to SAP AI Launchpad and go to Generative AI Hub → Model.

2. Select the model to deploy

Pick the model you want from the list. In this example it is Claude Sonnet 4.5.

3. Start the deployment

Click the Deploy button. The deployment process begins immediately.

4. Track the deployment status

Go to ML Operations → Deployment to monitor progress. The first deployment can take a while to finish, so do not be surprised if the status sits in an intermediate state for several minutes.

Before it is ready:

After it is ready:

Important: you must wait until the deployment reaches the RUNNING status before you try to use it. Calling the endpoint before it is running will fail.

5. Copy the generated URL

Once the deployment is running, copy the generated deployment URL. This is the endpoint your applications, SDKs and REST clients will use to run inference.

What's next

With a running deployment and its URL in hand, you are ready to start calling the model. From here you can:

• Consume the endpoint from the ABAP AI SDK via ISLM scenarios
• Send raw REST requests with a tool like Bruno or Postman, as we do with SAP-RPT-1
• Wire the endpoint into an orchestration or RAG pipeline

Deployment is a one-time setup per model, but it is the gate everything else passes through, so it is worth getting comfortable with these steps early.

Where Boidra fits

We build custom agents that call models through AI Core, so your agents inherit the same scaling, governance and model choice, and stay portable over open protocols like A2A and MCP. The runtime is SAP's; the agents on top are yours.