> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# NIM REST API

> Send synchronous prediction requests to Kumo Relational through its NIM-compatible REST API

Kumo Relational exposes the Universal Structured Data Foundation Models API, a shared REST request and response format for structured-data models.
Use the Python SDK for standard prediction workflows.
Call the REST API directly when integrating Kumo Relational into a service or generating a client from its OpenAPI schema.

## Choose an endpoint

* **Self-hosted NVIDIA NIM:** NVIDIA NIM is the deployed inference service that hosts Kumo Relational and exposes its REST endpoint.
  Use the base URL of your deployment.
  The NIM API contract does not require authentication, but a gateway in front of the deployment might require an API key or another credential.
* **NVIDIA API Catalog:** Use the server URL and bearer-token authentication shown in the published API reference.

## Run a one-shot prediction

The `POST /v1/predictions` endpoint accepts one complete request in the shared API format.
The request specifies the model, prediction task, relational schema, labeled context, rows to predict, requested outputs, and optional inference settings and metadata.
The endpoint returns the prediction synchronously and does not create a durable prediction resource that you can retrieve later.

To send a request:

1. Copy the complete `multitable_binary_classification` request example from the published Kumo Relational NIM API reference and save it as `request.json`.
2. Set `KUMO_RELATIONAL_API_ENDPOINT` to the base URL of your deployment, such as `http://localhost:8000`.
3. Run the following command:

```bash
curl --fail-with-body \
  --request POST \
  --url "${KUMO_RELATIONAL_API_ENDPOINT}/v1/predictions" \
  --header "Content-Type: application/json" \
  --data @request.json
```

If your gateway requires authentication, add its required authentication header to the command.
For the NVIDIA API Catalog endpoint, add `--header "Authorization: Bearer ${NVIDIA_API_KEY}"`, as shown in the generated API examples.

## Response and errors

A successful response contains a response-level correlation ID, the model name, prediction items, and metadata.
Each prediction item includes a zero-based `row_index` that identifies its position among the requested rows.
It can also include the input `instance_id` in the `id` field and requested output fields such as `prediction` and `probabilities`.

Errors use the `application/problem+json` media type with fields such as `type`, `status`, `title`, `detail`, and `code`.
See the generated API reference for the complete request schema, response schema, error model, and runnable examples.