> ## Documentation Index
> Fetch the complete documentation index at: https://docs.open-metadata.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Large Language Models

> Catalog large language model services and models through the REST API and Java SDK

# Large Language Models

The large language model (LLM) catalog uses a service-and-model hierarchy:

```text theme={null}
LLM Service
└── LLM Model
```

An **LLM Service** represents a provider or deployment boundary. Common types include OpenAI, Anthropic, and Amazon Bedrock. Other types cover Vertex AI, Ollama, and custom providers. An **LLM Model** captures identity, specifications, and evaluation results. It also tracks deployment, cost, and governance status.

An LLM service name is globally unique. An LLM model uses the fully qualified name `service.model`, such as `openai_production.gpt_5`.

## LLM Service Endpoints

| Method   | Endpoint                                             | Description                              |
| -------- | ---------------------------------------------------- | ---------------------------------------- |
| `GET`    | `/v1/services/llmServices`                           | List LLM services                        |
| `GET`    | `/v1/services/llmServices/{id}`                      | Retrieve a service by ID                 |
| `GET`    | `/v1/services/llmServices/name/{name}`               | Retrieve a service by name               |
| `POST`   | `/v1/services/llmServices`                           | Create a service                         |
| `PUT`    | `/v1/services/llmServices`                           | Create or update a service               |
| `PATCH`  | `/v1/services/llmServices/{id}`                      | Patch a service by ID                    |
| `PATCH`  | `/v1/services/llmServices/name/{fqn}`                | Patch a service by name                  |
| `PUT`    | `/v1/services/llmServices/{id}/testConnectionResult` | Store a connection test result           |
| `PUT`    | `/v1/services/llmServices/{id}/followers`            | Add the authenticated user as a follower |
| `DELETE` | `/v1/services/llmServices/{id}/followers/{userId}`   | Remove a follower                        |
| `GET`    | `/v1/services/llmServices/{id}/versions`             | List versions                            |
| `GET`    | `/v1/services/llmServices/{id}/versions/{version}`   | Retrieve a version                       |
| `DELETE` | `/v1/services/llmServices/{id}`                      | Soft-delete or hard-delete by ID         |
| `DELETE` | `/v1/services/llmServices/async/{id}`                | Delete asynchronously                    |
| `DELETE` | `/v1/services/llmServices/name/{name}`               | Soft-delete or hard-delete by name       |
| `PUT`    | `/v1/services/llmServices/restore`                   | Restore a soft-deleted service           |

The required create fields are `name` and `serviceType`. Valid service types are `OpenAI`, `Anthropic`, `AzureOpenAI`, `Bedrock`, `VertexAI`, `Ollama`, `HuggingFace`, and `CustomLLM`.

## LLM Model Endpoints

| Method   | Endpoint                                | Description                                        |
| -------- | --------------------------------------- | -------------------------------------------------- |
| `GET`    | `/v1/llmModels`                         | List models, optionally filtered by `service`      |
| `GET`    | `/v1/llmModels/{id}`                    | Retrieve a model by ID                             |
| `GET`    | `/v1/llmModels/name/{fqn}`              | Retrieve a model by fully qualified name           |
| `POST`   | `/v1/llmModels`                         | Create a model                                     |
| `PUT`    | `/v1/llmModels`                         | Create or update a model                           |
| `PATCH`  | `/v1/llmModels/{id}`                    | Patch a model by ID                                |
| `PATCH`  | `/v1/llmModels/name/{fqn}`              | Patch a model by fully qualified name              |
| `PUT`    | `/v1/llmModels/{id}/followers`          | Add the authenticated user as a follower           |
| `DELETE` | `/v1/llmModels/{id}/followers/{userId}` | Remove a follower                                  |
| `GET`    | `/v1/llmModels/{id}/versions`           | List versions                                      |
| `GET`    | `/v1/llmModels/{id}/versions/{version}` | Retrieve a version                                 |
| `DELETE` | `/v1/llmModels/{id}`                    | Soft-delete or hard-delete by ID                   |
| `DELETE` | `/v1/llmModels/async/{id}`              | Delete asynchronously                              |
| `DELETE` | `/v1/llmModels/name/{fqn}`              | Soft-delete or hard-delete by fully qualified name |
| `PUT`    | `/v1/llmModels/restore`                 | Restore a soft-deleted model                       |

The required create fields are `name`, `baseModel`, and the parent service's fully qualified name in `service`. Optional metadata covers `modelVersion`, `modelProvider`, and `modelSpecifications`. Evaluation, cost, and deployment fields track operations. Regulatory compliance, ownership, domains, and data products add governance context.

## Create a Service and Model

The Java example assumes a configured `OpenMetadataClient` named `client`.

<RequestExample dropdown>
  ```java Java SDK theme={null}
  import org.openmetadata.schema.api.ai.CreateLLMModel;
  import org.openmetadata.schema.api.services.CreateLLMService;

  var service = client.llmServices().create(
      new CreateLLMService()
          .withName("openai_production")
          .withDisplayName("OpenAI Production")
          .withServiceType(CreateLLMService.LlmServiceType.fromValue("OpenAI"))
  );

  var model = client.llmModels().create(
      new CreateLLMModel()
          .withName("gpt_5")
          .withDisplayName("GPT-5")
          .withBaseModel("gpt-5")
          .withModelProvider("OpenAI")
          .withService(service.getFullyQualifiedName())
  );

  System.out.println(model.getFullyQualifiedName());
  ```

  ```bash REST API theme={null}
  curl -X POST "{base_url}/api/v1/services/llmServices" \
    -H "Authorization: Bearer {access_token}" \
    -H "Content-Type: application/json" \
    -d '{
      "name": "openai_production",
      "displayName": "OpenAI Production",
      "serviceType": "OpenAI"
    }'

  curl -X POST "{base_url}/api/v1/llmModels" \
    -H "Authorization: Bearer {access_token}" \
    -H "Content-Type: application/json" \
    -d '{
      "name": "gpt_5",
      "displayName": "GPT-5",
      "baseModel": "gpt-5",
      "modelProvider": "OpenAI",
      "service": "openai_production",
      "governanceStatus": "Approved"
    }'
  ```
</RequestExample>

The v2.0 Java SDK exposes both resources through the direct client. The v2.0 Python core SDK has no typed routes for LLM services or models, so use the REST API.

## Update and Delete

Send an RFC 6902 JSON Patch document with `Content-Type: application/json-patch+json` to a PATCH endpoint. Set `hardDelete=true` on a DELETE request for permanent deletion. Restore a soft-deleted entity by sending its `id` to the corresponding `/restore` endpoint.