## Introduction  
The Model Download API provides external access to retrieve download URLs for artifacts generated in Rasa Studio, facilitating CI/CD integration.  
Built with GraphQL, this API allows queries to request trained assistant models and associated input data in YAML format.

## Requirements  
- Required API Role `Model download urls`. See [Authorization](/content/docs/reference/api/studio/api-authorization/index.html) to learn how to get an access token.

- Assistant API ID. See [authorization guide for configuration](/content/docs/studio/installation/setup-guides/authorization-guide/index.html) for more details.

- Assistant Version Name. See [user guide for training](/content/docs/studio/test/train-your-assistant/index.html) for more details.

## Available APIs

### Query: `ModelDownloadUrls`  
**Schema:**  
```graphql
  type Query {
    modelDownloadUrls(input: ModelDownloadUrlsInput!): ModelDownloadUrlsOutput!
  }

input ModelDownloadUrlsInput {
    assistantId: ID!
    assistantVersionName: ID
  }

type ModelDownloadUrlsOutput {
    modelFileUrl: String!
    modelInputUrl: String!
  }
```

#### Input: `ModelDownloadUrlsInput`  
| Field                    | Description                                             |
|--------------------------|---------------------------------------------------------|
| `assistantId`           | Unique identifier of the assistant.                     |
| `assistantVersionName`   | Optional version of the trained assistant. Default is `latest`. |

#### Output: `ModelDownloadUrlsOutput`  
| Field           | Description                                               |
|-----------------|-----------------------------------------------------------|
| `modelFileUrl`  | Url to download the model file `<assistant-version-name>.tar.gz`. |
| `modelInputUrl` | Url to download the model input data in YAML format.    |

## Usage  
The query can be used to fetch the download URLs for the trained model and the input data in YAML format for an assistant.

### Example `curl` Request  
Request the Model Download API using e.g. `curl`:
```bash
curl -X POST https://<api-url>/api/graphql \
-H "Content-Type: application/json" \
-H "Authorization: Bearer <access-token>" \
-d '{\
    "query": "query ModelDownloadUrls($input: ModelDownloadUrlsInput!) { modelDownloadUrls(input: $input) { modelFileUrl modelInputUrl } }",\
    "variables": {\
      "input": {\
        "assistantId": "<assistant-id>",\
        "assistantVersionName": "<assistant-version-name>"\
      }\
    }\
  }'
```

### Example Response  
The response will contain the download URLs for the trained model and the model input data.
```bash
{"data":
  {"modelDownloadUrls":
    {
      "modelFileUrl":"https://<api-url>/api/v1/download/model-file/<assistant-version-name>",
      "modelInputUrl":"https://<api-url>/api/v1/download/model-input/<assistant-version-name>"
    }
  }
}
```

### Download model file  
To download the model file, request the `modelFileUrl` with authorization header.
```bash
curl "https://<api-url>/api/v1/download/model-file/<assistant-version-name>" \
    -H "Authorization: Bearer <access-token>" \
    -O -J
```
The downloaded model file `<assistant-version-name>.tar.gz` can be used to deploy the trained model.

### Download model input data  
To download the model input YAML files, request the `modelInputUrl` with authorization header.
```bash
curl "https://<api-url>/api/v1/download/model-input/<assistant-version-name>" \
    -H "Authorization: Bearer <access-token>" \
    -O -J
```
The downloaded model input zipped directory `<assistant-version-name>_model-input.tar.gz` contains the model input data in YAML format:
- `data/flows.yaml`
- `data/nlu.yaml`
- `config.yaml`
- `credentials.yaml`
- `domain.yaml`
- `endpoints.yaml`