Create a service

Available in VPC.

To test retrieval models using the RAG system after completing the service subscription, you must index the data files that will be used for search. This section describes the service creation process, data configuration, and service deletion, and provides guidance on the RAG Development Guide.

The service creation process consists of the following steps:

• Create a service
  • 1. Service settings
  • 2. Data settings
  • 3. Retrieval settings
  • 4. Generation settings
  • 5. Final confirmation
• Check service
• Delete service
• RAG Development Guide

Create a service

This section describes how to create a service from the Services menu.

1. In the VPC environment of the NAVER Cloud Platform console, navigate to > Services > AI Services > RAG.
2. Click Services.
3. Click [Create service] button.
4. On the Service creation page, enter the service settings and click [Next] button.
5. In the Data settings step, enter the required information and click [Next] button.
6. In the Retrieval settings step, enter the required information and click [Next] button.
7. In the Generation settings step, connect to CLOVA Studio and enter the required information, then click [Next] button.
8. Review the entered information in the final confirmation step.
9. Click [OK] button.

1. Service settings

Enter the following information in Service settings.

Service settings

• Service name: Can enter 3 to 20 characters.
• Service description: Up to 50 characters

2. Data settings

Configure the following items in Data settings.

Select data integration method

• Integration method Choose either NCP Object Storage or RESTful API.
  • NCP Object Storage Indexes data stored in an Object Storage bucket.
  • RESTful API: Directly calls index data through an API.

Data settings

If you select NCP Object Storage as the integration method, configure the data files to be indexed.

• Object Storage buckets Shows the number of buckets created in Object Storage.
• [Refresh]: Refreshes the file list in the selected bucket.
  • Object Storage buckets Shows available buckets created by Object Storage.
    • Maximum file size: 100 MB
    • Image formats: .jpg, .png, .tiff
    • Document formats: .pdf, .pptx, .docx, .hwp, .hwpx, .txt
    • Data file formats: .md, .json, .xml
• File list Shows the number of data items stored in the bucket.
  • File name Name of the data file in the bucket.
  • Last modified Last modified date of the data file.
  • Document Click [View document] button to open a preview interface.
  • [Add to index] button Select files from the list and click [Add to index] button to add them to the index target list.
• Index targets Shows the number of files added as index targets.
• [Refresh]: Refreshes the index target list.
  • File name Name of the file added for indexing.
  • Bucket Object Storage bucket name.
  • Last modified Last modified date of the data file.
  • File size: Size of the data file.
  • Document Click [View document] button to open a preview interface.
  • Index status: Status of index processing (Failed, Completed, Attention needed, Indexing, Waiting).
  • Delete Click [Delete] button to remove the file from the list.

Data integration settings

If indexing image files, configure OCR-related settings.

• Model selection Only the default model is currently provided.
• OCR option Extracts text in images using CLOVA OCR.
  • Up to 10 pages supported for PDFs.

3. Retrieval settings

Enter the following information in Retrieval settings.

Chunking method

• Method: Splits long text into smaller sections for easier retrieval. Only the default option is available.
• Passage: Defines the size of text segments used for retrieval.
  • Enter 256 to 4096
  • (default: 512).
• Overlap A configuration value that adjusts the overlap ratio between individual areas in the search results to properly separate the text.
  • You can enter a value from 0 up to half of the Passage value (0 to Passage/2).
  • (default: 128).

Embedding model

• Model name: Choose the model used for embedding. Only the default option is provided.

Vector DB

• DB: Selects the vector database. Only the default option is provided.
• K: Number of sources displayed in search results.
  • Enter 1 to 10
  • (default: 5).

4. Generation settings

Enter the following information in Generation settings.

Model settings

• [CLOVA Studio API Guide]: Link to the required integration instructions
  • [Visit link]: Opens the CLOVA Studio service interface
• Model name: Name of the LLM used to generate answers
  • Check model: Opens the CLOVA Studio interface
• CLOVA Studio API Key: API Key required to connect to CLOVA Studio

Prompt settings

• Enter instructions that define how the service should generate responses.
  • Default settings Applied when no custom prompt is provided.

Parameter settings

• Top P: Calculates the probability of each token by checking how often the language model attempts to generate it. If a token's probability does not meet the threshold P, the token is excluded from the result.
• Top K: Includes only the top K tokens ranked by relevance and excludes all remaining tokens.
• Maximum Tokens: Maximum number of tokens used for generation. A value between 300 and 500 is recommended.
• Temperature: Lower values generate more deterministic responses. Higher values increase output diversity but may reduce response quality.

5. Final confirmation

Review the information entered in steps 1-4 and click [OK] button.

Check service

You can view the list of created services in the NAVER Cloud Platform console under AI Services > RAG > Services.

To check service information from the Services menu:

1. In the VPC environment of the NAVER Cloud Platform console, navigate to > Services > AI Services > RAG.
2. Click Services.
3. In the Services list, click the button for the service you want to view.
4. When the service information area expands, review the Service informationdetails:
   • Service name: Name set when creating the service.
   • Service description: Description set when creating the service
   • Index status: Indexing status for data files
     • Failed: Indexing failed for all data files
     • Completed: Indexing completed successfully for all data files
     • Attention needed: One or more data files failed to index
     • Indexing: One or more data files are currently being indexed
     • Waiting: Any other state that is not Failed, Completed, Attention needed, or Indexing

Delete service

To delete a service from the list:

1. In the VPC environment of the NAVER Cloud Platform console, navigate to > Services > AI Services > RAG.
2. Click Services.
3. In the Services list, click button for the service you want to delete.
4. Click menu, then the Delete menu item.
5. In the delete service, click [Delete] button.
6. The service is removed from the Services list.

Check RAG development guides

If you integrate the RAG service with APIs, see the following development guides:

• RAG Service API guide: Opens the API guide for the RAG service.
• CLOVA Studio user guide: User guide for CLOVA Studio, the required integration service for using LLMs.
• API call and authentication guide:: Describes how to use NAVER Cloud Platform RESTful APIs.