Setting up Kuroco RAG

Overview

Kuroco RAG is a service that enhances the accuracy and reliability of generated AI models using information obtained from external sources.

This document aims to understand what can be achieved with Kuroco RAG by creating a GPT that generates answers by referencing content information stored in Kuroco.
The explanation will proceed using a Kuroco account, so if you do not have a Kuroco account, please register for an account from the free trial.

info

If you wish to use Kuroco RAG, please contact us below.

• Support

What You Will Learn

In the following steps, we will create a custom GPT that generates answers by referencing content information stored in Kuroco.

• Prerequisites
• Setting Up Kuroco
  • Enabling OpenAI
  • Registering Content
  • Setting Up the API
• ChatGPT Configuration
  • Creating GPT

Prerequisites

You will need the following accounts, so please create them if you do not have them.

• Kuroco Account: You need a valid Kuroco account. If you do not have an account yet, please register for an account from the free trial.

• ChatGPT Plus, Team, or Enterprise Plan: To use GPTs (customized ChatGPT features), you need to have either the ChatGPT Plus, ChatGPT Team, or ChatGPT Enterprise plan.
  If you do not have one, please consider upgrading.

Setting Up Kuroco

Enabling OpenAI

Click on [External System Integration] -> [AI].

Enable "OpenAI" by checking the checkbox and click [Update].
The API key field can be left blank as we will be using Kuroco's API key.

Registering Content

Register the content that ChatGPT will reference in Kuroco.
The registered content will be set up so that it can only be accessed via the API by the customized ChatGPT and cannot be referenced by others externally.

Creating the Content Structure

Click [Add] on the Content structure list page.

Configure the settings with the following details.

General

Field	Setting
Name	RAG demo

Field Settings

Field	Slug	Field Name	Field Setting
ext_1	None	WYSIWYG	WYSIWYG

Search Settings

Item	Setting
Convert to vector data	Enable
Embedding model	text-embedding-3-small
Keyword template (AI/Vector)	Default

Once the settings are done, click on [Add] to add the content definition.

Adding the contents

Register information for OpenAI to use as content.

Click on [Add] from the Content list screen.

Here, we have added the following 3 contents as examples.

Remote Work Guidelines

Employee Benefits Guidebook

Contact Information

info

The example shows the registration of internal documents for an AI assistant intended for internal use.
In the tutorial, there are only three contents, but in actual operation, the entire employment regulations will be saved and used to allow the AI to search and check desired items from a large amount of text on behalf of the user.

Checking Vector Data

When a content is added or updated, it is automatically converted to vector data.
Click on [More] -> [Vector Template] from the content editing screen to check the generated keywords.

Additionally, you can check the progress and results of vector data generation on the [External System Integration] -> [AI] page.

Setting up the API

Next, prepare the API for the custom ChatGPT to retrieve content information.

Adding the API

Click "Add" from the API section in the Kuroco admin panel.

In the API creation screen, enter the following and click on "Add".

Item	Setting
Title	RAG demo
Version	1.0
Description	This API allows you to retrieve information about the work regulations of Kuroco RAG Demo Corporation.
Order	0

info

These settings will be used in ChatGPT later, so please write descriptions that explain what this API is for.

Security Settings

Next, configure the security settings.
Click on [Security] in the API page.

Set the security to [Static Access Token] and click on [Save].

CORS Configuration

Next, configure CORS settings.
Click on [Operation CORS].

Click on [Add Origin] for CORS_ALLOW_ORIGINS and add the following:

• Admin panel URL
• https://chatgpt.com/

Click on [Add Method] for CORS_ALLOW_METHODS and add the following:

• GET

Click on [Add Header] for CORS_ALLOW_HEADERS and add the following:

• x-rcms-api-access-token

Enable [Allow Credentials] in CORS_ALLOW_CREDENTIALS

Verify that the checkbox for [Allow Credentials] is checked.

If everything is correct, click on [Save].

Issuing a Static Access Token

Click on [Swagger UI].

Set the expiration date for the static access token and click on [Generate].

Once the static access token is issued, make a note of the value.

Creating an Endpoint

Click on [Add a new endpoint] from the RAG demo API.

Create the following endpoint:

Field	Setting
Path	search
Category	Content
Model	Topics
Operation	list
topics_group_id	ID of the content structure created earlier
cnt	Set to 3 to limit the number of entries returned in the API response
required_param	vector_search

Once configured, click on [Add] to add the endpoint.

Exporting the OpenAPI Configuration File

Kuroco has the functionality to export OpenAPI, allowing you to output API information in a format compliant with OpenAPI specifications.

From the endpoint list screen, click on [Export OpenAPI].

A popup for the OpenAPI schema file download will appear.
Select the format as YAML, select the desired endpoints, and click [Export].

The downloaded YAML file contains all the settings for the endpoint URL and available parameters.
We will need to make some adjustments so that we are only allowing the vector_search parameter for ChatGPT by remove all parameters except vector_search from the parameters section.

Also, set the required field for vector_search to true.

The YAML file will look like the following example.

openapi: 3.1.0
info:
  title: 'RAG demo'
  version: '1.0'
  description: 'This API allows you to retrieve information about the work regulations of Kuroco RAG Demo Corporation.'
servers:
  -
    url: 'https://your-domain-here.g.kuroco.app'
    description: 'API Backend'
paths:
  /rcms-api/8/search:
    get:
      tags:
        - Content
      summary: ''
      description: |
        
        ### **Topics::list (v1)**
        
        
        ## Controller parameters
        
        > **topics_group_id** `27`
        
        > **cnt** `3`
        
        > **required_param** `vector_search`
        
      parameters:
        -
          name: vector_search
          schema:
            type: string
            format: ''
          in: query
          required: true
          style: form
          explode: true
          description: 'Vector search'
      responses:
        '200':
          description: 'Topics data successfully fetched'
        '404':
          description: 'Topics data could not be found'
      security:
        -
          Token-Auth: []
      operationId: getRcmsApi8Search
components:
  schemas: {  }
  securitySchemes:
    Token-Auth:
      type: apiKey
      in: header
      name: X-RCMS-API-ACCESS-TOKEN

Once this is done, the preparation on the Kuroco side is complete.

ChatGPT Configuration

Creating GPT

Access https://chat.openai.com/ and log in.

Click on [Explore GPTs] in the sidebar, then click on [+ Create] in the top right corner of the page.

The GPT creation screen will open, configure it as follows in the [Configure] tab:

Field	Value
Name	Kuroco RAG Demo Co., Ltd. Internal Rules Assistant
Description	Answers questions based on the employment regulations of Kuroco RAG Demo Co., Ltd.
Instructions	To provide efficient and accurate responses, follow the guidelines below when interacting with users: 1. Always send a request to the specified endpoint before answering. 2. Guessing answers based on information not obtained from the endpoint is prohibited. 3. Do not answer questions that are deemed irrelevant to the business.

Scroll down to the bottom of the screen and click on [Create new action].

Paste the content of the modified openapi.yaml file that was downloaded from Kuroco earlier, then click on the gear icon next to Authentication.

Configure as follows and click on [Save]:

Field	Value
Authentication Type	API Key
API Key	Static access token issued by Kuroco
Authentication Type	Custom
Custom Header Name	X-RCMS-API-ACCESS-TOKEN

Click on [Test].

GPT will attempt to communicate with Kuroco, click on [Allow].
Verify that the data is retrieved to complete this configuration.

Click [Create] and specify the publish settings to complete the creation of the GPT.

Verification of Operation

When you access the GPT you have created and ask a question, you can see that it will answer by referring to the content you have stored in Kuroco.
If the answer is not as intended, you can update and adjust the instructions for the GPT.

With this, the creation of GPT using Kuroco RAG is now complete.
Kuroco RAG plays a role in facilitating the management and coordination of information provided to GPT, while GPT enhances the accuracy of the generated AI by referencing external content.