RAG Conversation

RAG Conversation allows you to search for information within your documents based on a query, maintaining the conversation memory and taking into account the historical context, in order to provide always relevant and contextualized answers. The service combines semantic search and natural language output generation, returning responses within a dialogue that can be resumed using the session ID, ensuring continuity and coherence across interactions. All this is done securely, without the risk of sensitive data leaks.

The RAG-as-a-Service technology by Openapi therefore allows companies to query their proprietary data and obtain reliable, contextualized answers. The service reduces the risk of "hallucinations" typical of Large Language Models (LLMs), and provides answers that are consistent and relevant to the organization's specific knowledge.

With RAG-as-a-Service, integrating RAG functionalities into applications is simple and scalable, without the need to manage complex data systems. Your content is automatically indexed, allowing the model to access up-to-date and relevant data in real time. In this way, Openapi offers the combination of artificial intelligence and contextual search, providing accurate answers based solely on your company knowledge.

Openapi also offers two other RAG services for document search: Search, which returns one or more text extracts, and Conversation, which provides single natural language answers not embedded in a conversation. All three services are valid globally.

To obtain consistent and relevant results with RAG Conversation, three main steps must be followed:

Thanks to indexing, it is possible to quickly retrieve the most relevant information in response to a query. Every time the RAG is modified (document uploaded or removed), it is essential to restart indexing to ensure updated and accurate responses.

Create a New RAG

The first step is to create an empty RAG via the POST /rag endpoint, where documents can be added later.

REQUEST EXAMPLE

The request should specify:

{
  "name": "RAGTest",
  "options": {
    "ocr": false,
    "autoIndexingIntervalHours": 0
  },
  "callback": {
    "method": "POST",
    "field": "string",
    "url": "https://www.mysite.it/callback.php",
    "data": {}
  }
}

RESPONSE EXAMPLE

The response contains RAG details (e.g., ID and name, state, options, settings such as OCR, total and indexed documents, creation and update dates, last indexing date).

{
  "data": {
    "id": "68947e1e0dab7e743108abfb",
    "name": "RAGTest",
    "state": "ready",
    "options": {
      "ocr": false,
      "autoIndexingIntervalHours": 0
    },
    "totalDocuments": 0,
    "totalIndexedDocuments": 0,
    "sizeMegaByte": 0,
    "createdAt": "2025-09-04T10:45:42.036Z",
    "updatedAt": "2025-09-04T10:45:42.036Z",
    "indexedAt": "2025-09-04T10:45:42.036Z",
    "indexingRunAt": "2025-09-04T10:45:42.036Z",
    "callback": {
      "method": "POST",
      "field": "string",
      "url": "https://www.mysite.it/callback.php",
      "data": {}
    }
  },
  "success": true,
  "message": "",
  "error": null
}

You can retrieve the list of all RAGs or the details of a single RAG at any time using the GET /rag and GET /rag/{id} endpoints.

Upload a Document

Once the RAG is created, you can proceed with adding a document to it through the POST /rag/{id}/documents endpoint.

REQUEST EXAMPLE

The request must specify the ID of the RAG where the document will be uploaded.

The payload must include:

{
  "fileName": "documentTest",
  "fileContent": "string",
  "metadata": {
    "custom_key_str": "string",
    "custom_key_date": "2025-09-04T10:51:06.259Z",
    "custom_key_int": 0,
    "custom_key_geo_latitude": 0,
    "custom_key_geo_longitude": 0,
    "custom_key_bool": true
  }
}

RESPONSE EXAMPLE

The response provides document details including ID, name, state, metadata, size, file type, creation and update dates.

{
  "data": [
    {
      "id": "string",
      "name": "string",
      "state": "string",
      "metadata": {
        "custom_key_str": "string",
        "custom_key_date": "2025-09-04T10:51:06.260Z",
        "custom_key_int": 0,
        "custom_key_geo_latitude": 0,
        "custom_key_geo_longitude": 0,
        "custom_key_bool": true
      },
      "sizeMegaByte": 0,
      "mimeType": "string",
      "createdAt": "2025-09-04T10:51:06.260Z",
      "updatedAt": "2025-09-04T10:51:06.260Z",
      "paymentDueDate": "2025-09-04T10:51:06.260Z"
    }
  ],
  "success": true,
  "message": "",
  "error": null
}

You can retrieve the list of documents within a RAG at any time via the GET /rag/{id}/documents endpoint.

Start Indexing Process

As mentioned above, to obtain contextual and accurate answers, it is essential to start indexing every time a RAG is modified (i.e., a document is uploaded or deleted). If this step is not performed, the system will continue to rely only on previous data and not on the updated information.

Indexing can be started manually via PATCH/rag/{id}; alternatively, automatic indexing can be set up, which runs every X hours (as defined by the user). The latter is configured when creating the RAG, via the POST /rag endpoint. Indexing, whether automatic or manual, does not incur additional costs.

REQUEST EXAMPLE

The request must specify the ID of the RAG for which the manual indexing process is being started.

The request payload must include the startIndexing parameter set to true to start the indexing process:

"startIndexing": true

RESPONSE EXAMPLE

The response returns details about the RAG name, state (e.g., ready, indexing), settings (e.g., OCR yes/no, automatic indexing yes/no), number of documents present and indexed, current RAG size in MB, and creation and update dates.

{
  "data": {
    "id": "68947e1e0dab7e743108abfb",
    "name": "RAGTest",
    "state": "indexing",
    "options": {
      "ocr": false,
      "autoIndexingIntervalHours": 0
    },
    "totalDocuments": 0,
    "totalIndexedDocuments": 0,
    "sizeMegaByte": 0,
    "createdAt": "2025-09-04T10:55:37.306Z",
    "updatedAt": "2025-09-04T10:55:37.306Z",
    "callback": {
      "method": "POST",
      "field": "string",
      "url": "https://www.mysite.it/callback.php",
      "data": {}
    }
  },
  "success": true,
  "message": "",
  "error": null
}

RAG CONVERSATION

After completing the three steps mentioned above, you can proceed with the actual search through the POST /rag/conversation endpoint.

As stated at the beginning, RAG Conversation allows—via API and in real time—to find the most relevant document or information within your files, based on a query and taking previous conversations into account. All this is done securely, without the risk of sensitive data leaks.

REQUEST EXAMPLE

The request should specify:

{
  "ragId": "688b2ab3bccc6a603202f87c",
  "query": "what services are available?",
  "max_len": 5,
  "metadata_filter": "(custom_key_int<175 AND custom_key_bool = \"true\") OR (custom_key_int<125 AND custom_key_bool = \"false\")",
  "settings": {
    "prompt_style": "basic"
  },
  "session": "string"
}

RESPONSE EXAMPLE

RAG Conversation returns information about the most relevant documents for your search, including text extracts, which help quickly locate the desired data.

The response payload specifically provides:

{
  "data": [
    {
      "answer": "string",
      "search_result": [
        {
          "index": 0,
          "snippets": [
            "string"
          ],
          "id": "689c70313f292734010a7dda",
          "ragId": "689c6ed7b283ac0aa40c1555",
          "name": "documentTest",
          "state": "indexing",
          "metadata": {
            "custom_key_str": "string",
            "custom_key_date": "2025-09-25T10:34:00.005Z",
            "custom_key_int": 0,
            "custom_key_geo_latitude": 0,
            "custom_key_geo_longitude": 0,
            "custom_key_bool": true
          }
        }
      ],
      "session": "projects/702697495064/locations/global/collections/default_collection/RAG/689c6ed7b283ac0aa40c1555/sessions/14709352892326605"
    }
  ],
  "success": true,
  "message": "",
  "error": null
}

With RAG Conversation, you can start continuous conversations with the system and quickly query your documents, ensuring responses are as accurate as possible and based solely on the data actually uploaded and indexed. All information remains fully secure.

The service is particularly useful for companies managing large amounts of documents (contracts, reports, manuals), e.g.:

In general, RAG Conversation is perfect for any organization or professional needing fast access to accurate information within their documents, optimizing time and search reliability.