OpenSearch:Rich text feature

How it works

During indexing, the RAG pipeline extracts rich text elements (images, tables, and code) from each document and saves them as rich text tags alongside the corresponding text chunks. When a query is submitted through the SearchKnowledge API:

1. Open Search retrieves the relevant chunks, including any embedded rich text tags.

2. The LLM generates an answer using the retrieved chunks. Rich text tags are preserved in the answer.

3. Depending on the rich_text_strategy value you set, Open Search either renders the tags inline (Markdown) or returns the raw content separately for your app to render.

Enable the rich text feature

Set the rich_text_strategy parameter on the Other Parameters tab of the Q&A Test page. Leave the field blank to disable rich text.

The rich_text_strategy drop-down list provides two options: inside_response (embeds rich text content inline in the response body) and extend_response (returns rich text content as an extension).

The data array in the response may include items with "type": "IMAGE". These are image results from the retrieval phase—not output from the rich text feature. Rich-text-generated images appear in data.answer (for inside_response) or in data.rich_text_ref (for extend_response), and have better contextual relevance than retrieval-phase images.

Choose a strategy

The rich_text_strategy parameter accepts two values. Choose based on how your app renders content:

Configure the parameter

Add rich_text_strategy to the options.chat object in your SearchKnowledge API request:

{
  "question": {
    "text": "How do I configure a sensitive data identification rule?",
    "type": "TEXT"
  },
  "options": {
    "chat": {
      "rich_text_strategy": "inside_response"
    }
  }
}

Parse the response