Skip to main content

​
Create an Index for Agentic Retrieval

Indexes for agentic retrieval require specific configurations to support multi-query patterns and semantic reranking.

​
Requirements

1

Semantic Configuration

Required for semantic reranking of subquery results
2

Vector Fields

Recommended for hybrid search across text and embeddings
3

Filterable Fields

Optional but useful for pre-filtering results
4

Default Scoring Profile

Optional for custom relevance tuning

​
Index Schema Example

{
  "name": "products-index",
  "fields": [
    {
      "name": "productId",
      "type": "Edm.String",
      "key": true
    },
    {
      "name": "description",
      "type": "Edm.String",
      "searchable": true
    },
    {
      "name": "descriptionVector",
      "type": "Collection(Edm.Single)",
      "searchable": true,
      "dimensions": 1536,
      "vectorSearchProfile": "my-vector-profile"
    }
  ],
  "semantic": {
    "defaultConfiguration": "my-semantic-config",
    "configurations": [
      {
        "name": "my-semantic-config",
        "prioritizedFields": {
          "titleField": {"fieldName": "title"},
          "contentFields": [
            {"fieldName": "description"}
          ]
        }
      }
    ]
  }
}

​
Best Practices

  • Use descriptive field names
  • Include both text and vector fields for hybrid search
  • Configure semantic ranking on most important fields
  • Add filters for common query patterns

​
Next Steps

Create Knowledge Source

Wrap your index in a knowledge source

Vector Search

Add vector capabilities to your index

Build docs developers (and LLMs) love

Get started for freeTalk to us