Configuration Reference

Introduction

The configuration is a metadata object that lists all the database entities — such as collections and fields — that the data connector has to know about in order to serve queries. It never changes during the lifetime of the data connector service instance. When your database schema changes you will have to update the configuration accordingly, see updating with introspection.

Structure

The configuration object is a JSON object with the following fields:

{
  "config": {
    "collection_names": [],
    "object_fields": {},
    "object_types": {},
    "collection_vectors": {},
    "functions": [],
    "procedures": []
  }
}

Property: Collection Names

This is an array of collection (table) names that are included in the configuration. This is a required field.

Example:

{
  "collection_names": ["multivector_collection", "terraforming", "star_charts"]
}

Property: Object Fields

This is a JSON object that maps each collection name to its field names. These field names correspond to the attributes or columns available in each collection.

Example for a single collection:

{
  "multivector_collection": ["name", "type", "id", "score", "vector"]
}

Property: Object Types

This is a JSON object providing detailed type information for each field in every collection. It includes both required and nullable fields, using type descriptors.

Example:

{
  "multivector_collection": {
    "fields": {
      "id": {
        "type": {
          "type": "named",
          "name": "Int"
        }
      },
      "score": {
        "type": {
          "type": "nullable",
          "underlying_type": {
            "type": "named",
            "name": "Float"
          }
        }
      }
    }
  }
}

For nullable fields, the type is wrapped in a nullable descriptor:

Nullable field example:

{
  "vector": {
    "type": {
      "type": "nullable",
      "underlying_type": {
        "type": "array",
        "element_type": {
          "type": "named",
          "name": "Float"
        }
      }
    }
  }
}

Property: Collection Vectors

This field maps collection names to a boolean indicating whether the collection supports multi-vector search.

Example:

{
  "terraforming_plans": true,
  "multivector_collection": false
}

Property: Functions and Procedures

Arrays of function and procedure definitions.

{
  "functions": [],
  "procedures": []
}

Updating with introspection

Whenever the schema of your database changes you will need to update your data connector configuration accordingly to reflect those changes.

Running update in a configuration directory will do the following:

• Connect to the database and refresh the collection metadata
• Update the collection_names, object_fields, and object_types to reflect the current database schema
• Preserve any custom configuration that doesn't conflict with the updated schema