Skip to main content
A payload is a JSON object that stores metadata for a point in VectorAI DB. Use payloads to attach contextual information such as categories, prices, timestamps, and custom attributes relevant to your app. While vectors represent the semantic meaning of your data, payloads store the descriptive attributes that make your data actionable. For example, a product image vector might have a payload storing product name, category, price, availability status, and rating. Hybrid search combines vector similarity with metadata filtering through payload fields. With payloads, you can build queries like:
  • Find similar products under $100 that are in stock.
  • Search for related articles published after 2024.
  • Retrieve images tagged with specific keywords.

​
Payload structure

Each payload is a JSON object. VectorAI DB supports standard data types, including strings, numbers, booleans, arrays, nested objects, and null values.

​
Strings

Store text values like names, categories, and descriptions.

​
Numbers

Store integers and floating-point values for prices, quantities, and scores.

​
Booleans

Store true or false flags for status fields like availability and feature toggles.

​
Arrays

Store lists of values such as tags, labels, or related identifiers.

​
Nested objects

Store structured data by nesting objects within payload fields.

​
Null values

Set a field to null when the value is unknown or not yet available.
Payloads do not exist on their own. You attach them to points, alongside the point ID and vector.

​
Common use cases

The following examples show how payloads support common application patterns. Attach product attributes like category, price, brand, rating, and availability. Filter by price ranges, brands, or availability while finding similar products based on descriptions or images.

​
Document retrieval

Store document metadata like title, author, publication date, and tags. Search for semantically related content within specific categories or time periods to build knowledge bases or search engines. Include image metadata like filename, location, timestamp, and tags. Find visually similar images that match specific criteria like location or date for photo management or visual search apps.

​
Content recommendation

Track user preferences, interaction history, and content attributes. Recommend similar items that match user preferences or constraints for personalized content delivery.

​
Task guides

The following guides cover payload operations step by step.

​
Next steps

  • Filtering — Build must, should, and must-not queries against payload fields.
  • Points — Understand the point structure that payloads belong to.
  • Search operations — Combine payload filters with similarity search.
  • Vectors — Understand the embeddings paired with payloads.