Vector and Text Search¶

Vector and Text Search in ScyllaDB¶

ScyllaDB’s Vector and Text Search feature combines two complementary ways of searching your data: vector search, which retrieves data by similarity, and full text search, which retrieves data by matching terms in text.

Vector search is a powerful method for efficiently searching and retrieving high-dimensional data based on similarity rather than exact matches. It is particularly useful in AI and machine learning applications, where data is often represented as vectors — mathematical representations of objects such as text, images, audio, or video. In these applications, you typically need to retrieve data that is similar to a given query, rather than relying on keyword-based search or exact matches.

Full text search complements this by matching and ranking documents against the terms in a query, which is what you want when the exact words matter. Combining both lets you build hybrid search, where semantic similarity and term matching are used together to improve relevance.

Vector and Text Search allows you to store, index, and query high-dimensional vector data at scale. It is built to work within your existing ScyllaDB infrastructure, taking advantage of its high-performance and highly available architecture.

Getting Started¶

• Quick Start Guide

  Walk through setting up a vector-enabled table, inserting data, and running your first similarity search in minutes.

Understanding Vector and Text Search¶

• Vector and Text Search Concepts

  Architecture overview, HNSW algorithm, CDC-based indexing, and data flow between storage and indexing nodes.
• Glossary

  Definitions of key terms including ANN, HNSW, embeddings, similarity functions, quantization, filtering, and more.

Deployment and Operations¶

• Vector and Text Search Deployments

  Create, enable, resize, disable, and monitor Vector and Text Search clusters in ScyllaDB Cloud via the UI or API.
• Sizing and Capacity Planning

  Estimate memory requirements, understand quantization impact, and choose instance types for your workload.
• Security

  Authentication, authorization, service-level isolation, and network security for vector search.

Working with Vectors and Text¶

• Working with Vector and Text Search

  CQL usage guide covering the vector data type, vector indexes, similarity functions, index tuning, ANN queries, and driver integration.
• Working with Full Text Search

  CQL usage guide for full text search: creating a full-text index, the BM25() query shape, boolean and phrase queries, relevance ranking, and combining text search with vector search.
• Vector Search with Alternator

  Use Vector Search through the DynamoDB-compatible Alternator API, with Python examples and a feature-availability comparison with CQL.
• Filtering Vector and Text Search Results

  Combine similarity search with metadata constraints using global and local vector indexes.
• Quantization and Rescoring

  Reduce index memory usage with quantization (f16, i8, b1) and recover precision with oversampling and rescoring.
• LangChain and CassIO Compatibility

  Use the LangChain Cassandra connector (CassIO) with ScyllaDB through Storage Attached Index (SAI) compatibility, including requirements, limitations, and a runnable RAG example.

Troubleshooting and Reference¶

• Troubleshooting

  Common issues and solutions for index creation, query results, data insertion, performance, and connectivity.
• FAQ

  Frequently asked questions about similarity functions, dimensions, latency, filtering, quantization, and more.
• Reference

  Technical reference for instance types, CQL syntax, index options, and Cloud API endpoints.

Examples¶

• Example Applications

  Learn how to use ScyllaDB vector search to build RAG applications, semantic caching layers, and how it integrates with popular LLM libraries like LlamaIndex and LangChain.
• LangChain and CassIO Compatibility

  Run the LangChain Cassandra connector against ScyllaDB, with a complete Retrieval-Augmented Generation (RAG) example.