Research summary on VectorDB

[yuki-2025]

Sharing the Research about VectorData

1. 65 Types of Vector Databases on the Market

Currently, not all information has been filled in. We welcome your help to provide more details.

Download or Edit Online: (https://docs.google.com/spreadsheets/d/18iQl6hUkXny7daCj3xO7u_OmWW2yZJtg/edit?usp=sharing&ouid=105396890618967809956&rtpof=true&sd=true)

2. Research Summary on Top 20 Vector Database

Download or Edit Online (中文/English) : https://docs.google.com/spreadsheets/d/12Q9LRGrlLuyZC52m9nxNghT58Jdyzigv/edit?usp=sharing&ouid=105396890618967809956&rtpof=true&sd=true

<style> </style>

Vector Database	Features (Including Platform and Developers)	Advantages	Application Scenarios	Launch Year	Stability	Ease of Use	Performance	Performance Rating	Cloud Expansion	Open Source
Weaviate	Graph-based storage system developed by SeMI Technologies. Supports semantic search and automatic classification.	Supports graph data structure for complex queries and relationships. Strong natural language understanding (NLU) capabilities for exceptional semantic search performance.	Knowledge graphs, semantic search.	2019	Mature with several production-level customers	Comprehensive documentation, good community support	High QPS support depending on configuration and optimization	8/10	Yes	Yes
Pinecone	Highly scalable vector search service provided by Pinecone Systems Inc., focused on supporting vector and metadata search.	Highly scalable service design, sub-second level vector search, especially suitable for applications requiring precise similarity matching and ranking.	Recommendation systems, personalized search.	2021	Emerging and rapidly developing	User-friendly, clear documentation	Designed for high performance, strong scalability	9/10	Yes	No
Pgvector on PostgreSQL	PostgreSQL extension supported by Supabase, allows vector search on the widely used database system.	As an extension of PostgreSQL, enables existing PostgreSQL users to easily integrate vector search, seamlessly combining relational data and vector search.	Integration into existing database applications.	2021	Newer, for PostgreSQL users	Integrates with PostgreSQL, moderate documentation	Performance dependent on PostgreSQL configuration	6/10	Yes (via cloud-provided PostgreSQL services)	Yes
Milvus	Open-source vector database developed by Zilliz, scalable, supports massive vector searches.	Open-source and scalable, specially optimized for real-time vector search of massive data, supports efficient retrieval of hundreds of millions of vectors, suitable for bioinformatics, image retrieval, etc.	Bioinformatics, image retrieval.	2019	Mature, active community	Easy to learn, rich documentation	High performance, optimized for real-time search of massive data	9/10	Yes	Yes
Qdrant	Open-source database supporting vector and metadata filtering, developed by the Qdrant team.	Offers filtering queries for vectors and metadata, supports complex condition combinations, widely applied in multimodal search and recommendation systems.	Multimodal search, recommendation systems.	2020	Newer but rapidly developing	Friendly API, documentation in progress	Performance optimized, supports high concurrency queries	8/10	Possible (via Kubernetes, etc.)	Yes
Chroma	Vector database focused on image search, specific development team information not detailed.	Focuses on vectorization and search of image content, optimized for image similarity search performance, suitable for visual content retrieval and management systems.	Image retrieval, visual search.	Not detailed	Information not detailed	Information not detailed	Optimization for image search might improve QPS in specific scenarios	7/10
Kdb.ai	High-performance time-series database KDB+ extension by Kx Systems, integrates vector search.	Efficient combination of time-series data analysis and vector search.	Financial services, real-time monitoring.	Not detailed	KDB+ is a long-standing stable product	Friendly for KDB+ users, detailed documentation	Excellent performance in time-series data	8/10	Possible (needs self-configured cloud deployment)	No (base KDB+)
Elasticsearch	Full-text search engine, supports basic vector search, developed by Elastic N.V.	Mature search engine based on Lucene, provides powerful full-text search and data analysis capabilities, widely used in log analysis, full-text search, etc.	Log analysis, full-text search.	2010	Very mature, widely used in production	Very comprehensive documentation, strong community support	Mature and stable, but may not be as effective in pure vector search as specialized vector databases	7/10	Yes	Yes
Hyperspace	Index system extension for Apache Spark, developed by Microsoft.	Provides indexing for Apache Spark, significantly improving Spark's data processing and query performance, suitable for big data analysis and processing.	Data lake acceleration, big data analysis.	2020	Emerging, actively being developed	Requires Apache Spark knowledge, documentation gradually improving	Accelerates Spark queries, performance gains depend on specific use cases	7/10	Possible (needs self-configured cloud deployment)	Yes
Datastax Astra/Apache Cassandra	Distributed NoSQL database, Apache Cassandra supported by Apache Software Foundation, DataStax Astra is a cloud service based on Cassandra.	High availability and scalability, suitable for scenarios requiring large-scale data distributed storage and access, like online applications, big data platforms, etc.	Large-scale data management, real-time applications.	2008 (Cassandra)	Very mature, widely used	Learning curve for new users, comprehensive documentation	High scalability and fault tolerance	7/10	Yes	Yes (Apache Cassandra)
Vectara	Serverless, AI-driven search service provided by Vectara.	Provides a serverless AI-driven search service, simplifying deployment and management of search systems, making it convenient in enterprise search and content discovery.	Enterprise search, content discovery.	2021	Emerging, with growth potential	User-friendly interface, documentation in progress	Serverless architecture may provide high performance	8/10	Yes	?
Cratedb	Distributed SQL database, supports spatio-temporal data, developed by Crate.io.	Optimized for IoT and big data real-time analytics processing, supports SQL queries, suitable for applications needing fast data analysis and processing.	IoT, real-time analysis.	2014	Mature, focused on real-time analytics	Easy to learn, comprehensive documentation	Optimized for real-time analysis of IoT and big data	7/10	Yes	Yes
TimescaleDB	A PostgreSQL extension for time-series data, developed by TimescaleDB, Inc.	Extends PostgreSQL to optimize the processing of time-series data, offers efficient data compression and partitioning, suitable for monitoring, IoT, etc.	Time series analysis, monitoring.	2017	Relatively mature, focused on time-series data	PostgreSQL compatible, easy to migrate and learn	Simplifies the management and analysis of time-series data	10-Jul	Yes (via cloud providers)	Yes
Deep Lake	A data lake launched by Databricks, optimized for AI workloads, based on Apache Spark and Delta Lake technologies.	Designed for large-scale data sets, optimized for AI and machine learning workloads, supports complex data analysis and processing.	Machine learning, data lakes.	2021	Backed by Databricks for confidence	Requires Databricks platform knowledge, detailed documentation	Efficient big data processing and AI workloads	10-Aug	Yes (as part of Databricks)	No
Vespa	Large-scale real-time computing engine developed by Yahoo, supports dynamic data's real-time indexing and querying.	Supports large-scale data's real-time indexing and querying, and instant deployment of machine learning models, suitable for dynamic content recommendation and advertising systems.	Large-scale recommendation systems, real-time decisions.	2017 (open source)	Long-term internal use by Yahoo	Learning curve present but comprehensive documentation	Designed for real-time processing of large-scale data	10-Sep	Yes (self-configuration required)	Yes
Faiss	A high-efficiency similarity search library developed by Facebook AI Research, designed for deep learning applications.	Developed by Facebook AI Research, optimized for efficient similarity search in large datasets, especially suitable for deep learning applications.	Similarity item retrieval, deep learning.	2017	Endorsed by Facebook AI Research, good stability	Detailed API, supports multiple languages	Excellent performance in similarity search libraries	10-Sep	Possible (self-configured cloud deployment required)	Yes
Marqo	An open-source vector database for tensor search, developed by Marqo.	Simplifies the indexing and searching process of deep learning models, supports searching of various data types, making AI model and data retrieval more accessible and efficient.	Semantic search, document retrieval.	2022	Newer, growing community	Documentation improving, friendly API	Optimized for deep learning model search	10-Aug	Yes (can be deployed in cloud environments)	Yes
Google Vector Search	Vector search service provided by Google Cloud, utilizing Google's AI technology.	Utilizes Google Cloud infrastructure to provide high-performance vector search services, integrating advanced AI features, suitable for scenarios requiring large-scale, high-performance search solutions.	Enterprise-level search, multimedia content retrieval.	2021	Google's backing provides high confidence	Integrated with Google Cloud services, comprehensive documentation	Utilizes Google infrastructure for excellent performance	10-Sep	Yes	No
Neo4jvector	Vector search extension for Neo4j, developed by Neo4j.	Combines the capabilities of Neo4j graph database with vector search functionality, enabling the use of vector similarity in graph queries, suitable for applications requiring both graph data and vector search.	Vector search in graph databases	Not detailed	Emerging, focused on graph database extension	Requires understanding of Neo4j, documentation improving	Combines the advantages of graph databases and vector search, performance rating depends on specific use cases	10-Aug	Possible (self-configured cloud deployment required)	Yes (base Neo4j)
Azure Cognitive Search	Search service on the Microsoft Azure platform	Integrates various Azure AI tools, offering rich data processing and search capabilities, suitable for scenarios requiring intelligent search solutions.	AI-enhanced search	Not detailed	Mature, Microsoft Azure product	Easy to use, integrates with Azure services	Integrates various Azure AI tools, enhancing search intelligence, excellent performance	10-Aug	Yes	No
Baidu Vector DB	Developed by Baidu, designed for AI applications	A high-performance vector retrieval system designed for AI and machine learning applications, supports efficient processing of large-scale data, suitable for AI application scenarios.	Vector retrieval	Not detailed	Baidu background, high stability expected	Requires adaptation to Baidu ecosystem, documentation primarily in Chinese	High performance, suitable for large-scale AI scenarios, performance depends on specific use cases	10-Aug	Yes	No
Redis	Open-source, supports multiple data structures, maintained by Redis Labs	Offers multiple data structure support and high-speed caching, suitable as a real-time application's data storage and processing backend.	High-performance key-value storage	2009	Very mature, widely used	Easy to use, rich documentation	Fast read/write, supports a rich set of data types, exceptionally high performance	10-Sep	Yes (via cloud providers)	Yes
DynamoDB	Fully managed database service provided by Amazon Web Services	Offers a highly scalable and fully managed NoSQL database solution, suitable for applications requiring high availability and large-scale data processing.	NoSQL database service	2012	Very mature, widely used	AWS integration, comprehensive documentation	High scalability, supports key-value and document data, high performance	10-Sep	Yes	No

Reference:
https://lakefs.io/blog/12-vector-databases-2023/
https://archive.is/cPx8g
https://superlinked.com/vector-db-comparison/
https://www.vecdbs.com/
https://milvus.io/docs/index.md

https://milvus.io/docs/v2.2.x/install-pymilvus.md