Vector databases
A vector database stores embeddings, the numeric vectors that represent text, images, or other data, and returns the entries closest to a query vector using approximate nearest neighbor indexes. In retrieval-augmented generation, the vector database supplies the passages a model reads before it answers.
The options fall into four groups: purpose-built engines such as Pinecone, Weaviate, Qdrant, Milvus, and Chroma; embedded libraries such as LanceDB, Faiss, and sqlite-vec that run inside an application process; extensions to existing systems, such as pgvector for PostgreSQL, Redis, MongoDB, and Elasticsearch; and cloud services such as Amazon S3 Vectors, Azure AI Search, and Google Cloud Vector Search.
Most now combine vector similarity with keyword or sparse search and metadata filtering, because hybrid retrieval finds exact terms, such as product codes, that pure embedding search can miss. The right choice depends on where the data already lives, the expected corpus size, and how much infrastructure the team wants to run.
Vector databases compared
| Platform | Open source | Deployment | Best for |
|---|---|---|---|
| Amazon S3 Vectors | No | Managed cloud, API | Large, infrequently queried vector sets on AWS |
| Astra DB | No | Managed cloud, Self-hosted | Vector search on Cassandra-based infrastructure |
| Azure AI Search | No | Managed cloud, API | Retrieval for Azure and Microsoft Foundry applications |
| Chroma | Yes (Apache-2.0) | Library, Self-hosted, Managed cloud | Prototyping retrieval locally, then moving to a hosted service |
| Cloudflare Vectorize | No | Managed cloud, API | Retrieval inside Cloudflare Workers applications |
| Elasticsearch | Yes (AGPL-3.0-only OR SSPL-1.0 OR Elastic-2.0) | Self-hosted, Managed cloud | Adding vector search to existing Elasticsearch deployments |
| Faiss | Yes (MIT) | Library | Building custom vector indexes in-process |
| Infinity | Yes (Apache-2.0) | Self-hosted, Library | Hybrid retrieval with tensor and full-text ranking |
| LanceDB | Yes (Apache-2.0) | Library, Managed cloud | Embedded and multimodal retrieval on object storage |
| Meilisearch | Yes (MIT AND BUSL-1.1) | Self-hosted, Managed cloud | Developer-friendly hybrid search for apps and sites |
| Milvus | Yes (Apache-2.0) | Library, Self-hosted, Managed cloud | Large deployments that need horizontal scaling |
| MongoDB Vector Search | No | Managed cloud, Self-hosted | Vector search inside an existing MongoDB deployment |
| OpenSearch | Yes (Apache-2.0) | Self-hosted, Managed cloud | Apache-licensed search with vector and keyword retrieval |
| pgvector | Yes (PostgreSQL) | Self-hosted, Managed cloud | Teams already running PostgreSQL |
| pgvectorscale | Yes (PostgreSQL) | Self-hosted | Scaling pgvector workloads beyond memory |
| Pinecone | No | Managed cloud, API | Managed vector search without running infrastructure |
| Qdrant | Yes (Apache-2.0) | Self-hosted, Managed cloud | Filtered vector search with detailed performance tuning |
| Redis | Yes (AGPL-3.0-only OR SSPL-1.0 OR RSALv2) | Self-hosted, Managed cloud | Low-latency vector search alongside caching |
| SingleStore | No | Managed cloud, Self-hosted | Vector search combined with real-time SQL analytics |
| sqlite-vec | Yes (MIT OR Apache-2.0) | Library, Local | Local and on-device vector search with SQLite |
| Supabase Vector | Yes (Apache-2.0) | Managed cloud, Self-hosted | Apps already built on Supabase Postgres |
| turbopuffer | No | Managed cloud, API | Large multi-tenant indexes on object storage |
| Typesense | Yes (GPL-3.0) | Self-hosted, Managed cloud | Site and app search with added vector retrieval |
| Upstash Vector | No | Managed cloud, API | Serverless and edge apps that need a hosted index |
| Vald | Yes (Apache-2.0) | Self-hosted | Kubernetes-native vector search at scale |
| Vector Search (Gemini Enterprise Agent Platform) | No | Managed cloud, API | Teams building retrieval on Google Cloud |
| VectorChord | Yes (AGPL-3.0-only OR Elastic-2.0) | Self-hosted | Large vector indexes inside PostgreSQL |
| Vespa | Yes (Apache-2.0) | Self-hosted, Managed cloud | Large-scale search with custom ranking logic |
| Weaviate | Yes (BSD-3-Clause (enterprise code under the Weaviate License)) | Self-hosted, Managed cloud | Hybrid search with built-in embedding integrations |
Profiles

Amazon S3 Vectors
Amazon Web Services · aws.amazon.com
Amazon S3 Vectors adds vector buckets and vector indexes to Amazon S3 for storing and querying embeddings without provisioning servers, and works as a vector store for Amazon Bedrock Knowledge Bases and alongside Amazon OpenSearch Service.

Astra DB
IBM (DataStax) · ibm.com
Astra DB is a serverless database built on Apache Cassandra that supports vector search alongside NoSQL data, offered by IBM with the Hyper-Converged Database for self-managed deployments and Langflow for building RAG applications.
IBM completed its acquisition of DataStax in May 2025, and Astra DB now sits within IBM's watsonx data portfolio.
Azure AI Search
Microsoft · learn.microsoft.com
Azure AI Search is Microsoft's managed search service for vector, keyword, and hybrid retrieval with semantic ranking and integrated vectorization, and it underpins Foundry IQ, the knowledge layer that agents in Microsoft Foundry use for agentic retrieval.

Chroma
trychroma.com
Chroma is open-source search infrastructure for AI that supports vector, full-text, regex, sparse, and metadata search, runs embedded in an application or on a team's own servers, and offers Chroma Cloud as a managed service.

Cloudflare Vectorize
Cloudflare · developers.cloudflare.com
Vectorize is Cloudflare's globally distributed vector database for applications built on Cloudflare Workers, storing embeddings with metadata filtering and pairing with Workers AI embedding models for retrieval close to users.
Elasticsearch
Elastic · elastic.co
Elasticsearch is a distributed search and analytics engine built on Apache Lucene that combines dense and sparse vector search with BM25 keyword search for hybrid retrieval, and runs self-managed or on Elastic Cloud with hosted Jina embedding models.

Faiss
Meta · faiss.ai
Faiss is an open-source library from Meta for similarity search and clustering of dense vectors, with CPU and GPU implementations of flat, IVF, HNSW, and product quantization indexes, and it serves as the index engine inside several vector databases.

Infinity
InfiniFlow · infiniflow.org
Infinity is an open-source database from InfiniFlow, the team behind RAGFlow, that combines dense vector, sparse vector, tensor, and full-text search with fused ranking such as reciprocal rank fusion for RAG workloads.

LanceDB
lancedb.com
LanceDB is an open-source embedded retrieval library built on the Lance columnar format that stores vectors, metadata, and multimodal data on local disk or object storage, with a managed multimodal lakehouse for distributed production search.
Meilisearch
meilisearch.com
Meilisearch is a search engine written in Rust that merges full-text and vector results in hybrid search through configurable embedders, with an MIT-licensed community edition, a BUSL-licensed enterprise edition, and Meilisearch Cloud.
Milvus
Zilliz · milvus.io
Milvus is an open-source vector database that runs as Milvus Lite inside a Python process, as a standalone server, or as a distributed cluster for billions of vectors, with Zilliz Cloud from the same team as its fully managed service.
MongoDB Vector Search
MongoDB · mongodb.com
MongoDB Vector Search adds a $vectorSearch aggregation stage for approximate and exact nearest neighbor queries over documents in MongoDB Atlas, and is also generally available in self-managed Community Edition through the SSPL-licensed mongot search process.
OpenSearch
opensearch.org
OpenSearch is an Apache-2.0 search and analytics suite run as a Linux Foundation project, with a k-NN plugin for approximate vector search, neural and hybrid search features, and a managed version in Amazon OpenSearch Service.

pgvector
github.com
pgvector is an open-source PostgreSQL extension that adds a vector type with exact and approximate nearest neighbor search through HNSW and IVFFlat indexes, and major managed PostgreSQL services support it, so embeddings live beside application data.

pgvectorscale
Tiger Data (formerly Timescale) · github.com
pgvectorscale is an open-source PostgreSQL extension from Tiger Data that complements pgvector with a StreamingDiskANN index, statistical binary quantization, and label-based filtering, aimed at vector workloads too large to keep fully in memory.
Pinecone
Pinecone Systems · pinecone.io
Pinecone is a managed vector database with serverless indexes, dense and sparse vectors, metadata filtering, and Dedicated Read Nodes, sold as part of a knowledge platform that also includes Pinecone Assistant and the Nexus knowledge engine for agents.
Qdrant
qdrant.tech
Qdrant is an open-source vector search engine written in Rust that supports payload filtering, dense and sparse vectors for hybrid search, multi-vector representations, and vector quantization, and runs self-hosted or on Qdrant Cloud.
Redis
redis.io
Redis supports vector search through the Redis Query Engine, with FLAT, HNSW, and SVS-VAMANA indexes over hashes or JSON documents, adds a vector set data type in Redis 8, and runs self-managed or on Redis Cloud.
SingleStore
singlestore.com
SingleStore is a distributed SQL database for transactions and analytics with a native vector type, indexed approximate nearest neighbor search, and full-text search, so vector queries can join relational data; it runs as the managed Helios service or self-managed.
sqlite-vec
alexgarcia.xyz
sqlite-vec is an open-source SQLite extension written in C that stores and queries float and binary vectors in virtual tables, and runs wherever SQLite does, including mobile devices, browsers through WebAssembly, and servers.

Supabase Vector
Supabase · supabase.com
Supabase Vector is the vector toolkit in Supabase's Postgres platform, using pgvector to store, index, and query embeddings next to application data, with client libraries and Edge Functions for generating embeddings and building AI features.

turbopuffer
turbopuffer.com
turbopuffer is a managed search engine that stores vector and BM25 full-text indexes on object storage with caching in front, organizing data into namespaces so applications can keep separate indexes per user or tenant.

Typesense
typesense.org
Typesense is an open-source, typo-tolerant search engine written in C++ that adds vector, semantic, and hybrid search, can generate embeddings with built-in or external models, and runs self-hosted or on Typesense Cloud.
Upstash Vector
Upstash · upstash.com
Upstash Vector is a serverless vector database for RAG and semantic search with built-in embedding models, hybrid search, metadata filtering, and namespaces, accessed over a REST API that also works from edge runtimes.

Vald
vald.vdaas.org
Vald is an open-source distributed vector search engine written in Go that runs on Kubernetes, uses the NGT approximate nearest neighbor library, and spreads indexes across agent pods with automatic indexing and index backups.

Vector Search (Gemini Enterprise Agent Platform)
Google Cloud · docs.cloud.google.com
Vector Search is Google Cloud's managed approximate nearest neighbor service, built on vector search technology from Google Research, that serves low-latency similarity queries at scale and plugs into the platform's RAG Engine as a vector store.
Formerly Vertex AI Vector Search; Google replaced the Vertex AI brand with Gemini Enterprise Agent Platform in April 2026.
VectorChord
TensorChord · vectorchord.ai
VectorChord is a PostgreSQL extension for disk-friendly vector search that succeeds pgvecto.rs, combines IVF indexing with RaBitQ quantization, and stays compatible with pgvector data types so existing schemas can move over.
Vespa
Vespa.ai · vespa.ai
Vespa is an open-source AI search platform that combines vector search, full-text search, structured filtering, and tensor-based ranking in one engine for search, recommendation, and RAG, and runs self-hosted or on Vespa Cloud.

Weaviate
weaviate.io
Weaviate is an open-source vector database written in Go that stores objects alongside their vectors, combines keyword and vector search in hybrid queries, and connects to embedding providers through modules, with Weaviate Cloud as the managed option.
How to choose
- 01Deployment
- A fully managed service, a self-hosted server or cluster, an embedded library, or an extension to a database you already operate.
- 02Search features
- Hybrid dense and sparse retrieval, full-text search, metadata filtering, multi-vector support, and built-in reranking.
- 03Scale and indexing
- Index types such as HNSW, IVF, or disk-based DiskANN, plus vector quantization, determine memory use, cost, and recall at large corpus sizes.
- 04Operations
- Backups, replication, multi-tenancy, and how inserts, updates, and deletes affect index freshness.
- 05License
- Permissive licenses such as Apache-2.0 or MIT, copyleft or source-available terms such as AGPL-3.0 or SSPL, or a proprietary managed service.
- 06Ecosystem
- Client SDKs and integrations with frameworks such as LangChain and LlamaIndex, and with embedding providers, reduce glue code.
Frequently asked questions
What is a vector database used for?
- A vector database is used to find items that are semantically similar to a query. Common uses include retrieval for RAG applications, semantic and hybrid search, recommendations, deduplication, and long-term memory for AI agents. It stores embeddings alongside metadata and uses approximate nearest neighbor indexes to return the closest matches quickly, even across millions of records.
Do I need a vector database for RAG?
- Not always. A small corpus can be searched in memory with a library such as Faiss, and teams already running PostgreSQL, MongoDB, or Elasticsearch can add vector search there. A dedicated vector database becomes useful when the corpus reaches millions of chunks, when you need hybrid search and filtering at low latency, or when retrieval must scale independently of the main application database.
Is Postgres a vector database?
- PostgreSQL is a relational database, but extensions turn it into a capable vector store. pgvector adds a vector type with exact search and HNSW and IVFFlat indexes, while pgvectorscale and VectorChord add disk-friendly indexes for larger workloads. Embeddings then sit in the same tables as application data and use SQL joins, transactions, and backups.
What is the difference between a vector database and a relational database?
- A relational database stores rows in tables and answers exact queries, such as all orders from a given customer, using indexes on column values. A vector database stores high-dimensional embeddings and answers similarity queries, such as the passages closest in meaning to a question. Many relational and document databases now include vector indexes, so the two roles often share one system.
What is the best open source vector database for RAG?
- No single open-source vector database fits every RAG system. Qdrant, Weaviate, and Milvus are standalone engines with hybrid search and horizontal scaling, Chroma and LanceDB suit embedded and local use, and pgvector fits teams already on PostgreSQL. Check the license as well: some projects are Apache-2.0 or MIT, while others use AGPL-3.0, SSPL, or source-available terms.
Related guides
RAG and vector search hubOther categories
- Model providers
- Inference providers
- GPU clouds
- Serving engines
- Embedding and reranking models
- Document parsing and data ingestion
- Fine-tuning and post-training
- Agent frameworks
- Agent infrastructure
- Web search APIs
- MCP platforms
- Voice AI
- AI gateways
- LLM observability and evaluation
- Guardrails and AI security
- AI coding tools
Product details change quickly; check each vendor’s site for current features, licensing and pricing.