Elasticsearch iconElasticsearch

Distributed search engine with native vector search, combining dense, sparse and BM25 keyword retrieval in one hybrid query

78k+ GitHub Stars
AGPL Open Source Option (2024)
2010 First Release

Overview

Elasticsearch is the widely used distributed search engine from Elastic, and it is also a full vector database. It stores dense vectors with HNSW indexing, sparse vectors (including Elastic's ELSER model), and traditional inverted indexes, so a single query can blend semantic similarity, BM25 keyword relevance, filters, geo and reranking. Quantization options such as Better Binary Quantization shrink vector indexes substantially, and a semantic_text field type handles chunking and embedding for you.

Elastic moved Elasticsearch off Apache 2.0 in 2021, then added AGPLv3 as an open-source license option in August 2024 alongside SSPL and the Elastic License. You can self-manage it, use Elastic Cloud (hosted or serverless), or run it through cloud marketplaces. Also from Elastic: Jina AI embeddings.

The Verdict

Who Should Use Elasticsearch for Vectors?

Best For

  • Hybrid search that mixes keywords, filters and semantics
  • Teams already running Elastic for search or logs
  • Enterprise search and RAG at large scale
  • Ecommerce and site search with relevance tuning

Not Ideal For

  • Small projects that only need simple vector lookup
  • Tight budgets: clusters are resource-hungry
  • Teams without search engineering experience

What's Great

  • Best-in-class hybrid search: BM25, dense and sparse vectors together
  • Mature, proven at very large scale
  • Quantization (BBQ) cuts vector memory dramatically
  • Powerful query DSL, aggregations and reranking
  • Huge ecosystem and integrations (LangChain, LlamaIndex)

Watch Out For

  • Cost is the most common complaint at scale
  • Steep learning curve for mappings, tuning and operations
  • Resource-intensive compared with focused vector databases
  • License history (Apache, then SSPL, now AGPL option) worries some teams

Pricing

View all features & details

Vector Features

  • Dense vectors with HNSW
  • Sparse vectors and ELSER
  • Hybrid queries with reciprocal rank fusion
  • Better Binary Quantization (BBQ)
  • semantic_text field with automatic embedding
  • Learning to Rank and rerankers

Integrations

  • LangChain, LlamaIndex
  • Inference API for OpenAI, Cohere, Hugging Face and others
  • AWS, Azure and Google Cloud marketplaces
  • Kibana for exploration

How It Compares

Feature Elasticsearch OpenSearch Vespa Pinecone
Open source AGPL option Apache 2.0 Apache 2.0 No
Hybrid keyword plus vector Yes Yes Yes Sparse vectors
Self-hosting Yes Yes Yes No
Managed option Elastic Cloud AWS Vespa Cloud Fully managed
Operational effort High High High Low

Is this your tool? Add a badge