Elasticsearch

Elasticsearch is Elastic's open source search engine and vector database for developers who need keyword, vector and hybrid search, self-hosted or managed.

Elasticsearch indexes documents across a cluster of nodes and answers keyword queries, vector similarity queries or both at once, in near real time. Its developer is Elastic, a company listed on the New York Stock Exchange as ESTC, which also sells Kibana, Elastic Observability and Elastic Security. The engine holds documents, embeddings and search logic in one place, so a team does not have to keep a separate vector store beside its search index. Because it is open source, it can run on a company's own servers or as a managed service on Elastic Cloud. Vectors live in a dense_vector field indexed with HNSW, and the int8, int4 and BBQ quantization options shrink that index in memory at some cost in accuracy. Similarity can be cosine, dot product, L2 norm or maximum inner product. A semantic_text field sends its text to a configured inference endpoint as it is indexed, so embeddings arrive without a separate pipeline. Hybrid search mixes keyword matching, nearest neighbor vectors and sparse learned retrieval in one workflow, and semantic reranking with Elastic Rerank reorders the top results. Underneath, primary shards spread an index over nodes, replica shards take over when a node goes down, and snapshot and restore copies an index to another cluster. Developers adding search to an application, teams building retrieval-augmented generation on their own content and platform teams that want search and analytics in one engine are its core users. Operations and security teams also rely on it for log analytics and security analytics. A team that wants no cluster to manage starts a Serverless project on Elastic Cloud. A team that wants to choose its deployment runs Elastic Cloud Hosted, and one bound to its own infrastructure runs Elasticsearch self-managed. On a product or documentation site, Elasticsearch answers the search box with hybrid queries and reranks the best hits before they are shown. In an AI assistant, it retrieves the passages a language model answers from, after embedding them on ingest through semantic_text. Log and event data usually reach it through Logstash, Beats or Elastic Agent, and analysts explore that data in Kibana. New documents become searchable almost as soon as they are written, so the same index can serve a live search box and fresh operational data.