← Vector Databases for AI Agents

Elasticsearch

hybridFree (self-hosted, Elastic License v2 / SSPL)Elastic Cloud Hosted $99+/mo (Standard, Gold, Platinum, Enterprise)Elasticsearch Serverless (pay per use)open source
https://www.elastic.co/elasticsearch ↗View as JSON →

Best for

Hybrid search at scale on top of mature search infrastructure — combine BM25, ELSER sparse vectors, and dense vectors in one query

Limitations

Heavier and more memory-hungry than purpose-built vector engines; license is no longer pure Apache 2.0; vector-only workloads are overserved by Elasticsearch's full feature surface

Features

Hybrid Search
Combined dense vector + sparse/keyword (BM25) search in one query
Metadata Filtering
Pre/post filtering by structured metadata — JSON payloads, tags, dates
Multi Tenancy
Native namespace or tenant isolation for per-user or per-agent vector spaces
Managed Embeddings
Built-in embedding generation at ingest and query time — no separate embedding service required
Mcp Server
Official Model Context Protocol server exposing vector operations as agent tools
Serverless
Pay-per-use scaling with separated storage and compute

Frameworks

langchainllamaindexvercel-aihaystack

SDK Languages

pythonjavascripttypescriptgojavarubyphpcsharprust

Compliance

soc2hipaagdprpci-dssiso27001

Elasticsearch

Elasticsearch added dense vector storage and HNSW indexing in 7.x and has steadily made vector search a first-class workload alongside its long-standing BM25 search. The result is one of the strongest hybrid-search stories on the market: dense vectors, sparse vectors (ELSER), and lexical search can be combined in a single _search query with reciprocal rank fusion.

For AI agents that need to ground answers in technical, proper-noun-heavy, or domain-specific corpora — where pure vector search underperforms — Elasticsearch's hybrid retrieval is hard to beat. The inference API hosts embedding and reranker models (E5, ELSER, Cohere, OpenAI, Hugging Face) so you can ingest raw text without a separate embedding service. The official Elasticsearch MCP server exposes index, search, and inference operations to agents.

Elastic Cloud Serverless decouples storage from compute and bills per usage, making bursty agent workloads cheaper than the classic node-based pricing. Multi-tenancy is handled through dedicated indices, index aliases, or document-level security depending on isolation needs.

The tradeoffs are weight and license. Elasticsearch runs on the JVM, wants a lot of memory, and brings a much larger feature surface (full-text search, observability, security analytics) than a pure vector workload needs. The license shift to Elastic License v2 / SSPL also matters for some users — though Elasticsearch added an AGPL option in 2024.

Agent-specific features:

  • Hybrid retrieval (dense + ELSER sparse + BM25) with RRF in a single query
  • inference API hosts embedding and reranker models inside the cluster
  • Document-level security for multi-tenant agent apps
  • Official Elasticsearch MCP server for agent-driven search and ingest
  • Elastic Cloud Serverless with separated storage and compute
Last verified: 2026-06-10Verified by: editorial
docs ↗pricing ↗changelog ↗Spot a mistake? Open a PR →