OpenSearch Project Blog

At OpenSearchCon Europe 2026, Fernando Rejon Barrera from Zeta Alpha challenged standard vector search benchmarks by evaluating Lucene’s HNSW, Faiss, and jVector under realistic production workloads.

Learn how vector search in OpenSearch eliminates zero-result pages in e-commerce by acting as a semantic fallback when lexical query relaxing falls short.

Learn how one engineer went from OpenSearch user to plugin maintainer through community engagement, consistent contributions, and a passion for search relevance.

At OpenSearchCon Europe 2026, Laysa Uchoa from NordCloud reframed generative AI hallucinations as a retrieval problem rather than an LLM limitation. She demonstrated production RAG architectures where hallucination rates dropped by over 80% by prioritizing hybrid retrieval (combining BM25 lexical scoring and k-NN vector search), implementing disciplined chunking strategies, and utilizing rigorous evaluation frameworks over simple "vibes".

Learn how OpenSearch 3.5's LLM as a Judge feature improves search relevance evaluation by using large language models to automatically assess search result quality at scale.

The OpenSearch agent server is a multi-agent orchestration platform that routes tasks to specialized AI agents using MCP tools and intelligent context-aware routing.

At OpenSearchCon Europe 2026, Carl Meadows announced three major shifts positioning OpenSearch as the foundation for agentic AI: OpenSearch Launchpad for natural language application deployment, automated hybrid search configurations, and high-cardinality telemetry tracking. Additionally, CERN joined the OpenSearch Software Foundation to handle massive scale workloads.

A single space in a KNN field name broke OpenSearch snapshots. Learn how Aiven traced the bug, reported it, and helped the community ship a fix in OpenSearch v2.17.

Asymmetric embedding models in OpenSearch 3.5 improve semantic search relevance by up to 125% over symmetric models. Learn how they work and how to implement them.

At OpenSource Summit North America,OpenSearch open source community member Bobby Mohammed took the stage to unpack how OpenSearch is bridging the gap between traditional retrieval and the future of AI. If you weren’t able to catch the session live, here is your breakdown of how OpenSearch is providing the critical primitives required to power multi-agent AI architectures at scale.

Across all SDLC phases, AI agents now generate code, root-cause analyses, remediation plans, performance fixes, and standard operating procedures faster than humans can meaningfully review them. The bottleneck is no longer generation; it's deciding whether to trust what has been generated. This post shares four agents the OpenSearch team built across the SDLC and the harness-based verification loops that make them reliable.

Contact center customers shouldn't have to repeat themselves on every call. This post shows how OpenSearch agentic memory provides your AI agent with persistent memory across sessions. In this implementation, semantic and hybrid search retrieve relevant information on demand, so your agent remembers what matters.

The OpenSearch Project introduces a faster, smarter way to optimize your search experience. The OpenSearch Relevance Tuning Agent is designed to automate the heavy lifting for search relevance tuning, allowing you to move from "best guess" tuning to data-driven precision in a fraction of the time. OpenSearch Relevance Agent continuously analyzes user behavior signals and query patterns, generates data-driven hypotheses, and validates improvements through rigorous offline evaluation.

The OpenSearch Hadoop connector 2.0 adds Apache Spark 3.5 and 4 support, OpenSearch 3.x compatibility, Amazon OpenSearch Serverless support, and AWS SDK v2 migration.

Last week at OpenSearchCon Europe in Prague, the OpenSearch Project reached a milestone that many are calling a defining turning point. While the event was filled with news, from CERN joining the foundation to the release of OpenSearch 3.6, the headline that continues to make waves throughout the ecosystem is the official launch of the Long-Term Support (LTS) program.

Documents with tables, charts, and diagrams need more than text-based search. We benchmarked three multimodal document search approaches in OpenSearch on 1,000 report pages. ColPali late interaction reranking achieved the highest search quality (92% recall) but with higher latency, BDA modality-aware embedding offered a balanced middle ground, and text-only chunking was simplest to implement but scored lowest on visually rich content.

Day two of OpenSearchCon Europe in Prague ended the way the best conferences do: with something the community has been waiting for.

What a start to OpenSearchCon Europe in Prague. This morning's keynotes set the tone for what promises to be a great two days, and a few themes emerged that speak directly to where we are headed as a project and as a community.  

The OpenSearch Software Foundation is pleased to announce that CERN has joined as an associate member. The intergovernmental organization operates the world's largest particle accelerator and is responsible for landmark discoveries in particle physics, including the Higgs boson.