The buzz this week at Open Source Summit North America is undeniable. And it’s centered entirely around the monumental shift from traditional search infrastructure to autonomous, intelligent, and agentic systems.
On Monday, Bobby Mohammed, Principal Product Manager at AWS and core OpenSearch open-source community member, 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.
The Paradigm Shift: Why Traditional Search Isn’t Enough for Agents
For the past decade, search and observability architectures were built around a straightforward concept: single-term, single-shot retrieval. You ask a question, the database returns a matching keyword or vector embedding.
But we are living in a new era. Multi-agent architectures don’t just search; they reason, plan, adapt, and execute autonomously. They iterate in loops, meaning they require multi-hop retrievals, stateful long-term memory, and dynamic access to diverse data sources.
As Mohammed noted on stage during his presentation:
“If you look at the last decade since OpenSearch branched from Elastic, it evolved significantly from keyword retrieval to semantic search, and over the last few years into a premier vector database. But over the last year or two, we’ve seen a major transformation. Adopting agentic AI requires a fundamentally different way of approaching the infrastructure and the primitives you use to solve problems.”
To solve this, OpenSearch has introduced over ten agentic features in its latest OpenSearch 3.6 release.
Three game-changing capabilities that are fundamentally changing the developer experience:
1. OpenSearch Agent Skills (Powered by Anthropic’s Framework)
Enterprise LLMs are brilliant at general knowledge but completely blind to your highly specific domain data or complex search query logic. Borrowing from the widely adopted skills framework released by Anthropic, OpenSearch now provides pre-packaged, portable Agent Skills (configured via simple .env files) that extend an LLM’s capabilities instantly.
- The Search Skill: Enables developers to spin up a fully functioning lexical or semantic search application, complete with a user interface, in less than 10 minutes without needing deep search expertise.
- The Migration Skill: Makes migrating massive legacy workloads from Apache Solr to OpenSearch seamless. By executing a simple npx command, developers can use natural language to let an agent handle up to 90% of the migration heavy lifting.
2. Agentic Search Pipelines
Traditional search engines expect structured Domain Specific Language (DSL) queries. Agents expect natural language. OpenSearch now features built-in Agentic Query Rewriting, which translates natural language prompts into optimized sub-queries on the fly.
Even better, OpenSearch 3.6 introduces two out-of-the-box packaged agents engineered for maximum real-time performance:
- The Flow Agent: Optimized for low-latency, single-shot RAG (Retrieval-Augmented Generation) workflows, returning answers in a matter of seconds.
- The Conversational Agent: Seamlessly manages short-term, long-term, and session-based memory using the OpenSearch Memory API, shifting context intelligently so your agent personalizes results based on historical user behavior.
3. The Relevance Agent: A Search Engineer in a Box
Tuning search relevance has historically been a painful, manual process. Engineers had to comb through click logs, manually rewrite queries, test a tiny subset of data, and pray it worked in production.
The new OpenSearch Relevance Agent automates this entirely using a multi-agent architecture:
- User Behavior Insights (UBI): Automatically tracks and analyzes how users interact and click
- Hypothesis Agent: Formulates testing experiments (e.g., boosting specific fields or modifying aggregations)
- Evaluation Agent: Scores the results against performance baselines
Developers can now simply open their OpenSearch Dashboards, click “Ask AI,” and use natural language to ask:
“What is the worst-performing query in our system right now, and how do we fix it?”
The Relevance Agent takes care of the rest, yielding benchmarked accuracy improvements ranging from 36% to over 200% on complex queries.
Built for the Community, by the Community
None of these innovations happen in a vacuum. As an Apache 2.0-licensed project, OpenSearch thrives entirely on its community. Bobby highlighted that the project now boasts over 3,000+ active contributors and 500+ participating organizations globally.
Whether you are looking to ingest data via Data Prepper, scale your analytics, or build autonomous AI workflows, the OpenSearch ecosystem provides a complete, end-to-end open-source platform to do it securely and transparently.
Ready to Dive In?
The agentic revolution is here, and you don’t need a PhD in search engineering to leverage it.
- Download OpenSearch 3.6 today to test out the new Flow and Conversational Agents.
- Spin up the Agent Skills framework to build your first semantic search app in under 10 minutes.
- Join a local OpenSearch User Group or start your own to collaborate with the thousands of builders shaping the future of open-source AI.
