OpenSearch 3.3's New AI Agents Now Generally Available for Developers (thenewstack.io)

OpenSearch 3.3 (released Oct. 14, 2025) makes its headline feature—AI agentic search and agentic memory APIs—generally available, letting developers embed autonomous plan-execute-reflect agents directly into their data stacks. These agents interpret natural-language queries, break down and rewrite complex questions into smarter retrieval plans, execute queries across sources and tools, and synthesize concise, relevant answers instead of returning raw document hits. That GA milestone, paired with expanded relevance scoring and fine-tuning controls, moves OpenSearch further toward deep semantic search and makes it a viable platform for large-scale, custom generative-AI and inference workflows. The release also advances ML and observability tooling: the ML Commons plugin gains experimental batch inference for distributed processing over huge vector datasets, and the new Seismic neural sparse-search algorithm boosts vector search performance. UI and observability upgrades include a redesigned Discover interface for log analytics and distributed tracing, multi-source dashboards, searchable snapshots across clusters, and improved OpenTelemetry instrumentation. Operational improvements add rule-based auto-tagging, query monitoring, expanded gRPC and experimental Apache Arrow Flight streaming, plus safety limits (JSON nesting and property name lengths) to protect cluster stability. OpenSearch 3.3’s combination of agentic AI, vector-scale inference, and unified observability targets developers building production-grade generative search and monitoring systems across cloud and on-prem environments (Linux/Windows/Docker/FreeBSD/Arch).