OpenSearch Project has released OpenSearch 3.8, adding expanded Model Context Protocol support, gRPC token streaming for machine-learning predictions, vector-search performance improvements and new PPL-based tools for observability analysis.
OpenSearch Project has released OpenSearch 3.8, expanding its open-source search and analytics platform with updates for AI integrations, vector workloads and observability.
According to the project’s release announcement, the version broadens support for the Model Context Protocol (MCP), a standard used to connect AI systems with external tools and data sources. OpenSearch says the integration now supports more agent architectures, potentially giving developers more options for connecting AI-driven applications to search and analytics capabilities.
Vector search is a central focus of the 3.8 release. OpenSearch says it has added Base64 vector ingestion and improved the performance of vector-oriented operations. The project reports vector ingestion can be up to 4.16 times faster, while radial-search throughput can improve by as much as 2.1 times. Those figures are project-reported benchmarks, and actual results will depend on data, indexing choices and deployment configuration.
The release also introduces gRPC transport for streaming machine-learning predictions. OpenSearch positions the change as a way to reduce latency when applications need generated tokens delivered incrementally rather than waiting for a complete model response.
OpenSearch 3.8 expands connector support for LLM-as-a-Judge workflows, according to the 3.8.0 release notes. These workflows use large language models to help assess search-result quality, which can be useful when teams are testing ranking changes or comparing relevance strategies.
The project says the broader provider support is intended to make it easier to scale relevance evaluation across more model options. The supplied release materials do not specify every newly supported provider or prescribe a single evaluation methodology.
For observability users, OpenSearch 3.8 adds a visual builder for Piped Processing Language (PPL) queries, alongside SQL-query support and an onboarding canvas aimed at simplifying initial analytics workflows. The project says these additions are designed to help users investigate logs and metrics with less manual query construction.
New PPL commands also target time-series analysis. OpenSearch describes the capabilities as tools for shaping, pivoting and comparing time-series data, which could help users reorganize and examine operational data during investigations.
At the engine level, the project’s GitHub release entry also lists Hive ingestion and HTTP/3 client support among the 3.8.0 updates. OpenSearch and OpenSearch Dashboards 3.8.0 are identified as compatible releases.
OpenSearch 3.8 is available through the project’s download page. Organizations evaluating the release should review the full release notes for component-level changes and compatibility details.
OpenSearch Project has released OpenSearch 3.8, expanding its open source search and analytics platform with updates for AI integrations, vector workloads and observability.
According to the project’s release announcement, the version broadens support for the Model Context Protocol (MCP), a standard used to connect AI systems with external tools and data sources.
OpenSearch says the integration now supports more agent architectures, potentially giving developers more options for connecting AI driven applications to search and analytics capabilities.
Continue reading
