Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 48 additions & 0 deletions release-notes/opensearch-ml-commons.release-notes-3.8.0.0.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,48 @@
## Version 3.8.0 Release Notes

Compatible with OpenSearch and OpenSearch Dashboards version 3.8.0

### Features

* Support MCP for Flow and Conversational Flow agents ([#4776](https://github.com/opensearch-project/ml-commons/pull/4776))
* Add REST API to list all tools available in an MCP server ([#4705](https://github.com/opensearch-project/ml-commons/pull/4705))
* Introduce gRPC streaming support for ML prediction and agent execution ([#4790](https://github.com/opensearch-project/ml-commons/pull/4790))
* Add memory retention policy data model, pinned field, and API wiring (1/2) ([#4914](https://github.com/opensearch-project/ml-commons/pull/4914))
* Add memory retention job, cluster defaults, and hardening (2/2) ([#4918](https://github.com/opensearch-project/ml-commons/pull/4918))

### Enhancements

* Support tool description override in MCP connector ([#4766](https://github.com/opensearch-project/ml-commons/pull/4766))
* Refine planner agent prompt ([#4822](https://github.com/opensearch-project/ml-commons/pull/4822))
* Add JSON logs to agent flow ([#4740](https://github.com/opensearch-project/ml-commons/pull/4740))
* Serve `/_plugins/_ml/mcp` on demand instead of a per-node cache ([#4907](https://github.com/opensearch-project/ml-commons/pull/4907))

### Bug Fixes

* Fix MCP agent execution hang after MCP SDK 1.1.1 bump ([#4816](https://github.com/opensearch-project/ml-commons/pull/4816))
* Return HTTP 429 instead of 500 on connection pool acquire timeout ([#4852](https://github.com/opensearch-project/ml-commons/pull/4852))
* Fix `system_prompt` not reaching Bedrock Converse ([#4871](https://github.com/opensearch-project/ml-commons/pull/4871))
* Fix SearchIndexTool frequently failing ([#4530](https://github.com/opensearch-project/ml-commons/pull/4530))
* Retry transient HTTP errors from any remote service ([#4882](https://github.com/opensearch-project/ml-commons/pull/4882))
* Fix ModelGuardrail and LocalRegexGuardrail fail-open bugs (now fail-closed) ([#4904](https://github.com/opensearch-project/ml-commons/pull/4904))
* Fix gRPC stream double-termination ([#4927](https://github.com/opensearch-project/ml-commons/pull/4927))
* Validate model archives to block arbitrary code execution ([#4929](https://github.com/opensearch-project/ml-commons/pull/4929))
* Return correct HTTP status on memory-container failures instead of 500 ([#4930](https://github.com/opensearch-project/ml-commons/pull/4930))

### Infrastructure

* Pin GitHub Actions to commit SHAs ([#4828](https://github.com/opensearch-project/ml-commons/pull/4828))
* Onboard new reusable backport-pr GitHub workflow ([#4873](https://github.com/opensearch-project/ml-commons/pull/4873))
* Update maven2 mirror repository URL order ([#4887](https://github.com/opensearch-project/ml-commons/pull/4887))
* Fix flaky IT `testChatAgentWithMcpStreamableHttpConnector` and add timeout ([#4851](https://github.com/opensearch-project/ml-commons/pull/4851))
* Retry integration tests on transient remote-service 5xx errors ([#4896](https://github.com/opensearch-project/ml-commons/pull/4896))
* Disable fail-fast on CI matrix so Java 21/25 jobs run independently ([#4899](https://github.com/opensearch-project/ml-commons/pull/4899))
* Fix flaky CI: httpbin dependency, MCP tool sync race, and SearchModelGroupITTests hang ([#4903](https://github.com/opensearch-project/ml-commons/pull/4903))
* Add `jiapingzeng` as maintainer ([#4910](https://github.com/opensearch-project/ml-commons/pull/4910))
* Bump bc-fips to 2.1.3 in plugin test classpath to fix CVE-2026-8149 ([#4923](https://github.com/opensearch-project/ml-commons/pull/4923))

### Documentation

* Pass Gemini API key via header instead of URL in connector blueprint ([#4874](https://github.com/opensearch-project/ml-commons/pull/4874))
* Add LLM judgment connector blueprints ([#4878](https://github.com/opensearch-project/ml-commons/pull/4878))
* Use http instead of https for Ollama connector endpoint blueprint ([#4885](https://github.com/opensearch-project/ml-commons/pull/4885))
Loading