3.3.0.0

Version 3.3.0 Release Notes

Compatible with OpenSearch and OpenSearch Dashboards version 3.3.0

Features

• Add WriteToScratchPad and ReadFromScratchPad tools (#4192)
• Add global resource support (#4003)
• Add ml-commons passthrough post process function (#4111)
• Add output transformation support with mean pooling for ML inference processors (#4236)
• Adding query planning tool search template validation and integration tests (#4177)
• Onboards to centralized resource access control mechanism for ml-model-group (#3715)
• Refactor Agentic Memory (#4218)
• Search Template Support for QueryPlanningTool (#4154)
• [Agentic Search] Support Query Planner Tool with Conversational Agent (#4203)
• [FEATURE] Add Index Insight Feature (#4088)
• [FEATURE] Agent Execute Stream (#4212)
• [FEATURE] Predict Stream (#4187)
• [MCP Connector] MCP Connectors for streamable HTTP (#4169)
• Add processor chain and add support for model and tool (#4093)
• Add create session API; add message id to working memory; fix update api (#4246)

Enhancements

• Add Get Agent to ML Client (#4180)
• Parameter Passing for Predict via Remote Connector (#4121)
• Support multi-tenancy for LocalRegexGuardrail (#4120)
• [MCP Server] Support Streamable HTTP and deprecate SSE in MCP server (#4162)
• [Memory] Add updated time to message (#4201)
• [Metrics] Introduce agent metrics & Add is_hidden tag for model metrics (#4221)
• Update interaction with failure message on agent execution failure (#4198)
• Add PlainNumberAdapter and corresponding tests for Gson in SearchIndexTool (#4133)
• Move HttpClientFactory to common to expose to other components (#4175)
• Change the setting name to same naming convention with others (#4215)
• Enabling agentic memory feature by default as we are going GA (#4240)
• Add parameter to control delete memories when delete container (#4238)
• Refactor and add more validation to processor chain (#4260)
• [Agentic Search] Use same model for Agent and QPT (#4262)
• Improve semantic fact extraction prompt and add JSON enforcement (#4282)
• Improve user preference extraction prompt with XML-based structure (#4288)
• Enable execute tool feature flag by default (#4296)

Bug Fixes

• Fix NPE when execute flow agent with mutli tenancy is off (#4189)
• Fix claude model it (#4167)
• Fix error_prone_annotations jar hell (#4214)
• Fix failing UTs and increment version to 3.3.0-SNAPSHOT (#4132)
• Fix jdt formatter error (#4151)
• Fix missing RAG response from generative_qa_parameters (#4118)
• Fix model deploy issue and address other comments in #4003 (#4207)
• Fix: refactor memory delete by query API to avoid anti pattern (#4234)
• Fix MLTaskState enum serialization errors (#4158)
• Fix connector tool IT (#4233)
• Agent/Tool Parsing Fixes (#4138)
• [Metrics Framework] Fix version checking logic for starting the stats collector job (#4220)
• Fixing build issue in ml-commons (#4210)
• Fixing metrics correlation algorithm (#4200)
• Fixing validate access for multi-tenancy (#4196)
• Make MLSdkAsyncHttpResponseHandler return IllegalArgumentException (#4182)
• Skip the model interface validation for batch predict (#4219)
• Fix: use builtin BulkByScrollResponse parser to parse delete by query response in memories (#4237)
• Fix stream predict with security enabled (#4248)
• Fix wrong field name in get working memory API (#4255)
• Fix: allow only container owner to delete memory container (#4258)
• [Agent Framework] Exception handling for runtime exceptions during async execution (#4263)
• Fix json parsing error by extracing json first; add for each processor; support input processor (#4278)
• Fix: add validations during create and update memory container (#4284)
• Fix agent streaming with security enabled + error handling (#4256)
• Fix llm result path; convert message to user prompt string (#4283)
• Fix llm result path error (#4292)
• Fix dimension update flow to allow embedding type update (#4297)
• Verify llm before summarize session (#4300)

Infrastructure

• Update maintainer list (#4139)
• Downloads test certificates from security plugin (#4245)
• Update approver matching to be exact match (#4247)
• Fix approver parsing bug in require-approval workflow (#4259)

Documentation

• Add colpali blueprint (#4130)
• Add tutorial for agentic search (#4127)
• Ollama connector blueprint (#4160)
• Tutorial on agentic memory with strands agents (#4125)
• Change suggested instance type in tutorial (#4145)

Maintenance

• Adding more unit tests (#4124)
• Adding more unit tests (#4126)
• Move common string (#4173)
• Fix Cohere IT (#4174)
• Updating gson version to resolve conflict coming from core (#4176)

3.2.0.0

Version 3.2.0 Release Notes

Compatible with OpenSearch and OpenSearch Dashboards version 3.2.0

Features

• Initiate query planning tool (#4006)
• Add Execute Tool API (#4035)
• Implement create and add memory container API (#4050)
• Enable AI-Oriented memory operation on Memory APIs (Add, Search, Update & Delete) (#4055)
• Support output filter, unify tool parameter handling and improve SearchIndexTool output parsing (#4053)
• Delete memory container API (#4027)
• GET memory API (#4069)

Enhancements

• Add Default System Prompt for the query Planner tool (#4046)
• Add support for date time injection for agents (#4008)
• Expose message history limit for PER Agent (#4016)
• [Enhancement] Enhance validation for create connector API (#3579)
• [Enhancements] Sparse encoding/tokenize support TOKEN_ID format embedding (#3963)
• Add validation for creating uri in connectors (#3972)
• Enhance tool input parsing and add agentic rag tutorial (#4023)
• Run auto deploy remote model in partially deployed status (#3423)
• [ExceptionHandling] Throw proper 400 errors instead of 500 for agent execute and MCP (#3988)
• Tuning PER Agent Prompts (#4059)
• Add feature flag for agentic search (#4021)
• Adding feature flag for agentic memory (#4067)
• Add feature flag to delete mem container (#4072)

Bug Fixes

• Fix class cast exception for execute API (#4010)
• Fix delete connector/model group exception handling (#4044)
• Fix exposed connector URL in error message (#3953)
• Fix is_async status of agent execution task (#3960)
• Fix update model config invalid error (#3994)
• [FIX] allow partial updates to llm and memory fields in MLAgentUpdateInput (#4040)
• Fix the error status code and message for empty response (#3968)
• [MLSyncUpCron] Change info log to debug log to reduce logging (#3948)
• Fixing unit test for user_requested_tenant_access (#4037)
• Ensure chat agent returns response when max iterations are reached (#4031)

Maintenance

• Increase mcp code coverage and address comments in PR: #3883 (#3908)
• Fix: change log level for sync up job (#3948)
• Keep .plugins-ml-config index for Integration test (#3989)
• Adding unit tests for create and get memory container functionalities (#4056)
• Adding more unit tests and upgrading jacoco (#4057)
• CVE fix: beanutils (#4062)

Infrastructure

• Bump gradle to 8.14 and update JDK to 24 (#3983)
• Updating gradle version (#4064)
• [FIX] Update lombok version for jdk24 (#4026)

Documentation

• Add multi modal tutorial using ml inference processor (#3576)
• Add blueprint for semantic highlighter model on AWS Sagemaker (#3879)
• Add Documentation for creating Neural Sparse Remote Model (#3857)
• Add tutorials for language_identification during ingest (#3966)
• Update link to the model in the aleph alpha blueprint (#3980)
• Add agentic rag tutorial (#4045)
• Notebook for step by step in multi-modal search in ml-inference processor (#3944)

2.19.3.0

Version 2.19.3.0 Release Notes

Compatible with OpenSearch 2.19.3

Maintenance

• Force commons-beanutils to 1.11.0 (#4000)
• Force runtime class path commons-beanutils:commons-beanutils:1.11.0 to avoid transitive dependency (#3935)
• Fix commons-beanutils (#3996)
• Exclude commons-beanutils (#3991)
• Fix CVE-2025-48734 (#3986)
• Fix CVE-2025-48924 (#3985)
• Fix CVE-2025-27820 (#3984)
• Update the maven snapshot publish endpoint and credential (#3955)
• Adds Json Parsing to nested object during update Query step in ML Inference Request processor (#3868)

3.1.0.0

Version 3.1.0.0 Release Notes

Compatible with OpenSearch 3.1.0.0

Enhancements

• Support persisting MCP tools in system index (#3874)[https://github.com//pull/3874]
• [Agent] PlanExecuteReflect: Return memory early to track progress (#3884)[https://github.com//pull/3884]
• Add space type mapping for pre-trained embedding models, add new additional_config field and BaseModelConfig class (#3786)[https://github.com//pull/3786]
• support customized message endpoint and addressing comments (#3810)[https://github.com//pull/3810]
• Add custom SSE endpoint for the MCP Client (#3891)[https://github.com//pull/3891]
• Expose Update Agent API (#3820)[https://github.com//pull/3902]
• Use function calling for existing LLM interfaces (#3888)[https://github.com//pull/3888]
• Add error handling for plan&execute agent (#3845)[https://github.com//pull/3845]
• Metrics framework integration with ml-commons (#3661)[https://github.com//pull/3661]

Bug Fixes

• Fix connector private IP validation when executing agent without remote model (#3862)[https://github.com//pull/3862]
• for inline model connector name isn't required (#3882)[https://github.com//pull/3882]
• fix the tutorial in AIConnectorHelper when fetching domain_url (#3852)[https://github.com//pull/3852]
• Adds Json Parsing to nested object during update Query step in ML Inference Request processor (#3856)[https://github.com//pull/3856]
• adding / as a valid character (#3854)[https://github.com//pull/3854]
• quick fix for guava noclass issue (#3844)[https://github.com//pull/3844]
• Fix python client not able to connect to MCP server issue (#3822)[https://github.com//pull/3822]
• excluding circuit breaker for Agent (#3814)[https://github.com//pull/3814]
• adding tenantId to the connector executor when this is inline connector (#3837)[https://github.com//pull/3837]
• add validation for name and description for model model group and connector resources (#3805)[https://github.com//pull/3805]
• Don't convert schema-defined strings to other types during validation (#3761)[https://github.com//pull/3761]

Maintenance

• [Code Quality] Adding test cases for PlanExecuteReflect Agent (#3778)[https://github.com//pull/3778]
• Add Unit Tests for MCP feature (#3787)[https://github.com//pull/3787]
• exclude trusted connector check for hidden model (#3838)[https://github.com//pull/3838]
• add more logging to deploy/undeploy flows for better debugging (#3825)[https://github.com//pull/3825]
• remove libs folder (#3824)[https://github.com//pull/3824]
• Downgrade MCP version to 0.9 (#3821)[https://github.com//pull/3821]
• upgrade http client to version align with core (#3809)[https://github.com//pull/3809]
• Use stream optional enum set from core in MLStatsInput (#3648)[https://github.com//pull/3648]
• change SearchIndexTool arguments parsing logic (#3883)[https://github.com//pull/3883]

Infrastructure

• change release note (#3811)[https://github.com//pull/3811]

Documentation

• Replace the usage of elasticsearch with OpenSearch in README (#3876)[https://github.com//pull/3876]
• added blueprint for Bedrock Claude v4 (#3871)[https://github.com//pull/3871]

3.0.0.0

Version 3.0.0.0 Release Notes

Compatible with OpenSearch 3.0.0.0

Breaking Changes

• Use _list/indices API instead of _cat/index API in CatIndexTool (#3243)
• Deprecate the restful API of batch ingestion (#3688)

Features

• Onboard MCP (#3721)
• Plan, Execute and Reflect Agent Type (#3716)
• Support custom prompts from user (#3731)
• Support MCP server in OpenSearch (#3781)

Enhancements

• Support sentence highlighting QA model (#3600)
• Add parser for ModelTensorOutput and ModelTensors (#3658)
• Function calling for openai v1, bedrock claude and deepseek (#3712)
• Update highlighting model translator to adapt new model (#3699)
• Implement async mode in agent execution (#3714)

Bug Fixes

• Fix building error due to a breaking change from core (#3617)
• Fixing the circuit breaker issue for remote model (#3652)
• Fix compilation error (#3667)
• Revert CI workflow changes (#3674)
• Fix config index masterkey pull up for multi-tenancy (#3700)
• Fix error message when input map and output map length not match (#3730)
• Agent Framework: Handle model response when toolUse is not accompanied by text (#3755)
• Allow user to control react agent max_interations value to prevent empty response (#3756)
• Agent framework: Fix SearchIndexTool to parse special floating point values and NaN (#3754)
• Directly return Response objects from metadata client responses (#3768)
• Remove opensearch-ml-2.4.0.0.zip file that was added by random mistake (#3763)
• Replace null GetResponse with valid response and not exists (#3759)
• Fix ListIndexTool and SearchIndexTool (#3720)
• Support MCP session management (#3803)
• Support customized message endpoint and addressing comments (#3810)
• Excluding circuit breaker for Agent (#3814)

Maintenance

• Update CB setting to 100 to bypass memory check (#3627)
• Use model type to check local or remote model (#3597)
• Fixing security integ test (#3646)
• Remove forcing log4j version to 2.24.2 (#3647)
• Improve test coverage for MLHttpClientFactory.java (#3644)
• Improve test coverage for MLEngineClassLoader class (#3679)
• Typo: MLTaskDispatcher _cluster/settings api (#3694)
• Add more logs to troubleshoot flaky test (#3543)
• Add package for security test (#3698)
• Add sdk implementation to the connector search (#3704)
• Sdk client implementation for search connector, model group and task (#3707)
• Add Feature Flag for MCP connectors Feature ((#3738)
• Support phasing off SecurityManager usage in favor of Java Agent (#3729)
• Add java-agent gradle plugin (#3727)

Documentation

• Add tutorial for RAG of openai and bedrock (#2975)
• Fix template query link (#3612)
• Add standard blueprint for vector search (#3659)
• Add blueprint for Claude 3.7 on Bedrock (#3584)
• Add standard blueprint for azure embedding ada2 (#3725)

3.0.0.0-beta1

Version 3.0.0.0-beta1 Release Notes

Compatible with OpenSearch 3.0.0.0-beta1

Breaking Changes

• Deprecate the restful API of batch ingestion (#3688)[https://github.com//pull/3688]

Enhancements

• Add parser for ModelTensorOutput and ModelTensors (#3658)[https://github.com//pull/3658]
• Function calling for openai v1, bedrock claude and deepseek (#3712)[https://github.com//pull/3712]
• Update highlighting model translator to adapt new model (#3699)[https://github.com//pull/3699]
• Plan, Execute and Reflect Agent Type (#3716)[https://github.com//pull/3716]
• Implement async mode in agent execution (#3714)[https://github.com//pull/3714]

Bug Fixes

• Fixing the circuit breaker issue for remote model (#3652)[https://github.com//pull/3652]
• Fix compilation error (#3667)[https://github.com//pull/3667]
• Revert CI workflow changes (#3674)[https://github.com//pull/3674]
• Fix config index masterkey pull up for multi-tenancy (#3700)[https://github.com//pull/3700]

Maintenance

• Remove forcing log4j version to 2.24.2 (#3647)[https://github.com//pull/3647]
• Improve test coverage for MLHttpClientFactory.java (#3644)[https://github.com//pull/3644]
• Improve test coverage for MLEngineClassLoader class (#3679)[https://github.com//pull/3679]
• Typo: MLTaskDispatcher _cluster/settings api (#3694)[https://github.com//pull/3694]
• Add more logs to toubleshot flaky test (#3543)[https://github.com//pull/3543]
• Add package for security test (#3698)[https://github.com//pull/3698]
• Add sdk implementation to the connector search (#3704)[https://github.com//pull/3704]
• Sdk client implementation for search connector, model group and task (#3707)[https://github.com//pull/3707]

Documentation

• Add standard blueprint for vector search (#3659)[https://github.com//pull/3659]
• Add blueprint for Claude 3.7 on Bedrock (#3584)[https://github.com//pull/3584]

3.0.0.0-alpha1

Version 3.0.0.0-alpha1 Release Notes

Compatible with OpenSearch 3.0.0.0-alpha1

Breaking Changes

• Use _list/indices API instead of _cat/index API in CatIndexTool (#3243)[https://github.com//pull/3243]

Bug Fixes

• Fix building error due to a breaking change from core (#3617)[https://github.com//pull/3617]

Documentation

• Add tutorial for RAG of openai and bedrock (#2975)[https://github.com//pull/2975]
• Fix template query link (#3612)[https://github.com//pull/3612]

Enhancements

• Support sentence highlighting QA model (#3600)[https://github.com//pull/3600]

Maintenance

• Update CB setting to 100 to bypass memory check (#3627)[https://github.com//pull/3627]
• Use model type to check local or remote model (#3597)[https://github.com//pull/3597]
• Fixing security integ test (#3646)[https://github.com//pull/3646]

2.19.1.0

Version 2.19.1.0 Release Notes

Compatible with OpenSearch 2.19.1

Bug Fixes

• fix ignoreFailure flag in ml inference search processors (#3570)[https://github.com//pull/3570]
• updating bulk update to use sdkclient (#3546)[https://github.com//pull/3546]
• Add interface for no postprocessing function case (#3553)[https://github.com//pull/3553]
• add edge case for models that are marked as not found in cache (#3523)[https://github.com//pull/3523]
• applying sdkclient changes to config index (#3521)[https://github.com//pull/3521]
• remainig sdk client changes for search (#3522)[https://github.com//pull/3522]
• adding tenantId for inline connector in model registration (#3531)[https://github.com//pull/3531]
• Remove the integ test case cause cohere classify is deprecated in its llm service (#3513)[https://github.com//pull/3513]
• fix null decrypted credential while getting batch predict job status (#3518)[https://github.com//pull/3518]

2.19.0.0

Version 2.19.0.0 Release Notes

Compatible with OpenSearch 2.19.0

Enhancements

• adding multi-modal pre-processor for cohere (#3219)[https://github.com//pull/3219]
• [Enhancement] Fetch system index mappings from json file instead of string constants (#3153)[https://github.com//pull/3153]
• Retrieve remote model id from registration response in IT to avoid flaky (#3244)[https://github.com//pull/3244]
• [Refactor] Remove bloat due to unnecessary setup in test and add retry for potential flaky behavior due to timeout (#3259)[https://github.com//pull/3259/files]
• [Enhancement] Enhance validation for create connector API (#3260)[https://github.com//pull/3260]
• Add application_type to ConversationMeta; update tests (#3282)[https://github.com//pull/3282]
• Enhance Message and Memory API Validation and storage (#3283)[https://github.com//pull/3283]
• Use Adagrad optimiser for Linear regression by default (#3291)[https://github.com//pull/3291/files]
• [Enhancement] Add schema validation and placeholders to index mappings (#3240)[https://github.com//pull/3240]
• add action input as parameters for tool execution in conversational agent (#3200)[https://github.com//pull/3200]
• Remove ignore decorator for testCohereClassifyModel (#3324)[https://github.com//pull/3324]
• refactor: modifying log levels and adding more logs to display error details (#3337)[https://github.com//pull/3337]
• Primary setup for Multi-tenancy (#3307)[https://github.com//pull/3307]
• apply multi-tenancy and sdk client in Connector (Create + Get + Delete) (#3382)[https://github.com//pull/3382]
• adding multi-tenancy + sdk client related changes to model, model group and connector update (#3399)[https://github.com//pull/3399]
• applying multi-tenancy to task apis, deploy, predict apis (#3416)[https://github.com//pull/3416]
• adding tenantID to the request + undeploy request (#3425)[https://github.com//pull/3425]
• Check before delete (#3209)[https://github.com//pull/3209]
• multi-tenancy + sdk client related changes in agents (#3432)[https://github.com//pull/3432]
• Introduce Ml Inference Search Request Extension (#3284)[https://github.com//pull/3284]
• Cherry-pick BWC fix for system prompt and user instructions (#3437)[https://github.com//pull/3437]
• Add deepseek as a trusted endpoint. (#3440)[https://github.com//pull/3440]
• applying multi-tenancy in search [model, model group, agent, connector] (#3433)[https://github.com//pull/3433]
• Added amazon rekognition as a trust endpoint (#3419)[https://github.com//pull/3419]
• adding multi-tenancy to config api and master key related changes (#3439)[https://github.com//pull/3439]
• Undeploy models with no WorkerNodes (#3380)[https://github.com//pull/3380]
• support batch task management by periodically bolling the remote task via a cron job (#3421)[https://github.com//pull/3421]
• Add pre and post process functions for Bedrock Rerank API #3254 (#3339)[https://github.com//pull/3339]
• [Backport 2.19] [BACKPORT 2.x] applying multi-tenancy in search [model, model group, agent, connector] (#3433) (#3469)[https://github.com//pull/3469]
• [Backport 2.19] fixing connector validation (#3471)[https://github.com//pull/3471]
• [BACKPORT 2.x] applying multi-tenancy in search [model, model group, agent, connector] (#3433) (#3443) (#3469)[https://github.com//pull/3469]

Bug Fixes

• getFirst is not allowed in java 17 (#3226)[https://github.com//pull/3226]
• Fix: ml/engine/utils/FileUtils casts long file length to int incorrectly (#3198)[https://github.com//pull/3198]
• fix for sync up job not working in 2.17 when upgraded from previous versions (#3241)[https://github.com//pull/3241]
• Fix remote model with embedding input issue (#3289)[https://github.com//pull/3289]
• Adds preset contentRegistry for IngestProcessors (#3281)[https://github.com//pull/3281]
• Revert "Add application_type to ConversationMeta; update tests (#3282)" (#3315)[https://github.com//pull/3315]
• Revert "Filter out remote model auto redeployment (#2976)" and related commits (#3104, #3214) (#3368)[https://github.com//pull/3368]
• Fix JsonGenerationException error in Local Sample Calculator and Anomaly Localization Execution Response (#3434)[https://github.com//pull/3434]
• [Backport 2.19] Fix guardrail it for 2.19 (#3468)[https://github.com//pull/3468]
• addressing client changes due to adding tenantId in the apis (#3474) (#3480)[https://github.com//pull/3480]

Maintenance

• Bump guava version to 32.1.3 (#3300)[https://github.com//pull/3300]
• Update Gradle to 8.11.1 (#3139)[https://github.com//pull/3139/files]
• Force version 3.29.0 of org.eclipse.core.runtime to mitigate CVE vulnerabilities (#3313)[https://github.com//pull/3313]
• Upgraded software.amazon.awssdk from 2.25.40 to 2.29.0 to address CVE… (#3320)[https://github.com//pull/3320]
• Adding back Mingshi as Maintainer. (#3367)[https://github.com//pull/3367]
• updating sdk client version (#3392)[https://github.com//pull/3392]
• downgrading codecov action (#3409) (#3410)[https://github.com//pull/3410]
• fix CVE from ai.djl dependency (#3478) (#3482)[https://github.com//pull/3482]

Infrastructure

• Enable custom start commands and options to resolve GHA issues (#3223)[https://github.com//pull/3223]
• Add Spotless Check to maintain consistency (#3386)[https://github.com//pull/3386]
• Add runs-on field to Spotless Check step in CI (#3400)[https://github.com//pull/3400]
• Checkout code from pull request head for spotless (#3422)[https://github.com//pull/3422]
• Fixes spotless on Java 11 (#3449)[https://github.com//pull/3449]
• add spotless to all build.gradle files (#3453)[https://github.com//pull/3453]
• Fixes Two Flaky IT classes RestMLGuardrailsIT & ToolIntegrationWithLLMTest (#3253)[https://github.com//pull/3253]
• Improve test coverage for RemoteModel.java (#3205)[https://github.com//pull/3205]
• Revert Text Block changes from "Enhance validation for create connector API" #3260 (#3329)[https://github.com//pull/3329]

Documentation

• Fix: typo in MLAlgoParams (#3195)[https://github.com//pull/3195/files]
• Add tutorial for bge-reranker-m3-v2, multilingual cross-encoder model on SageMaker (#2848)[https://github.com/opensearch-project/ml-commons/commit/https://github.com//pull/2848]
• [Backport main] adding blue print doc for cohere multi-modal model (#3232)[https://github.com//pull/3232]
• Tutorial for using Asymmetric models (#3258)[https://github.com//pull/3258]
• add tutorials for cross encoder models on Amazon Bedrock (#3278)[https://github.com//pull/3278]
• fix typo (#3234)[https://github.com//pull/3234]
• Tutorial for ml inference with cohere rerank model (#3398)[https://github.com//pull/3398]
• Add DeepSeek connector blueprint (#3436)[https://github.com//pull/3436]
• fix post_process_function on rerank_pipeline_with_bge-rerank-m3-v2_model_deployed_on_Sagemaker.md (#3296)[https://github.com//pull/3296]

2.18.0.0

Version 2.18.0.0 Release Notes

Compatible with OpenSearch 2.18.0

Enhancements

• Filter out remote model auto redeployment (#2976)[https://github.com//pull/2976]
• Allow llmQuestion to be optional when llmMessages is used. (#3072)[https://github.com//pull/3072]
• Enhance batch job task management by adding default action types (#3080)[https://github.com//pull/3080]
• use connector credential in offline batch ingestion (#2989)[https://github.com//pull/2989]
• change to model group access for batch job task APIs (#3098)[https://github.com//pull/3098]
• add rate limiting for offline batch jobs, set default bulk size to 500 (#3116)[https://github.com//pull/3116]
• Support ML Inference Search Processor Writing to Search Extension (#3061)[https://github.com//pull/3061]
• Enable pass query string to input_map in ml inference search response processor (#2899)[https://github.com//pull/2899]
• add config field in MLToolSpec for static parameters (#2977)[https://github.com//pull/2977]
• Add textract and comprehend url to trusted enpoints (#3154)[https://github.com//pull/3154]

Bug Fixes

• Fix ml inference ingest processor always return list using JsonPath (#2985)[https://github.com//pull/2985]
• populate time fields for connectors on return (#2922)[https://github.com//pull/2922]
• Fix for rag processor throwing NPE when optional parameters are not provided (#3057)[https://github.com//pull/3057]
• Fix PR #2976 bug due to missing adding function_name and algorithm in querying models (#3104)[https://github.com//pull/3104]
• Gracefully handle error when generative_qa_parameters is not provided (#3100)[https://github.com//pull/3100]
• Fix error log to show the right agent type (#2809)[https://github.com//pull/2809]
• fix model stuck in deploying state during node crash/cluster restart (#3137)[https://github.com//pull/3137]
• increase the wait timeout to fetch the master key (#3151)[https://github.com//pull/3151]
• Handle BWC for bedrock converse API (#3173)[https://github.com//pull/3173]

Maintenance

• Bump protobuf version to 3.25.5 to patch potential DOS (#3083)[https://github.com//pull/3083]
• removing api keys from the integ test log (#3112)[https://github.com//pull/3112]
• allowing backport prs to skip approval (#3132)[https://github.com//pull/3132]
• updating the approval requirement (#3148)[https://github.com//pull/3148]
• unblocking the integ test pipeline for release (#3159)[https://github.com//pull/3159]

Infrastructure

• Support index.auto_expand_replicas 0-all for .plugins-ml-config (#3017)[https://github.com//pull/3017]
• Add Test Env Require Approval Action (#3005)[https://github.com//pull/3005]
• upgrading upload artifact to v4 (#3162)[https://github.com//pull/3162]
• Bump actions/download-artifact from 3 to 4.1.7 in /.github/workflows (#2881)[https://github.com//pull/2881]

Documentation

• add tutorial for cross-account model invocation on amazon managed cluster (#3064)[https://github.com//pull/3064]
• support role temporary credential in connector tutorial (#3058)[https://github.com//pull/3058]
• connector blueprint for amazon bedrock converse (#2960)[https://github.com//pull/2960]
• Updates dev guide to inform the workflow approval step (#3062)[https://github.com//pull/3062]
• tune titan embedding model blueprint for v2 (#3094)[https://github.com//pull/3094]
• Add bedrock multimodal build-in function usage example in doc (#3073)[https://github.com//pull/3073]