databricks-ai-bridge

New Features

• databricks-ai-bridge: Durable, crash-resumable agent execution for MLflow ResponsesAgent handlers (#425)

Improvements

• databricks-ai-bridge / databricks-langchain / databricks-openai: migrate databricks-vectorsearch to databricks-ai-search (#437)

Bug Fixes

• databricks-openai: forward extra kwargs to maintain compatibility with newer versions of openai-agents (#426, #429)

New Features

• databricks-ai-bridge / databricks-langchain / databricks-openai: Agent Memory SDKs now allow custom schemas for Lakebase-backed checkpointers, stores, and sessions (#415)

Improvements

• databricks-langchain: ChatDatabricks now uses DatabricksOpenAI under the hood (#418)
• Agent Memory SDK cleanup across databricks-ai-bridge Lakebase utilities and databricks-langchain checkpoint/store (#420)

Bug Fixes

• databricks-ai-bridge: Use the last query attachment in _parse_attachments to support Genie self-correction (#419)

Release

• Support Lakebase autoscaling in stateful agent SDKs

Bug Fixes

• databricks-openai: Patch release to fix bug in 0.11.0 where importing classes crashed unless optional dependencies were installed

New Features

• databricks-ai-bridge: Added AsyncLakebaseSQLAlchemy class for SQLAlchemy async engine with automatic OAuth token rotation
• databricks-openai: Added AsyncDatabricksSession class supporting Session Protocol for stateful conversation management
• databricks-openai: Added support for querying Databricks apps via OpenAI client
• databricks-langchain: Added async support for similarity_search_by_vector_with_score
• databricks-langchain: Added async support to ChatDatabricks

Improvements

• Updated Lakebase token cache duration to 15 minutes per Databricks documentation

Testing & Infrastructure

• Added pytest coverage reporting to CI

• Lakebase Permission Granting SDK
• Do not convert numbers to scientific notation when converting to markdown for Genie
• Lazy imports in Genie
• Do not check workspace_client.current_user.me() in databricks-mcp

• LangChain MCP Server Adapters
• LangGraph DatabricksStore for long term memory on Databricks using Lakebase
• DCR Support for external UC Connections
• Single function and index MCP Servers for LangChain and OpenAI

• Addresses a dependency import bug introduced with the 0.10.0 release
• Adds support for passing conversation ID and return_pandas to return query results in pandas DataFrame format to databricks_langchain.GenieAgent

• Add support for LLM-generated filter parameters in VectorSearchRetrieverTool
• Support passing in Workspace Client to VSRetrieverTool
• Fix ChatModel Custom Outputs
• Vector Search add reranker parameter
• MCP Open AI Adapters
• Add Lakebase Pool Memory/Connection Pooling: LakebasePool and langchain CheckpointSaver

• Langchain V1 compatability
• Improve OBO user authz error message with guidance on WorkspaceClient instantiation
• Higher visibility of Genie API's execution steps in MLflow traces
• Add OBO Debug Mode
• MCP: Fixed the mcp tools_calls which required MCP_Session_id which is obtained in the session via initialize call

Improvements

• Add option to include similarity score in VectorSearchRetrieverTool result
• Introduce DatabricksMCPClient
• Include reasoning + SQL steps when Genie agent is queried
• Make ChatDatabricks compatible with ChatAgent and ResponsesAgent

Bug Fixes

• Fix Databricks Vector Search Filter implementation
• Improve performance for Genie Truncation
• Fix Bug for Genie Description
• Deprecate python 3.9 as a supported environment

Improvements

• Allow backticked vector search index names
• Include embedding model as a resource for VectorSearchRetrieverTool
• Support setting retriever schema (doc_uri, chunk_id) for VectorSearchRetrieverTool
• Support for passing filters at query time and forwarding arbitrary kwargs through VectorSearchRetrieverTool

Improvements

• Allow backticked vector search index names
• Include embedding model as a resource for VectorSearchRetrieverTool
• Support setting retriever schema (doc_uri, chunk_id) for VectorSearchRetrieverTool
• Support for passing filters at query time and forwarding arbitrary kwargs through VectorSearchRetrieverTool

Improvements

• Fix Databricks Connect and Tabulate Dependencies
• Remove default temperature for ChatDatabricks

Highlights

• Support On Behalf Of User rights with genie and VectorSearch Tool
• Update Genie Integration for PuPr APIs

Improvements

• Improve VectorSearchRetrieverTool by disabling unwanted notices
• Updated documentation for installing integration packages
• Update inline unitycatalog imports
• Better validation of VectorSearchRetrieverTools

Highlights

• Adding integration with databricks-openai
• Introducing api docs

Improvements

• Improving VectorSearchRetrieverTool API in both langchain and openai by introducing automatic description and resources needed to log the model
• Introducing model as an argument to be passed to ChatDatabricks in Langchain and deprecating notice endpoint to confer with the schema in Langchain.

This is a patch release that includes bugfixes.

Bug fixes:

• Update Genie API polling logic to account for COMPLETED query state (#16, @prithvikannan)

Initial version of databricks-ai-bridge and databricks-langchain packages

Features:

• Support for Databricks AI/BI Genie via the databricks_langchain.GenieAgent API in databricks-langchain
• Support for most functionality in the existing langchain-databricks under databricks-langchain. Specifically, this release introduces databricks_langchain.ChatDatabricks, databricks_langchain.DatabricksEmbeddings, and databricks_langchain.DatabricksVectorSearch APIs.