What's Changed

• Added LLMGraphTransformer for converting text documents into graph documents using an LLM, migrated from langchain-experimental.
• Added DataIngestionNotSupported exception, raised when attempting to ingest data into a relationship index.
• Skipped removal of the id property from search query results, so this field can now be present in retrieved documents' metadata.
• Replaced deprecated apoc.create.addLabels with native Cypher dynamic label assignment in Neo4jGraph.

New Contributors

• Rafal-Chrzanowski-IBM made their first contribution in #110
• Sanskar121543 made their first contribution in #119

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#0100

What's Changed

• Added LangChain-specific user-agent metadata to Neo4j driver initialization across graph, vectorstore, and checkpoint integrations.
• Fixed GraphCypherQAChain compatibility with non-Neo4j GraphStore implementations (for example AGEGraph) by not requiring the private _enhanced_schema attribute.

New Contributors

• Dependabot (@dependabot)[bot] made their first contribution in #94
• John Kennedy (@jkennedyvz) made their first contribution in #97

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#090

What's Changed

• Added Neo4jSaver, a synchronous checkpoint saver for persisting LangGraph graph state in Neo4j.
• Added AsyncNeo4jSaver, an asynchronous checkpoint saver for persisting LangGraph graph state in Neo4j.

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#080

What's Changed

• Added bearer token authentication support to Neo4jGraph via the token parameter
• Added support for Python 3.14
• Added support for Neo4j Python Driver v6.0.0+
• Switched project dependency management from Poetry to uv
• Updated docstrings and reference documentation formatting for the new LangChain reference docs site (MkDocs/Markdown)
• Fixed Neo4jVector to properly handle missing metadata in retrieval results
• Fixed Neo4jVector.from_existing_index retrieval query generation when using multiple text_node_properties
• Fixed Neo4jChatMessageHistory to use database_ instead of database in execute_query calls

New Contributors

• Mason Daugherty (@mdrxy) made their first contribution in #81
• Sefik Ilkin Serengil (@serengil) made their first contribution in #85
• Bayu Siddhi Mukti (@bayu-siddhi) made their first contribution in #75

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#070

What's Changed

• Makes LangChain Neo4j compatible with LangChain 1.0.0 by replacing langchain with langchain-classic as a dependency
• Removes support for Python 3.9 as this is no longer supported in langchain-classic
• Updates README examples to use langchain-classic

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#060

What's Changed

• Added support for custom database names in Neo4jChatMessageHistory (no longer limited to the default "neo4j").
• Introduced an optional mmr dependency group for maximal marginal relevance search, including numpy.
• Bumped the neo4j-graphrag dependency from ^1.5.0 to ^1.9.0.
• Prevented infinite loops in CypherQueryCorrector.extract_paths when a Cypher query contains repeated paths.
• Replaced direct result["text"] indexing with result.get("text") in Neo4jVector.similarity_search_with_score_by_vector to handle missing text fields gracefully.
• Updated the empty-list check in CypherQueryCorrector.detect_node_variables so nodes without an explicit variable are properly skipped, rather than misclassifying all labels under an empty variable.
• Stopped Neo4jChatMessageHistory from creating duplicate session nodes.
• Fixed a schema bug in Neo4jGraph (with enhance_schema=True) where relationships having boolean properties would trigger a malformed RETURN {} AS output Cypher query, preventing graph initialization.
• Ensured custom database names are correctly passed through to retrieve_vector_index_info and retrieve_existing_fts_index in Neo4jVector.

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#050

What's Changed

• Added the neo4j-graphrag package as a dependency.
• Renamed the type property to role on Message nodes in Neo4jChatMessageHistory for improved consistency with the neo4j-graphrag package.
• Updated GraphCypherQAChain to align with the schema format used by Neo4jGraph, ensuring a unified schema structure format across the package.
• Replaced code used to query vector and full text indexes in the vectorstores.neo4j_vector module with equivalents from the neo4j-graphrag package.
• Replaced database schema retrieval code in the graphs.neo4j_graph module with equivalents from the neo4j-graphrag package.
• Replaced the Cypher queries in Neo4jChatMessageHistory with equivalents from the neo4j-graphrag package.
• Refactored the construct_schema function in GraphCypherQAChain to utilise schema retrieval functions from neo4j-graphrag.
• Replaced the legacy extract_cypher function with an improved version from neo4j-graphrag, offering improved query extraction.
• Introduced a delete_session_node parameter to the clear method in Neo4jChatMessageHistory, allowing optional deletion of the Session node when deleting message chains.

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#040

What's Changed

• Disabled some Neo4jGraph driver warnings to reduce unnecessary log noise and improve user experience.
• Made the source parameter of GraphDocument optional, offering greater flexibility when inserting data into the database.
• Introduced an optional parameter for specifying the embedding dimension in Neo4jVector, reducing the number of potentially unnecessary database calls.
• Fixed Cypher query parsing to handle node names with spaces, ensuring more accurate query execution.
• Resolved multiple issues in the enhanced schema creation function, improving schema accuracy and eliminating previously raised deprecation warnings.
• Updated GraphStore to implement the Protocol interface, enabling any graph database class adhering to GraphStore’s method signature to be used with GraphCypherQAChain.
• Added support for removing the previously unhandled Lucene special character (’/’) in remove_lucene_chars, ensuring proper handling in Cypher queries that utilize vector and full-text indexes.
• Introduced a test_versioning_check test to ensure that version-checking logic accommodates Neo4j’s new calendar-based versioning introduced in early 2025.

New Contributors

• Dennis Liu (@tdennisliu) made their first contribution in #31
• Yuxin Chen (@chkaty) made their first contribution in #24

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#030

What's Changed

• Enhanced Neo4j driver connection management with more robust error handling.
• Streamlined connection state checks within the Neo4jGraph class for better performance and maintainability.
• Introduced the effective_search_ratio parameter in Neo4jVector, allowing for finer control over query accuracy by adjusting the candidate pool size in similarity searches.
• Fixed instantiation issues in GraphCypherQAChain by removing the deprecated LLMChain and resolving compatibility with the use_function_response parameter.
• Eliminated redundant # type: ignore comments, enhancing type safety and code clarity across the codebase.

New Contributors

• jason (@jtanningbed) made their first contribution in #14
• Tomaz Bratanic (@tomasonjo) made their first contribution in #18

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#020

What's Changed

• Removed dependency on LangChain Community package by integrating necessary components directly into the LangChain Neo4j codebase.
• Fixed bugs in the Neo4jVector and GraphCypherQAChain classes preventing these classes from working with versions < 5.23 of Neo4j.

Full Changelog: https://github.com/langchain-ai/langchain-neo4j/blob/main/CHANGELOG.md#011