1.16.0

Added

• Add close/aclose methods to LLMBase by @stellasia in #498
• Add warning to upsert_vectors + fix errors in doc generation by @stellasia in #515
• Update dependencies by @stellasia in #516
• Pass http_client to the proper (sync or async) client in OpenAILLM by @stellasia in #517
• Add an extra ORDER BY to cypher retrievers by @stellasia in #518
• Add Cypher 25 SEARCH clause for vector retrievers by @oskarhane in #488
• cypher query explained before executed to detect destructive queries by @serengil-neo4j in #520
• Use GraphSchema type instead of dict in ParquetWriter by @AmirLayegh in #523
• Include EXISTENCE constraints in metadata emitted by ParquetWriter by @adamnsch in #524

Fixed

• docs(llm): fix copy-pasted model names in docstrings by @chappiii in #504

New Contributors

• @serengil-neo4j made their first contribution in #520
• @chappiii made their first contribution in #504

Full Changelog: 1.15.0...1.16.0

1.15.0

Added

• feat: Add support for Markdown files by @ali-sedaghatbaf in #496
• feat: add Amazon Bedrock support for LLM and Embeddings by @adityachap in #475
• Introduce LLMBase by @ali-sedaghatbaf in #503
• Align KG builder schema (eg. GraphSchema) with Neo4j-style constraints
  • Add a separate schema model for LLM structured output by @adamnsch in #506
  • Introduce Neo4j-style EXISTENCE constraints for KG schemas by @adamnsch in #510
  • Add support for KEY constraints to KG builder schema by @adamnsch in #511
  • Support composite constraints in KG builder schemas by @adamnsch in #512
  • Ensure there is always a single-property key constraint by @adamnsch in #519
• Add token usage tracking by @ali-sedaghatbaf in #513

Fixed

• fix: extract_cypher fails on valid Cypher with map literals by @oskarhane in #509
• fix: Add default name property to node types defined without a properties key by @ali-sedaghatbaf in #499
• fix: build explicit PyArrow schema in Neo4jGraphParquetFormatter by @matteomedioli in #500
• fix: test embedding float32 followup (GENKGB-1065) by @matteomedioli in #514

New Contributors

• @adityachap made their first contribution in #475
• @matteomedioli made their first contribution in #500

Full Changelog: 1.14.1...1.15.0

1.14.1

What's Changed

• Forbid __ prefix and suffix in schema extraction for entity labels
• Use rate limit handler on VertexAILLM and MistralAILLM v2 invoke methods
• Fix ValueError in Neo4jGraphParquetFormatter when nodes of the same label have mixed property types

New Contributors

• @adamnsch made their first contribution in #492

Full Changelog: 1.14.0...1.14.1

1.14.0

What's Added

• Add Parquet writer by @ali-sedaghatbaf in #486

What's Changed

• Update gpt model references by @ali-sedaghatbaf in #482
• Fix deprecated apoc procedure addLabels by @NathalieCharbel in #484
• Enable structured output for SimpleKGPipeline by @NathalieCharbel in #481

New Contributors

• @ali-sedaghatbaf made their first contribution in #482

Full Changelog: 1.13.1...1.14.0

1.13.1

What's Changed

• Make rate limit handler open to which exceptions it can retry on by @jonnylaw in #474
• Fix vertexai init by @jonnylaw in #476
• Fix graph pruner lexical relationship validation by @AmirLayegh in #477

New Contributors

• @jonnylaw made their first contribution in #474

Full Changelog: 1.13.0...1.13.1

Neo4j GraphRAG Package for Python 1.13.0

What's Changed

• Add support for structured output in LLMInterfaceV2, OpenAILLM and VertexAILLM
• Enable Structured Output for schema from text extractor
• Implement tool calling for Ollama and include examples
• Support async embedding for Embedder base class and Ollama implementation
• Added cypher template example

New Contributors

• @kalrasamridhi7 made their first contribution in #441
• @zh-tan made their first contribution in #468

Full Changelog: 1.12.0...1.13.0

Neo4j GraphRAG Package for Python 1.12.0

Added in 1.12.0

• Support for Python 3.14
• Support for extracting the required property field from SchemaFromTextExtractor, enabling the LLM to distinguish mandatory node properties from optional ones when generating a schema.

Changed in 1.12.0

• Replaced poetry with uv as the project's dependency management tool
• Dropped support for python 3.9

Full Changelog: https://github.com/neo4j/neo4j-graphrag-python/blob/main/CHANGELOG.md#1120

Neo4j GraphRAG Package for Python 1.11.0

Full Changelog: 1.10.1...1.11.0

Added in 1.11.0

External retrievers

• Added an optional node_label_neo4j parameter in the external retrievers to speed up the search query in Neo4j.
• Added an optional id_property_getter callable parameter in the Qdrant retriever to allow for custom ID retrieval.

Get schema

• Exposed optional sample parameter on get_schema and get_structured_schema to control APOC sampling for schema discovery.

Pipeline

• Send pipeline failed events to the event callback

New Contributors

• @devatcadam made their first contribution in #437
• @androna-xm made their first contribution in #450
• @AmirLayegh made their first contribution in #455

Neo4j GraphRAG Package for Python 1.10.1

Added in 1.10.1

Full Changelog: 1.10.0...1.10.1

KG Construction

• Create Document node even from text input in SimpleKGPipeline
• Clean extracted schema json in the SchemaFromTextExtractor

Repository config

• Add issue templates for bugs, features, docs, and questions

Fixed in 1.10.0

• Add rate limit handling to embedders

• Fix e2e tests failure due to disk space

• Fix doc about custom data loader

• Fix LLMEntityRelationExtractor user guide

New Contributors

• @serengil made their first contribution in #429
• @martinohanlon made their first contribution in #438

Neo4j GraphRAG Package for Python 1.10.0

Full Changelog: 1.9.1...1.10.0

Added in 1.10.0

KG Construction

• Added schema_visualization function to visualize a graph schema using neo4j-viz.
• Added a new schema builder: SchemaFromExistingGraphExtractor component

Retrieval

• Added ToolsRetriever class and Retriever.convert_to_tool() method

Fixed in 1.10.0

• Bumped pypdf - CVE-2025-55197
• Fixed ingestion error due to invalid property type (map)

New Contributors

• @andrezz-b made their first contribution in #417