Neo4j has introduced Document Intelligence in preview for Aura, aiming to help teams turn document collections into knowledge graphs for connected retrieval and AI applications. Neo4j documentation describes support for cloud sources, local file streaming, and formats including PDF, Markdown, DOCX, TXT, and EPUB.
Neo4j announced a preview of Document Intelligence in Aura, a new capability intended to turn document collections into knowledge graphs for retrieval and AI application development.
In a Neo4j blog post titled “Introducing Document Intelligence: From documents to a knowledge graph, right inside Aura,” the company said Document Intelligence is now available in preview inside Aura, its managed cloud graph database service. Neo4j positions the feature as a way to move beyond document retrieval approaches that rely only on vector search or keyword indexes when applications need to reason across connected information.
The company’s core argument is that many enterprise documents contain relationships that are difficult to capture as isolated chunks. A contract may reference organizations, dates, obligations, locations, and related documents. A research archive may connect people, institutions, terms, and findings across many files. Neo4j says Document Intelligence is designed to extract such relationships into a knowledge graph, making them available for graph-based exploration and retrieval.
Neo4j’s Aura documentation describes Document Intelligence as a workflow for importing documents, generating graph models, and running imports inside the Aura interface. The documented visual tour includes areas for Graph models, Cloud data sources, Import jobs, prompt settings, model generation, and Run import.
According to Neo4j’s “Data provision” documentation for Aura Document Intelligence, users can connect cloud data sources or stream local files. The same documentation lists supported file formats including PDF, MD, DOCX, TXT, and EPUB. That range suggests Neo4j is targeting common enterprise and knowledge-work document collections rather than a single narrow file type.
The documentation also indicates that the product is built around a guided user interface rather than only a developer API. Users can define or generate a graph model, configure prompts, connect data, and run import jobs. In practice, that could make graph construction more accessible to teams that have documents and domain knowledge but do not want to hand-build every node and relationship type before testing a retrieval workflow.
Neo4j’s announcement places Document Intelligence in the broader context of AI systems that need reliable access to connected enterprise information. The company argues that vector search and keyword search can be useful, but may not be enough when questions require understanding relationships across documents and entities.
That framing aligns with a common limitation in retrieval-augmented generation systems: a retrieved passage may contain relevant text, but the system may still lack structured context about how a person, policy, product, event, or organization relates to other records. Neo4j’s approach is to represent extracted information as a graph so applications can traverse relationships, inspect provenance, and query connected facts.
Neo4j’s announcement does not, based on the provided source excerpts, claim that Document Intelligence eliminates the need for other retrieval methods. Instead, it presents knowledge graphs as complementary infrastructure when connected retrieval is important. That distinction matters because many production AI systems combine multiple techniques, including text search, embeddings, metadata filters, structured queries, and graph traversal.
Because Neo4j describes Document Intelligence as a preview, organizations should treat it as an emerging Aura capability rather than a fully mature replacement for existing ingestion and data-modeling processes. The provided Neo4j documentation shows the main workflow concepts and supported data inputs, but teams evaluating it would still need to test extraction quality, governance requirements, source coverage, and integration with their application stack.
For companies already using Neo4j Aura, the preview may reduce the effort required to experiment with document-derived knowledge graphs. For teams evaluating graph-based retrieval, it provides another sign that database vendors are packaging AI-oriented ingestion, modeling, and retrieval workflows directly into managed platforms.
The larger significance is not that every document repository should become a graph by default. It is that vendors such as Neo4j are trying to make graph construction less manual at a time when AI applications increasingly need structured, explainable access to enterprise knowledge.
Neo4j announced a preview of Document Intelligence in Aura, a new capability intended to turn document collections into knowledge graphs for retrieval and AI application development.
Neo4j positions the feature as a way to move beyond document retrieval approaches that rely only on vector search or keyword indexes when applications need to reason across connected information.
The company’s core argument is that many enterprise documents contain relationships that are difficult to capture as isolated chunks.
Continue reading
