October 2026 Summaries

17 posts from Neo4j

Filter
Month: Year:
Post Summaries Back to Blog
Microsoft Fabric and Neo4j Graph Intelligence are presented as a way for financial institutions to improve detection of organized peer-to-peer payment fraud by converting siloed transactional and customer data in OneLake Delta tables into connected property graphs. Conventional SQL-based AI agents can struggle with multi-hop relationships and generate false positives by treating common connections, such as public Wi-Fi addresses, the same as rare, high-confidence links like shared debit cards, phone numbers, or repeated payment paths. Neo4j’s graph algorithms, including Weakly Connected Components and PageRank, can filter high-traffic hubs, score meaningful relationships, identify likely fraud-ring clusters, and return concise evidence-based results. The integrated Fabric workflow allows teams to run graph analysis in notebooks, write risk scores and community identifiers back to Delta tables, and provide Fabric Data Agents with relationship-aware context while retaining enterprise controls such as single sign-on, Purview governance, and tenant isolation.
Oct 30, 2026 1,261 words in the original blog post.
Neo4j is launching The Analysis Room, a discussion series for intelligence practitioners and leaders focused on how modern tradecraft can respond to increasingly interconnected geopolitical, economic, criminal, cyber, and infrastructure threats. The initiative emphasizes that while AI can help analysts process, translate, compare, summarize, and connect large volumes of fragmented information, human judgment remains necessary to preserve source provenance, assess confidence, test alternative explanations, accommodate dissent, and ensure accountability. Neo4j positions its experience with relationship-focused data analysis as a basis for hosting conversations among experts across intelligence disciplines, including national security, policing, financial investigations, and threat intelligence. The first session, scheduled for December 2, 2026, will feature former CIA executive and Yale lecturer Dr. Chris Savos alongside Neo4j’s John Bender discussing AI’s effects on intelligence work, the continuing role of human analysts, and the skills likely to matter most in the future.
Oct 30, 2026 713 words in the original blog post.
MIT Technology Review Insights research cited by Neo4j finds that only 34% of agentic AI projects reach production, while a small group of “production leaders,” representing 13% of surveyed organizations, successfully deploys 61% of its projects by giving agents enterprise knowledge rather than data alone. The report identifies semantic knowledge of business concepts and relationships, procedural knowledge of how work is performed, and episodic memory of past actions and outcomes as essential for reliable agents, helping prevent errors such as flawed orders, database deletions, and inappropriate healthcare decisions. It argues that a knowledge layer built around knowledge graphs, ontologies, and agent memory can connect fragmented data, provide task-specific context, support multi-step reasoning, and make decisions more traceable. Examples from Cummins and Gilead Sciences illustrate how these capabilities can improve supply-chain forecasting and uncover indirect fraud networks, while the report concludes that stronger enterprise knowledge can increase trust, transparency, efficiency, and scalability for AI agents.
Oct 29, 2026 1,360 words in the original blog post.
Neo4j Laravel is an open-source package designed to integrate Neo4j graph databases into Laravel applications through familiar Laravel database conventions, allowing developers to use named connections, parameterized Cypher queries, the DB facade, and Eloquent-style models. It supports graph-focused relationship querying and management, including traversing connected nodes, loading related data, and operations such as attaching, syncing, detaching, and creating relationships. The package also includes practical application features such as transactions, pagination, soft deletes, connection configuration, query logging, and Laravel Debugbar support. Rather than replacing relational databases outright, it is intended to let Laravel applications use Neo4j alongside MySQL, PostgreSQL, or SQLite for recommendation systems, fraud detection, knowledge graphs, and other relationship-heavy workloads, while noting limitations for using Neo4j as Laravel’s default connection.
Oct 22, 2026 1,179 words in the original blog post.
Neo4j’s walkthrough demonstrates how Document Intelligence can convert 20 KYC documents into a knowledge graph and support beneficial-ownership investigations without manually defining extraction rules. In a fictional loan-screening case involving Anchor Point Partners LLC, Cypher ownership traversals reveal that guarantor Dorian Voss has two indirect ownership paths to the applicant despite not appearing on its cap table, including through Vantage Point Estates and through a chain involving Meridian Holdings, Northgate Ventures, and Cobalt Trading. Further graph analysis identifies a circular ownership structure among Meridian, Northgate, and Cobalt, while the Louvain community-detection algorithm separates Voss’s personal holding chain from the reciprocal-ownership core and from unrelated businesses. The post outlines how to create or restore the graph in Neo4j AuraDB, use an Aura Graph Analytics session for analysis and visualization, and explicitly remove temporary analytics resources afterward because they are billed while active.
Oct 22, 2026 2,452 words in the original blog post.
Neo4j’s Corporate KYC: Onboarding and Ownership Monitoring uses a graph-based approach to help financial institutions trace beneficial ownership through complex corporate structures, connect ownership positions with supporting documents and screening records, and identify material changes over time. In a synthetic case involving Meridian Trade Holdings, NorthStar Bank calculates effective ownership across multiple company layers, finding that Elena Park and Aleks Petrov meet a demonstration 25% threshold, while reviewed PEP and sanctions matches involving Marcus Lee and Petrov provide contextual evidence rather than automatic conclusions. The workflow also highlights shared device, IP address, and corporate-services-address connections among related entities as potential subjects for enhanced due diligence. When a new ownership declaration changes stakes, the system compares it with the accepted baseline, showing Marcus becoming a qualifying owner as Elena falls below the threshold and prompting renewed review of relevant screening evidence. Analysts can preserve their reasoning, task status, and evidence in a revision and generate a review package for subsequent authorized reviewers, while final customer approval or rejection remains the bank’s responsibility.
Oct 21, 2026 1,128 words in the original blog post.
Neo4j has announced Context Engine, a private-preview component of its Graph Intelligence Platform designed to help enterprise AI agents understand business knowledge, access authorized data, execute processes reliably, and provide explainable decisions. It addresses the “agent knowledge problem,” in which agents lack the organizational context embedded in application code, employee experience, and fragmented data systems. Its approach centers on a six-layer business ontology covering business processes, domain concepts, data products, physical systems, tools and integrations, and runtime knowledge, allowing agents to interpret terminology, identify relevant data, follow governance requirements, and learn from prior actions. The platform combines a Neo4j-based Business Ontology for Agents, Context Build and Runtime capabilities using governed GraphRAG retrieval, and Memory and Learning features that record decisions and reasoning paths over time. Context Studio provides a centralized interface for managing and governing the ontology, while organizations can retain their existing large language models and agent frameworks.
Oct 20, 2026 1,000 words in the original blog post.
A Neo4j and Graphicient blog post argues that tariff exposure analysis for forged industrial parts quickly becomes outdated because relevant information is dispersed across bills of materials, material certificates, supplier records, customs classifications, and traceability documents, while regulations and supply chains continually change. It critiques spreadsheet-based analyses, consulting studies, and BI dashboards as approaches that produce static answers for specific questions and may obscure data lineage or errors. The authors propose an ontology-driven knowledge graph approach in which business users model parts, materials, suppliers, programs, and their relationships; map existing files, databases, or data platforms to that model; and query the connected data in natural language or graph tools without first building new reports. They describe iterative refinement as a key benefit, allowing users to expose missing attributes or relationships and update the model rapidly. Examples using Neo4j Bloom demonstrate queries for the operational impact of a supplier failure and for a part’s end-to-end dependency chain, including materials, melt country, suppliers, programs, tariffs, and documentation. The post also presents Neo4j and Graphicient as graph-based platforms intended to improve industrial supply-chain visibility, traceability, compliance, and AI-supported decision-making.
Oct 19, 2026 2,860 words in the original blog post.
Agentic GraphRAG extends graph-based retrieval by allowing an AI agent to dynamically decide how to search, traverse, and update a knowledge graph rather than following a fixed retrieval path. Using iterative reasoning loops, agents can combine vector search with multi-step graph queries to answer complex questions, such as identifying customers affected by a shared issue, while preserving an explainable record of their retrieval decisions. Agents may also extract new entities and relationships from incoming information, resolve duplicates or conflicts, and update the graph under controlled permissions, validation rules, or human review. A context graph further incorporates conversations, tool calls, decisions, and outcomes, enabling agents to reuse prior operational experience alongside domain facts. The approach is most valuable when questions require unpredictable retrieval paths, a domain evolves with new concepts and relationships, or past decisions must guide future actions, while simpler fixed GraphRAG patterns may remain faster for predictable queries.
Oct 19, 2026 2,021 words in the original blog post.
Neo4j has released a preview Embeddings API in the Graph Data Science Python client 2.2+ for Aura Graph Analytics, allowing users to generate node embeddings with a single `gds.embedding.create` call. The default implementation uses the scalable FastRP algorithm to create vectors based on graph structure, supporting uses such as fraud detection, recommendations, segmentation, machine learning predictions, and retrieval for AI agents that combines graph and text embeddings. Generated embeddings remain as properties in the in-memory graph unless users choose to write them back to a connected database, and they can be streamed or used with algorithms such as k-nearest neighbors and k-means clustering. Users can customize dimensions, graph elements, property names, and random seeds, while safeguards prevent accidental overwriting of existing embeddings. For cases requiring node attributes or known labels, `gds.embedding.train` trains a GraphSAGE encoder using features and classifications, after which the saved model can generate task-specific embeddings through the same creation API. The preview is available through Aura Graph Analytics, including the AuraDB Free tier, supports constructing graphs from sources such as warehouses, Databricks, CSV files, and pandas DataFrames, and is expected to expand with additional capabilities and eventual Cypher API support.
Oct 16, 2026 1,514 words in the original blog post.
Neo4j’s Q3 2026 initiatives focused on making enterprise AI more secure, cost-efficient, and practical across Google Cloud, AWS, and Microsoft environments. The company reported expanded co-sell deployments involving property intelligence, AI-assisted codebase management, financial-services sales automation, and telecommunications security and churn prediction, with customers using Neo4j graphs to supply connected context for production AI systems. Neo4j also introduced Virtual Graph in public preview for Snowflake, Databricks, and Google BigQuery, allowing graph queries to run as optimized SQL against warehouse-resident data without copying or moving it. A reference architecture for Microsoft Copilot Studio, Okta, and Neo4j uses per-user authentication to enforce database-level access controls and reduce risks from shared AI-agent credentials. Benchmark results for Neocarta’s graph-based semantic layer indicated lower Text2SQL token use, improved query accuracy, and faster execution by retrieving only relevant catalog metadata and join paths. Looking toward Q4, Neo4j plans workshops, digital events, and joint field activities with cloud and data-platform partners to promote graph-grounded AI applications.
Oct 14, 2026 1,337 words in the original blog post.
Neo4j’s October 2026 community update highlights the upcoming NODES 2026 virtual conference on November 12, Road to NODES workshops, and several global events and meetups. Featured projects include agent orchestration using Neo4j Change Data Capture to trigger workflow handoffs, a GraphRAG system for U.S. levee infrastructure that reported 95.16% exact-match accuracy on its benchmark, and a real-time vehicle tracking architecture combining Neo4j, Databricks Lakebase, and OpenStreetMap. The update also covers Cogito Estella, an open-source, non-LLM tool for extracting text-based knowledge graphs into Neo4j, and BotMan-AI, which uses dependency graphs and agents to modernize legacy codebases. Community contributor Pratiksha Zalte is recognized for work on the Neo4j PHP Driver and will present at NODES, while Neo4j also promotes training resources, webinars, and its startup program offering eligible early-stage companies Aura credits and engineering support.
Oct 09, 2026 966 words in the original blog post.
AuraDB Business Critical and Virtual Dedicated Cloud now generally support multiple independent Neo4j databases within one instance, allowing organizations to consolidate workloads while preserving database-level separation and management. The feature targets SaaS multi-tenancy, agentic AI applications, departmental projects, workload segmentation, and nonproduction environments by enabling databases to share instance compute and storage rather than requiring separate instances. Each database can be created, managed, monitored, secured, backed up, restored, exported, and accessed independently through the Aura Console or API, while using tier-level security features and Neo4j role-based access control. Initially, instances support up to five databases per GB of RAM, with limits of 250 databases on Virtual Dedicated Cloud and 100 on Business Critical. Following a customer preview, the capability became generally available on September 30, 2026, and also helps self-managed Neo4j users migrate existing multi-database architectures to AuraDB without redesigning them.
Oct 08, 2026 923 words in the original blog post.
Modern government investigations often struggle not with insufficient data but with disconnected sources that delay analysts’ ability to identify meaningful relationships, a challenge Neo4j calls the “Investigation Gap.” The post argues that graph-powered intelligence analysis can connect entities such as people, organizations, locations, events, communications, financial records, and forensic evidence to reveal patterns, support multi-hop analysis, and produce explainable intelligence for decision-making. It promotes Neo4j GraphAware Intelligence Analysis, which combines Neo4j’s graph database, GraphAware Hume, and related services to help law enforcement agencies integrate evolving data sources while retaining flexibility and control. The company also highlights an on-demand webinar featuring Neo4j, GraphAware, and Western Australia Police, as well as its planned presence at the IACP 2026 conference in Orlando.
Oct 07, 2026 579 words in the original blog post.
Neo4j announced the upcoming general availability of Document Intelligence for AuraDB Free, Professional, and Business Critical tiers, enabling users to create knowledge graphs from text, PDF, and Word documents without code. Building on its preview release, the feature can process hundreds of documents through resilient background jobs, extract information from text, images, and tables, and connect new findings to existing AuraDB graph data through entity resolution. Users can define and refine extraction models with an interactive assistant and visual canvas, while sampled documents help identify recurring entities, properties, and relationships across a collection. The resulting graph combines extracted entity relationships with a lexical layer of document chunks and embeddings, allowing applications and AI agents to retrieve both connected business context and supporting source passages. Neo4j plans to add API, command-line, and Model Context Protocol access, along with improved controls for entity resolution, re-importing updated sources, testing, and broader enterprise deployment options.
Oct 06, 2026 1,120 words in the original blog post.
Neo4j describes a set of Neo4j Labs integrations intended to help developers connect AI agents built on platforms and frameworks such as AWS, Google, Vercel, Salesforce, LangGraph, and others to graph-based retrieval, actions, and persistent memory. The integrations use options including Neo4j’s MCP server, drivers, Aura Agents MCP, and Neo4j Agent Memory or its managed service, enabling agents to query and update graph data while retaining short-term, long-term, and reasoning context across sessions and among multiple agents. Examples include AWS Strands and AgentCore, Google ADK and Gemini Enterprise, Vercel AI SDK, and Salesforce Agentforce, with proposed applications in supply-chain analysis, customer assistance, and sales support. Neo4j says the integrations include runnable code, deployment and security guidance, and production-oriented setup materials in its GitHub repository and documentation, while future work is planned for CrewAI and Haystack 3.0.
Oct 06, 2026 1,575 words in the original blog post.
Tax filings and business registries may be accurate yet fail to reveal who actually controls a company, particularly in cases involving VAT carousel fraud, successor businesses, nominee directors, layered ownership structures, or entities split to remain below regulatory thresholds. Using Operation Admiral in Portugal as an example, the discussion explains how cross-border investigations can uncover extensive networks of companies and individuals whose relationships are not apparent from isolated records. It argues that investigators need both entity resolution to identify records referring to the same people or businesses and ownership-and-control tracing to examine links among companies, individuals, accounts, filings, and transactions without treating connected entities as a single taxpayer. The piece presents graph-based investigative knowledge layers, such as Neo4j’s platform, as a way to preserve distinct records while making relationships, evidence sources, dates, and uncertainty reviewable, reducing repeated manual work and allowing findings from one investigation to inform later cases while remaining open to revision.
Oct 01, 2026 1,495 words in the original blog post.