December 2025 Summaries

In a detailed exploration of Neo4j's new AI procedures in Cypher, Christoffer Bergman illustrates how these enhancements can be applied to complex cases, such as the unsolved murder of Swedish Prime Minister Olof Palme. The blog post describes the transition from requiring external functions for language model integration to having native AI functions incorporated directly within Cypher, thus removing the need for external support. Bergman provides an in-depth walkthrough of how to employ these new capabilities, including the use of vector embeddings and text completions, to analyze large datasets like the extensive public records of the Palme investigation. The process involves embedding questions as vectors, performing vector searches for relevant documents, and utilizing graph traversal techniques to collect contextual data, which is then used in querying language models for insights. While the exercise did not conclusively solve the Palme case, it demonstrated the potential of GraphRAG and AI-enhanced graph databases to provide valuable investigative insights, even suggesting that with full access to digitized, uncensored documents, significant advancements in such investigations could be made.

Context engineering in AI involves designing and delivering structured, relevant information that enables large language models (LLMs) to perform tasks more reliably and effectively, especially in complex applications like agentic systems. While prompt engineering focuses on crafting the questions or commands given to an AI model, context engineering emphasizes providing the necessary background information, tools, and constraints that guide the model's reasoning process. This approach involves clear task definitions, managing memory, and leveraging tools such as knowledge graphs to retrieve and organize the necessary data. By organizing the context into layers and focusing on the minimum viable context, AI systems can reduce errors, enhance multi-step reasoning, and ensure decisions are explainable and auditable. Context engineering, supported by technologies like Neo4j's GraphRAG, is essential for building intelligent, mission-critical agents that can handle complex workflows and provide consistent, grounded outputs.

Christoffer Bergman, Director of Engineering at Neo4j, explores the use of AI agents to enhance debugging processes for cash-handling devices, inspired by the capabilities of ChatGPT. Reflecting on his previous job in the cash management industry, Bergman describes how log files from devices such as ATMs and smart safes, which use Java-based event-driven software, are crucial yet cumbersome for debugging due to their size and complexity. By utilizing Neo4j's Aura, a cloud-based service, he simulates a data model with generated events and log files to illustrate how AI agents can identify software anomalies by analyzing patterns and discrepancies in transaction logs. In a case study, he demonstrates how an AI agent, configured with specific tools and prompts, can effectively detect a bug related to an unexpected behavior in a bill validator, leading to cash discrepancies. Bergman concludes by proposing further enhancements for integrating AI agents with source code and extending their capabilities for broader applications across event-based systems.

In the latest edition of "This Week in Neo4j," the focus is on various developments in the world of graph databases, including the completion of the Advent of Code 2025 using Cypher, and the announcement of the upcoming NODES AI conference, which will feature themes like GraphRAG and Graph Memory. The newsletter highlights the integration of LLMs with knowledge graphs for more reliable AI systems, the benefits of Neo4j's Fleet Manager for centralized monitoring and management of graph workloads, and the use of multi-layered ontologies to enhance legal advice provided by AI. It also emphasizes the importance of semantic tool discovery for AI agents working with extensive MCP tools to improve efficiency and accuracy. Featured community member Alessandro Negro discusses building graph-based memory systems, and the newsletter invites readers to various Neo4j events and workshops aimed at enhancing their graph technology skills.

Context engineering in AI focuses on providing large language models (LLMs) with the right instructions, tools, and evidence at the appropriate time to ensure reliable task completion, particularly in complex, multi-step applications. While prompt engineering shapes the initial queries, it is insufficient for advanced applications such as agents that require structured context, including task definitions, relevant information, and tool access to maintain coherence and avoid model failures. Context engineering involves designing, storing, and retrieving this information through frameworks like Neo4j, which utilizes knowledge graphs to create explainable and auditable AI workflows. By organizing context into layers—from stable system prompts to dynamic user queries—context engineering enhances reliability, reduces prediction errors, and improves decision-making transparency. This approach also emphasizes the importance of meta-context as a governance layer to maintain compliance and scope across various phases of the agent workflow. The use of knowledge graphs aids in structuring memory and reasoning, enabling the efficient assembly of relevant information and providing a robust framework to support AI agents in executing complex tasks with greater accuracy and governance.

In a project aiming to extend Neo4j's capabilities, a user-defined function (UDF) is developed to verify if an IP address belongs to a specific network segment, addressing the lack of native Classless Inter-Domain Routing (CIDR) validation in Neo4j. This UDF is particularly useful for network management applications, enabling developers to perform IP allocation verification within graph databases. The demonstration includes practical examples of querying network segments using Cypher, integrating the UDF with the Neo4j MCP server, and utilizing a language model like Claude Desktop to conduct detailed network analyses. This integration allows for tracking connected devices, identifying security issues, and providing actionable recommendations, ultimately enhancing network management and aiding informed decision-making based on comprehensive data analysis. The project showcases the potential of combining Neo4j, UDFs, and advanced machine learning tools to improve network management capabilities.

Model Context Protocol (MCP) is a universal protocol developed by Anthropic to seamlessly integrate AI models with external data sources, tools, infrastructure, and APIs, thereby enhancing GenAI applications by enabling a consistent and efficient way to plug into business systems, developer tools, and cloud platforms. By eliminating the need for custom, ad-hoc API integrations, MCP simplifies the process of connecting large language models (LLMs) to new services, allowing for richer, system-aware AI behaviors. This approach mirrors the impact of standards like HTTP and REST, fostering a growing ecosystem where MCP servers can interact with AI clients such as Claude, Cursor, and VS Code, enabling them to access databases, cloud services, and application APIs. MCP's architecture comprises the MCP host, client, and server, which work together to manage communication, tool discovery, and execution, providing a single protocol that facilitates structured data retrieval, API queries, and cloud service actions. Despite its growing adoption and potential to transform AI workflows, MCP faces challenges related to security, observability, and discovery, but ongoing developments aim to address these issues and improve its integration into the broader agent ecosystem.

The blog features a variety of articles centered around graph databases, highlighting their applications and advancements. Topics include the use of graph databases in supply chain management for improved sustainability, the potential benefits of connected data in enhancing decision-making within the US government, and the innovative property sharding technique in Infinigraph for scaling rich graph databases. Additionally, Neo4j's new architecture is discussed for its ability to unify transactional and operational workloads. The series also delves into the Cypher and GQL developer topics, exploring the Neo4j Spatial tool with a focus on path intersections using AIS data, layer management, spatial filtering, and basic querying. Overall, the blog provides insights into the diverse uses and technological developments of graph databases.

The text outlines a variety of articles and interviews focused on the application of graph technology across different sectors, particularly in network and IT operations. Key topics include the use of graphs for enhancing customer insights, fraud detection, and GDPR compliance, as well as their role in empowering IT operations management and driving fast innovation through lean graph data models. It also features real-world case studies and expert interviews, such as with David Fox from Adobe, discussing the advantages of graph databases in handling complex data relationships and facilitating advancements in machine learning and data analytics.

A diverse range of topics related to supply chain management is explored, highlighting the critical role of technology and data analytics in optimizing operations, enhancing sustainability, and mitigating risks. The articles delve into how graph databases and predictive analytics, including GenAI and Neo4j Aura Graph Analytics, can revolutionize supply chain dynamics by improving efficiency, risk management, and forecasting accuracy. The content underscores the importance of understanding and implementing various types of supply chain management software to enhance decision-making and ensure a robust, future-proof supply network. Additionally, real-world case studies provide insights into practical applications of these technologies, showcasing their potential to transform supply chain practices across industries.

Supply chain management faces significant challenges due to evolving regulations, disruptions in raw material sourcing, and shifting customer demands, which can lead to costly inefficiencies. Traditional rule-based systems often fail to provide the necessary visibility and context, as they focus on symptoms rather than root causes, resulting in companies remaining in a reactive mode. A graph database approach offers a solution by providing real-time visibility into the interconnected nature of supply chains, modeling them as a supplier graph that reflects the actual operations and dependencies. This comprehensive view enables organizations to identify potential bottlenecks and risks, improve coordination across procurement, planning, and logistics, and ensure a more resilient supply chain. Notable examples include BASF, which used a supplier graph to navigate the 2022 EU energy crisis, J.B. Hunt Transport Services, which optimized routes to reduce carbon footprint, and the U.S. Army, which enhanced inventory planning and decision-making speed. By extending this graph-based approach to other enterprise domains, companies can achieve a unified view of their operations, improve risk analysis, and make more informed, agile decisions.

Kapil Hetamsaria has joined Neo4j as the Chief Business Officer for Strategic Partnerships and Alliances, aiming to expand the company's global partner ecosystem and enhance enterprise adoption of its graph intelligence platform. With extensive experience from companies like McKinsey, Microsoft, and C3 AI, Hetamsaria is dedicated to prioritizing long-term customer value and fostering strategic co-creation with partners across various technology landscapes, including hyperscalers, data platforms, and AI application vendors. His vision involves transforming Neo4j's partner program into a central component of its go-to-market strategy, driving enterprise growth by providing tailored solutions that meet the rising global demand for connected data intelligence. Neo4j's current strong partnerships, such as those with AWS, Google Cloud, and Deloitte, serve as a foundation for this expansion, with successful implementations like Gilead Sciences' fraud detection and Uber's knowledge graphs exemplifying the platform's impact. Hetamsaria likens Neo4j to a diamond, emphasizing its role in converting data into actionable knowledge for intelligent applications and AI systems, and he is committed to broadening its reach as an indispensable asset for leading organizations worldwide.

Marco De Luca and Jelmer de Reus explore how integrating IT infrastructure, service, and security operations data can yield actionable insights into an organization's IT environment, focusing on the challenges and solutions of identifying and managing software vulnerabilities like Log4J. They highlight the importance of using comprehensive tools and processes, such as Software Composition Analysis (SCA) and Static Application Security Testing (SAST), to analyze software risks and the necessity of a connected data model that includes metadata and organizational structures. The authors present a case study involving a large government body, illustrating the use of graph databases to visualize and manage software vulnerabilities effectively. They emphasize the value of dashboards that provide an overview of software components and their vulnerabilities, enabling organizations to identify critical risks and the responsible teams efficiently. The article also discusses the potential of further expanding these data models with information about developers and teams, stressing the importance of securing sensitive data with role-based or attribute-based access control.

Neo4j has advanced its collaboration with Confluent by launching the Managed Neo4j Sink Connector as a fully managed service on Confluent Cloud's Kafka Connect, eliminating the need for customers to manually upload, create, or manage the connector themselves. Initially introduced as a Custom Connector supported by Neo4j for use in Confluent Cloud in 2023, this new managed service now allows enterprises to securely transfer data between cloud services, supporting private endpoints on Azure, AWS, and Google Cloud Platform. Customers can handle support issues directly with Confluent, with Neo4j providing assistance when necessary. The Managed Connector is accessible via the Confluent Marketplace, enabling users to sink data from Kafka Topics to Neo4j Aura databases or utilize Neo4j as a source of messages for Kafka Topics, with the option to use CDC or Cypher queries for detecting changes.

Neo4j's newly launched Fleet Manager is a centralized control plane designed to manage Neo4j databases across various environments, including cloud, on-premises, and hybrid setups. It provides comprehensive visibility over database instances, unified policy enforcement, and automated workflows for tasks such as provisioning, upgrades, and migrations. Fleet Manager supports multiple Neo4j editions, allowing for seamless integration and monitoring through the Aura console, where users can view database topology, performance metrics, and receive security advisories. The tool is beneficial for different organizational roles, offering database architects and engineers centralized management and developers access to built-in tools without administrative permissions. Fleet Manager encourages users to convert connections into fully monitored deployments to avoid configuration drift and improve response to outages. It includes features like database registration, topology views, and performance metrics, regardless of whether users are on the Enterprise or Community Edition. The service is accessible via a free Aura account, emphasizing collaboration with users to refine and enhance the platform's capabilities.

NODES 2025 was a 24-hour virtual event that gathered thousands of graph developers and data enthusiasts to participate in over 140 technical talks centered around AI engineering, graph-powered applications, data intelligence, and knowledge graphs. The event featured prominent keynotes from industry leaders such as Andrew Ng and Emil Eifrem, who discussed the future of agentic AI systems and the importance of clean, graph-shaped knowledge for reliability. Sessions covered a variety of topics, including the use of Neo4j for creating live architecture graphs, improving customer service through multimodal content understanding, and applying graph neural networks for fraud detection. Other presentations demonstrated the practical applications of knowledge graphs in fields like life sciences, supply chain, and cybersecurity, highlighting the versatility and impact of graph technologies in modern data solutions. The next iteration, NODES AI, is scheduled for April 15, with a focus on knowledge graphs, graph memory and agents, and graph + AI in production.

Neo4j Fleet Manager is introduced as a centralized control solution designed to manage and monitor Neo4j databases across diverse environments, including cloud, hybrid, and on-premises. It addresses challenges faced by organizations with multiple Neo4j deployments by providing fleet-wide visibility, unified policy enforcement, and automated workflows for tasks such as provisioning, upgrades, and migrations. The tool offers built-in observability and security features like console activity logging and role-based access control (RBAC), ensuring secure and efficient database operations. By consolidating management under a single interface, Fleet Manager enhances the performance, resilience, and uptime of Neo4j deployments, while also simplifying administrative tasks through features such as real-time monitoring, automated backups, and guided version upgrades. The first release emphasizes foundational visibility and management, encouraging user feedback to further evolve the platform's capabilities.

The blog post discusses building modular graph analytics pipelines using Neo4j Aura Graph Analytics, a serverless cloud offering that can execute graph algorithms without relying on a Neo4j database. It highlights the ability to integrate external data sources, such as relational databases or CSV files, through the use of existing MCP servers like Supabase, thereby avoiding the need for individual integrations for each data provider. The post explores challenges in connecting large language models (LLMs) with external data sources while keeping raw data out of the LLM context to prevent overwhelming it. By grouping graph algorithms semantically, such as centrality and community detection, the system allows LLM agents to fetch data, run graph algorithms, and receive insights without directly handling data rows, thereby maintaining scalability and efficiency. The approach is demonstrated through a proof of concept where data from Supabase is used to run graph algorithms, showcasing the versatility and efficiency of dynamically mounting different MCP servers for various data sources.

Neo4j's latest updates highlight the general availability of the Neo4j MCP Server, facilitating easier connections with Claude, VS Code, and other MCP hosts for various Neo4j deployments. The introduction of a native VECTOR data type promises to streamline application code by storing embedding vectors as fixed-length properties, enhancing safety and unlocking potential optimizations. The GraphAcademy is offering a new course on constructing knowledge graphs using the Neo4j GraphRAG package for Python, which allows for building enriched knowledge graphs from unstructured text and creating efficient retrievers. The community spotlight features AI engineer Vaibhava Lakshmi Ravideshik, who has contributed significantly to the development of adaptive GraphRAG architectures. Upcoming events include virtual conferences and livestreams, while the article also discusses deploying Neo4j securely on a VPS with guidance from Will Lyon.

In Microsoft Fabric, OneLake tables act as a centralized data source, ensuring consistency and trust across organizations, which is vital for effective data operations. The integration of Neo4j graph intelligence with OneLake can enhance retail recommendations by transforming these tables into connected graphs and applying graph algorithms to uncover relationships and scores. This process is demonstrated using a grocery store scenario, where personalized recommendations are improved by analyzing co-purchase patterns and node similarity rather than just frequency of purchase. The guide details setting up an AuraDB instance and utilizing graph models to create a recommendation system that filters out generic associations and highlights personalized product suggestions. This approach offers a more nuanced understanding of customer preferences, moving beyond simple co-purchase data to deliver recommendations that are both interesting and truly relevant to the customer.

Knowledge graphs provide a structured representation of real-world entities and their relationships, facilitating a comprehensive understanding of information. Constructing knowledge graphs from unstructured data involves complex processes such as data querying, cleansing, and transformation, which can be automated using text analysis capabilities of large language models (LLMs). The Neo4j GraphRAG for Python package, particularly the SimpleKGPipeline class, offers an efficient pipeline for creating knowledge graphs by loading text, splitting it into chunks, creating embeddings, extracting entities using an LLM, and writing the results to a Neo4j database. This process requires a Neo4j connection, an embedding model, and an LLM to convert documents into a knowledge graph. Neo4j GraphAcademy provides a course on using and customizing the SimpleKGBuilder, which covers the creation of text splitters, custom data loaders, schema definition for lexical graphs, and integration of structured and unstructured data into GraphRAG pipelines. The course is part of a broader offering from Neo4j GraphAcademy, which includes various free courses on Neo4j fundamentals and advanced applications.

The Neo4j Connector for AWS Glue streamlines the process of transforming and loading tabular data into graph databases within the AWS cloud, particularly for projects using Neo4j AuraDB or self-hosted Neo4j instances on Amazon EC2. By leveraging AWS Glue's serverless capabilities, data engineers can efficiently convert data from various Amazon services and popular data stores into Neo4j, using the Neo4j JDBC Driver to translate SQL commands into Cypher statements for optimal graph creation performance. The connector is especially useful for managing rich datasets, such as Transport for London's cycle data, allowing for the creation of detailed graphs that map stations, trips, and areas. This process involves setting up nodes and relationships, using best practices for data transformation, and ensuring compatibility with AWS Glue's schema requirements. Through visual ETL jobs, the connector facilitates the creation and synchronization of analytics and operational graph stores, enhancing the capability to perform route analytics, anomaly detection, and complex queries.

Neo4j has reinforced its collaboration with AWS by achieving three new AWS distinctions in 2025: AWS Government Competency, AWS Life Sciences Competency, and AWS Agentic AI Specialization, enhancing its total to seven AWS competencies. These achievements underline Neo4j's prowess in delivering AI, analytics, and mission-critical applications, demonstrating its broad industry coverage and technical excellence. The company has been recognized as an inaugural partner for the AWS Agentic AI Specialization, highlighting its leadership in advanced AI methodologies such as GraphRAG and knowledge graphs that provide context-aware and explainable data. Neo4j's solutions are employed by government agencies for secure and scalable data analytics, exemplified by the U.S. Army's use of Neo4j for operational planning and decision support. In the life sciences sector, Neo4j has enabled organizations like Gilead Sciences to unify complex research data, thereby accelerating drug discovery and research insights. Additionally, Neo4j introduces new AWS-aligned innovations, including a pay-as-you-go listing for AuraDB on AWS Marketplace and a Kubernetes Operator for Amazon EKS, aimed at simplifying graph technology adoption for AI and analytics.