July 2026 Summaries

AgentMemory for .NET is a newly developed, independent open-source project designed to provide persistent graph-native memory for AI agents within the .NET framework, serving as a counterpart to Neo4j Agent Memory. Created by José L. Latorre and supported by Neo4j Developer Advocate Zaid Zaim, this implementation is built from scratch and verified for compatibility using Neo4j's Test Compatibility Kit, achieving 178 out of 178 test scenarios successfully. Unlike existing Python and TypeScript SDKs, this .NET version integrates seamlessly with the Microsoft Agent Framework, offering three layers of memory—short-term, long-term, and reasoning—classified using the POLE+O ontology. It supports various retrieval methods, including vector and graph traversal, while incorporating .NET-native features such as OpenTelemetry and Semantic Kernel adapters. The project emphasizes the importance of native memory integration within agent frameworks to ensure consistent and reliable performance, and it is complemented by AgentEval, a toolkit for evaluating AI agent performance. The development team is also working on expanding backend options to include the Neo4j Agent Memory Service (NAMS) and plans to introduce new memory tiers for maintaining consistent agent behavior over time.

This week's Neo4j update dives into the importance of a shared, governed graph substrate, known as the Knowledge Layer, which addresses the challenge of scattered meaning in enterprise AI by providing a continuous querying platform for agents. It highlights a benchmark study demonstrating how agent steering can significantly improve recall and edge-attribute accuracy in knowledge graph extraction from pharmaceutical documents. The update also features a walkthrough of the Neo4j Agent Memory Service (NAMS), which utilizes three connected memory types supported by an Aura database to enhance agent functionality. Additionally, it offers insights into learning opportunities with Neo4j, including a masterclass on using Neo4j Aura and various workshops, as well as covering the launch of Velasight, a graph-native decision intelligence platform for commercial real estate. The update invites users to engage with Neo4j's development through its user research panel and explore upcoming events, such as live streams and conferences, while spotlighting community member Frédéric Valentin for his work on a Neo4j-backed legal agent.

Neo4j Virtual Graph, now in public preview, offers a zero-copy architecture that enables enterprises to transform existing data in warehouses like Snowflake, Databricks, and Google BigQuery into knowledge graphs without moving the data. This innovation allows users to leverage the power of knowledge graphs, which capture complex relationships better than flat retrieval methods, for tasks like agent workflows, batch enrichment, and analyst exploration. Users can generate a graph model using AI tools, adjust it, and execute Cypher queries, which are translated into optimized SQL, maintaining data governance and performance predictability. Virtual Graph serves workloads that can manage warehouse-grade latency, while native Neo4j is recommended for operations requiring millisecond response times. The public preview is free, with billing starting in September, and future developments include the ability to materialize virtual graphs into native Neo4j, federated queries, and support for additional data sources.

Public sector organizations are increasingly relying on AI to enhance efficiency, make faster decisions, and augment operational capabilities, especially in critical areas such as emergency response, defense, and healthcare. However, the effectiveness of AI is hindered by fragmented workforce data, necessitating a coherent architectural foundation known as the knowledge layer. This layer connects disparate data elements into a unified operational model, enabling AI to reason over complex workforce relationships, roles, and dependencies. The knowledge layer transforms static records into dynamic, connected intelligence that reflects real-time operational realities, allowing for more accurate and consistent decision-making. By establishing this foundation before deploying AI, organizations can avoid the pitfalls of automating flawed architectures and ensure that AI applications provide insights grounded in operational truth. The ultimate success of AI projects in the public sector relies not on the size of the AI models but on the strength of the underlying data architecture, which must be robust enough to reveal critical dependencies and enhance human judgment across various domains.

Agentic AI systems represent a significant advancement in artificial intelligence, offering both vast potential and notable risks, particularly in terms of reliability, as inaccuracies can lead to consequential errors such as filing wrong tickets or paying incorrect invoices. The National Innovation Centre for Data (NICD), in a study sponsored by Neo4j, found that the GraphRAG approach, which combines vector and graph RAG, significantly enhances the truthfulness and efficiency of AI agents compared to vector-only RAG. GraphRAG not only improved precision and recall but also reduced hallucinations and token usage, achieving a truthfulness score of 63 compared to 35 for vector-only RAG. This approach allowed AI agents to answer over twice as many questions, with a more than halved refusal rate for complex queries. The study demonstrated that organizations could improve AI agent reliability without extensive data projects by using existing resources like Wikipedia titles and sections. GraphRAG has thus become an essential tool for ensuring agent reliability, offering deep context and multi-hop reasoning for accurate, relevant, and explainable results.

MCP for Aura, introduced by Neo4j, streamlines the process for AI clients to access and interact with graph databases hosted on Neo4j Aura without the need for additional infrastructure setup. This hosted Model Context Protocol (MCP) server is integrated at no extra cost across Free, Professional, and Business Critical tiers, with Virtual Dedicated Cloud support forthcoming. It simplifies the connection process by eliminating the need for developers to manage separate servers and credentials, facilitating AI applications like Claude Desktop to query graph data seamlessly. Through a human-in-the-loop OAuth authentication, AI clients can access tools enabling schema exploration, read-only queries, and controlled read-write capabilities. This setup allows AI clients to interact with data securely, understanding the graph schema and executing queries without altering data unless explicitly permitted. The initiative is part of Neo4j's broader strategy to enhance AI's data interaction capabilities, offering an accessible framework for developers to integrate AI with graph databases efficiently.

AI agents hold the potential to revolutionize enterprise productivity by autonomously managing complex workflows and making decisions aligned with business objectives, yet they often falter due to fragmented context and lack of memory, leading to flawed decision-making. A context graph offers a solution by providing AI agents with a comprehensive, connected memory that encompasses business knowledge, conversation history, and decision traces, thus enabling decisions that are grounded, explainable, and improvable. The decision-making process of AI agents typically involves understanding goals, gathering context, deciding on the next step, acting, and learning, with each stage dependent on the previous one. Without a robust context graph, agents struggle with context fragmentation, lack of structure, absence of durable memory, and missing decision traces, leading to errors and inefficiencies. By utilizing a context graph, which connects long-term enterprise knowledge, short-term conversation history, and reasoning memory, AI agents can make more reliable decisions by understanding the full context of situations and learning from past outcomes. The context graph also enhances explainability and governance by allowing teams to inspect and debug the decisions made by agents. Through Neo4j's open-source library and its integration with various frameworks, organizations can implement context graphs to significantly improve the reliability and quality of AI decision-making.

Bloom, part of Neo4j’s suite of user tools, has expanded its capabilities to allow the execution of graph algorithms on the entire database, rather than just the current Bloom Scene, offering a more comprehensive view of data insights. This update enables users to run centrality and community detection algorithms across the entire database to uncover vital connections and groupings, enhancing the utility of graph algorithms in data analysis. Bloom provides a no-code interface for configuring and running these algorithms, allowing users to apply rule-based styling to visualize outcomes. With options to run algorithms on either a local scene or the entire graph, Bloom offers flexibility in data exploration, storing results locally or writing them back to the database for future reference. This functionality is available through Neo4j Instances that support the Graph Data Science plugin or the Aura Graph Analytics service, making it accessible without the need for coding. Users can initiate algorithm runs directly from the user interface while also having opportunities to fine-tune and schedule them outside Bloom using other Neo4j tools.

Jesús Barrasa's manifesto for context-rich enterprise AI emphasizes that failures in enterprise AI often stem from a lack of understanding of the business environment rather than issues with AI models or infrastructure. He introduces the concept of the Enterprise Knowledge Layer (EKL), a shared, governed substrate where organizational knowledge is stored, making it accessible to AI agents, tools, and applications. This layer integrates ontologies, grounding data, and memory, allowing agents to query and access authoritative, dynamic business knowledge rather than duplicating meanings across multiple agents, which leads to inconsistencies. The EKL ensures that enterprise AI systems are equipped with a unified understanding of business processes, relationships, and rules, facilitating scalable, intelligent actions. By encoding business knowledge in a centralized, queryable, and actionable manner, the EKL provides a foundation for integrating software development best practices into knowledge management, overcoming the limitations of traditional BI semantic and context layers. This approach not only enhances AI decision-making but also aligns business, data, and AI strategies under a common framework, emphasizing the crucial role of organizational knowledge as a competitive advantage in the AI era.

Public sector workforce management faces significant challenges due to outdated systems that cannot efficiently integrate data across HR, scheduling, security, and procurement platforms, resulting in delayed decision-making during critical times. As operational pressures become more concurrent and complex, the need for a cohesive understanding of workforce capabilities, clearances, and deployment readiness becomes crucial to maintaining national resilience and public safety. The OECD highlights these challenges, emphasizing the strain of aging workforces, shrinking talent pipelines, and simultaneous operational demands. To address these, Neo4j's Graph Intelligence Platform proposes a solution by creating a knowledge layer that unifies disparate data, allowing for real-time operational workforce intelligence. This connected intelligence enables leaders to make informed decisions quickly, identifying available personnel, prioritizing resource allocation, and assessing potential risks, thereby transforming traditional workforce management into a more dynamic and responsive system that anticipates and mitigates mission risks.

Neo4j is advancing enterprise AI by integrating its graph intelligence directly with cloud data platforms like Databricks and Snowflake, allowing organizations to leverage existing infrastructures without costly data migrations. The introduction of Neo4j Virtual Graph offers a zero-copy, zero-ETL solution for running Cypher queries and graph algorithms directly on cloud data, enhancing the speed and efficiency of deploying generative AI applications. Additionally, Neo4j's acquisition of GraphAware strengthens its security capabilities, enabling advanced investigative environments for detecting complex threats. The company's integration with Microsoft Fabric introduces an "Export to Lakehouse" feature, facilitating seamless synchronization of graph-enriched insights into Microsoft OneLake tables. Neo4j is also expanding its reach on Google Cloud with native integrations, supporting real-time workflows for AI applications. These efforts, combined with specialized reference architectures for Databricks and enhanced partner marketing initiatives, are positioning Neo4j as a leader in transforming raw cloud data into dynamic, action-oriented intelligence engines, providing significant value across various enterprise domains, including supply chain, IoT, commerce, and financial crime prevention.

At HackwithBay 3.0, held at the AWS Builder Loft in San Francisco, developers were challenged to create innovative solutions using Neo4j graph databases under the theme "Thoughtful Agents for Productivity." The event featured around 200 participants and resulted in approximately 50 submissions. The top three winning teams showcased unique applications that utilized Neo4j's capabilities effectively: VeriGraph transformed research papers into executable evidence by integrating claims and methods into a knowledge graph; Constructor served as an autonomous assistant for semiconductor research, dynamically evaluating techniques and updating a live graph with experiment results; and The Ultra Super Social Passport used the graph to connect event attendees based on shared interests and values. These projects highlighted the versatility and fast integration of Neo4j Aura in creating sophisticated, data-driven applications within a tight timeframe, supported by partners like RocketRide, Butterbase, and Daytona, which facilitated seamless integration and execution.

Neo4j's GraphAcademy has been revitalized with a new website and features, including the introduction of E.L.A.I.N.E, an AI learning assistant that helps users find courses and guides their learning journey. The platform now supports learning directly within integrated development environments via the GraphAcademy MCP server, allowing users to engage with Neo4j without leaving their editors. GraphAcademy offers both detailed courses and quick labs focused on specific skills, such as working with dates in Cypher or creating full-text search indexes. The initiative encourages community engagement through user research panels and workshops, offering opportunities for users to connect with Neo4j's product teams and to even lead workshops themselves. The platform also highlights successful community projects like Nivedita Thapa's Semantic Model Inspector, which assesses the readiness of semantic models for AI-powered analytics. Additionally, GraphAcademy emphasizes the application of graph technology in real-world scenarios, such as supply chain optimization demonstrated by EcocomityChain.AI, which uses a material genealogy graph to identify potential vulnerabilities in vehicle assembly chains.

Hybrid search in Neo4j is an advanced retrieval method that integrates lexical, semantic, and structural searches, enabling a comprehensive approach to finding relevant results by combining words, meanings, relationships, and structures in a single pipeline. This approach utilizes Cypher patterns and technologies such as Weighted Reciprocal Rank Fusion (WRRF) to re-rank results from multiple sources, ensuring a more precise retrieval than using any single search method. Lexical search focuses on exact vocabulary matches, semantic search identifies similar meanings even in different languages, and structural search examines graph contexts to determine relevance based on node relationships. By combining these methods, Neo4j's hybrid search provides a robust mechanism for tasks like support case investigations, allowing users to find related cases through a nuanced understanding of technical vocabulary, contextual similarities, and structural graph relationships. This approach is customizable, allowing teams to adjust weights and filters to better fit specific domain requirements, and demonstrates the unique value graph structures bring to search by not only improving post-search context but also guiding what should be retrieved initially.

The Aura Agent Hackathon, as recounted by Ari Waller, demonstrated the innovative applications of knowledge graphs in AI, with developers worldwide using Neo4j's new generation of Aura Agents to tackle real-world challenges across various sectors, from healthcare to finance and beyond. The competition saw 201 course completions and 40 AI agent submissions, emphasizing the significance of using graphs for intelligence rather than mere data storage. The standout projects included ConspiracyGraph Agent for misinformation analysis, RegulatoryRisk GraphAgent for financial compliance, and Korca Triage Agent for optimizing support ticket routing. These projects highlighted how knowledge graphs can seamlessly integrate with AI to provide context, enhance reasoning, and uncover complex relationships, proving that as AI systems evolve, the success of such technologies will heavily depend on sophisticated data architecture. The hackathon underscored the potential of graph intelligence in transforming diverse industries, setting a precedent for future innovations in graph-based AI solutions.

Christoffer Bergman discusses Neo4j's Virtual Graph feature, which allows users to treat data in relational databases like BigQuery as if it were native graph data, facilitating the use of Cypher queries and Graph Data Science algorithms without data extraction, transformation, and loading (ETL). By using Neo4j's Virtual Graphs, users can express graph queries and visualize data as graphs while maintaining data in its original relational format, albeit with some limitations in execution speed due to the underlying relational engine. Bergman uses the example of calculating the "Bacon number" of Swedish actor Björn Bengtsson through a series of database operations to illustrate the benefits and process of implementing Virtual Graphs. He explains that while Virtual Graphs provide a convenient and cost-effective way to explore graph data, they might not match the performance of native graph databases for real-time applications or complex traversals, but they offer an excellent starting point for users with existing data in relational warehouses.

In the rapidly evolving field of AI and graph technology, Neo4j has launched an updated GraphAcademy to enhance learning experiences with hands-on modules and collaborative tools. The revamped Neo4j Fundamentals course now allows users to design data models directly through a browser-based drag-and-drop interface, fostering a deeper understanding through active creation rather than passive learning. Users can continuously refine their data models, aided by an AI chatbot offering expert advice. The platform also emphasizes teamwork by introducing features that enable learners to form groups, track progress, compete in leaderboards, and maintain communication through a shared news feed. An onboarding assistant further personalizes the learning journey by suggesting courses based on individual goals and experience, while public profiles allow users to showcase their achievements. These features signify the beginning of ongoing enhancements to make graph learning more engaging and accessible.

AI-ready data is crucial for the success of enterprise AI projects, as it allows AI systems to reason, decide, and act effectively by providing data that is contextual, flexible, and standardized. Unlike traditional data infrastructures that are static and siloed, AI-ready data requires a knowledge layer that connects disparate data sources, enabling AI to navigate and reason about data connections. This knowledge layer, supported by components like knowledge graphs, context graphs, and GraphRAG, helps create a comprehensive framework where AI can access reliable and interconnected information. Such a structure facilitates more accurate, explainable, and governable AI outcomes. By implementing a knowledge layer, organizations can overcome common challenges in AI projects, such as data fragmentation and lack of context, ultimately enhancing AI performance and adoption. Case studies of companies like Klarna, Data², and Cummins demonstrate the effectiveness of leveraging a knowledge layer to drive successful AI initiatives in diverse industries.