Neo4j launches Context Engine to help enterprise AI agents act reliably

The agent knowledge problem

According to Neo4j, AI agents are "an entirely new kind of data consumer". For decades only applications and people consumed enterprise data, and its meaning lived in code and in human memory. The company gives an example: an application knows that an account status of "2" means "suspended pending review" because it is written in the code, while a person may know it from experience or a BI tool. Agents have no such knowledge.

Neo4j adds that business terms vary between organizations. "Customer" might mean one thing in Marketing and another in Finance. Fragmented data makes this worse, because agents often cannot reach the knowledge they need. Neo4j calls this the agent knowledge problem and describes it as "the trillion-dollar AI bottleneck", a claim that comes from the company itself.

Six-layer ontology at the core

Context Engine is built around an interconnected six-layer business ontology. Neo4j says it lets agents understand business concepts and how data entities relate, find the right data across clouds, platforms and vendors, and learn from every decision. The graph explicitly models relationships, so agents can reason across them and show how and why they reached an answer.

• Layers 1 and 2: the business process and domain layers. The first describes the rules and processes that determine how the organization operates, the second defines key terms and concepts.
• Layers 3, 4 and 5: the data products, physical systems, and tools and integration layers. They determine what information can be used, where it lives and how to connect to it.
• Layer 6: runtime knowledge, which lets agents store past decisions and actions and learn from them over time.

Three core components

Neo4j says Context Engine has three components that create and manage the ontology.

Business Ontology for Agents is a graph model that an enterprise defines, stores and enforces in Neo4j. It captures what business concepts mean, how processes work and how they relate to each other.

Context Build and Runtime lets agents access information wherever data lives. Built-in governance is meant to ensure agents can only access data they are authorized to see. Retrieval uses GraphRAG, which Neo4j describes as more accurate and cost-effective than vector-only RAG.

Memory and Learning connects short-term, long-term and reasoning memory in the same graph as business meaning. Every agent decision and decision path is tracked and stored, so agents can learn from what they do.

The product also includes Context Studio, a UI for managing and governing the business ontology from a single control plane. Neo4j says the components are designed to work together so teams spend less time building and maintaining fragmented context infrastructure.

Availability and what it means for teams

Neo4j says companies can keep the LLM and agent framework they already use, while, in its words, "dramatically reducing the risk of agent failure". That is a company claim, and the announcement gives no performance figures.

Context Engine is in private preview. Neo4j invites interested companies to become design partners, and those who want to get started can contact its sales team. No pricing or general availability date was given.