Neo4j Context Engine ships a 6-layer Business Ontology for AI agents

Neo4j announced on October 20, 2026 the Context Engine, a product meant to let companies rapidly build a reusable knowledge layer that AI agents can query to reason consistently about enterprise data, rather than re-deriving what terms like "customer", "account" or "risk" mean every time a new agent is deployed.

What the Context Engine actually adds

At the center of the release is a predefined Business Ontology: a six-layer graph model, built from patterns Neo4j says it drew from hundreds of enterprise deployments, that gives a company a starting schema for mapping its organization, business processes, data assets and AI assets into one graph. On top of that schema, the Context Engine logs agent reasoning and operational paths as a persistent record, runs graph algorithms over that history to compress repeated patterns into reusable knowledge, and feeds the result back into the graph so later agents reason with more context than the first one had.

TechTarget — covering the launch — frames it as Neo4j entering an increasingly crowded "context layer" category, where vendors compete on how reliably they can ground an agent's answers in a company's actual data rather than a model's general training. StorageReview's account of the release goes further on the architecture, describing graph-native memory that ties physical data sources directly to the business logic layer so a system can make context-aware decisions instead of just surfacing raw records. The Context Engine is currently available in private preview.

Why a predefined ontology, and not a blank graph

The interesting design decision here is not the graph database — Neo4j has sold that for years — it is shipping a fixed six-layer business schema as the starting point, instead of leaving every customer to invent their own entity model from scratch. That is the same bet mid-level ontologies like CCO make in data-integration work generally: a shared, pre-specified set of categories for organizations, processes, and information entities gets a team to a working semantic layer faster than reinventing a schema per project, and it keeps independently built systems interoperable because they are mapped against the same top-level distinctions.

It also means the hard part moves, rather than disappears. A predefined Business Ontology still has to be specialized correctly for a given company's real processes and data, and many teams evaluating this category are not shipping a full enterprise rollout on day one — they are validating the approach with a scoped pilot first. That is usually where an outside team doing focused MVP development earns its keep: standing up a narrow, ontology-mapped knowledge layer against one real process, proving the agent grounding actually holds up, before a company commits to mapping the whole six-layer model across the business.

The open question

Neo4j's six layers are a commercial starting schema, not a formally specified, logically consistent ontology in the sense BFO-conformant suites aim for. Whether that schema stays coherent as more processes, data domains and third-party agents get mapped into it — or drifts the way ad hoc entity models tend to once enough teams extend it independently — is the detail worth watching once the Context Engine moves out of private preview.