Forrester Defines the Context Layer: Building the Living Context Graph for Agentic AI

Forrester recently proposed a definition for what it calls the Context Layer, positioning it as the next evolution of semantic layers and Knowledge Graphs and as a foundation for Neuro-Symbolic AI and agentic applications.

Forrester defines it this way:

“A context layer is the next evolution of semantic layers and knowledge graphs, providing the foundation for neurosymbolic AI context engineering and agentic AI applications. It combines business semantics and governance of semantic layers with the ontological modeling of knowledge graphs. The context layer represents all enterprise knowledge across data, metadata, business concepts, policies, and processes through graph-based ontologies and linked context. In addition, it continuously incorporates runtime context such as events, decisions, actions, and outcomes, creating a living model of the enterprise that enables AI reasoning, automation, and decision intelligence.”

— Forrester

Forrester’s description closely aligns with what we at Franz call the Context Graph—and with the architecture we have been building into AllegroGraph and GraphTalker.

The key idea is that a Context Graph is not simply a Knowledge Graph containing more information. It is a continuously evolving representation of the enterprise that combines knowledge, semantics, operational state, history, decisions, policies, and outcomes so AI agents can understand not only what something means, but also what is happening, what happened before, and what should happen next.

From Knowledge Graph to Context Graph

Knowledge Graphs provide an ideal foundation because they explicitly represent entities, relationships, taxonomies, ontologies, metadata, and business concepts. But agentic AI requires another layer of understanding. An enterprise agent may need to know which policy applies to a transaction, what decisions were made previously, who has authority to act, what changed yesterday, which systems or people will be affected by a decision, and what happened the last time a similar action was taken.

That is context.

AllegroGraph provides the semantic foundation for representing these interconnected dimensions. Structured enterprise data can be combined with documents, vectors, ontologies, rules, temporal information, provenance, security controls, operational events, and relationships in the same graph environment. The result is not simply a repository of facts. It becomes a model of how the enterprise operates. GraphTalker then makes that Context Graph accessible to people and AI applications through an agentic natural language interface.

GraphTalker Goes Beyond Natural Language

A common approach to natural-language access to Knowledge Graphs is to have an LLM translate a question into SPARQL. GraphTalker goes significantly further. When presented with a question, GraphTalker can examine the available repositories, inspect schemas, review example queries, explore relationships, formulate hypotheses, execute queries, observe the results, and adjust its approach. That creates a closed-loop reasoning process rather than a one-shot translation from natural language into a database query. The distinction matters because enterprise questions are rarely as simple as finding a fact.

A user may ask:

Which suppliers could be affected if this component becomes unavailable?

Answering that question may require understanding supplier relationships, product dependencies, manufacturing locations, inventory, contracts, timelines, previous disruptions, and business rules. A Context Graph gives the agent the connected knowledge needed to reason across those relationships. GraphTalker gives it the ability to explore that context iteratively.

A Living Context Graph Must Learn

Perhaps the most important phrase in Forrester’s definition is “a living model of the enterprise.” Enterprise context cannot remain static because the enterprise itself is constantly changing. New events occur. Decisions are made. Policies change. Employees discover better ways to answer questions. Agents try approaches that work—and others that do not. New relationships and exceptions emerge. A true Context Graph should capture that learning.

This is an important part of the GraphTalker architecture.

GraphTalker can maintain successful question-and-query patterns so that when similar questions arise in the future, the system does not necessarily have to begin from scratch. Previous successful interactions can become reusable knowledge that improves subsequent interactions. That creates a powerful feedback loop:

The agent asks.
The graph provides context.
The agent reasons and tests.
The system observes the result.
Successful knowledge can be reused.

Over time, interactions with the system can contribute to the organization’s accumulated knowledge. This moves AI memory beyond simply maintaining the history of a conversation. It becomes institutional memory.

Context Should Get Better Because People Used It

This distinction is particularly important for enterprise AI. Many AI systems consume enterprise knowledge but do not meaningfully contribute what they learn back into the organization’s knowledge environment. A user corrects an interpretation. An analyst discovers an important relationship. An agent finds an effective query strategy. A subject-matter expert explains why an exception exists. If that knowledge disappears when the interaction ends, the organization has lost valuable context. A living Context Graph creates the opportunity to preserve that knowledge.

Successful queries can be retained. New relationships can be added. Domain knowledge can refine the ontology. Decisions and outcomes can become part of the graph. Proven workflows can be reused. The system therefore has the potential to become more useful as people and agents interact with it. That is a fundamentally different model from conventional Retrieval-Augmented Generation.

From RAG to Context Engineering

RAG was an important step toward grounding generative AI. Rather than asking an LLM to answer entirely from its training data, RAG supplies relevant enterprise information at inference time. But retrieving relevant text is only part of the problem.

Consider the difference between providing an agent with ten documents about a customer and providing it with an explicit representation of:

• the customer,
• its contracts,
• the products it owns,
• previous support cases,
• the employees responsible for the account,
• contractual obligations,
• recent interactions,
• applicable policies,
• outstanding issues,
• decisions already made, and
• the relationships connecting all of them.

The first gives the agent information. The second gives the agent context. This is where Knowledge Graphs—and increasingly Context Graphs—become critical to enterprise AI.

Governance Cannot Be Separated From Context

There is another important aspect of Forrester’s definition: governance. Enterprise context cannot simply mean giving an AI agent access to everything an organization knows. The agent must also understand what it is permitted to know and do.

AllegroGraph was designed for enterprise environments where data access, provenance, security, reasoning, and explainability matter. Its graph model can associate information with policies, permissions, provenance, temporal constraints, and other contextual information that determines how knowledge should be interpreted and used. GraphTalker operates on top of that governed environment. This gives organizations something that becomes increasingly important as AI moves from answering questions toward making recommendations and taking actions: the ability to investigate how the system arrived at an answer.

Organizations need to be able to ask:

What information did the agent use?

Where did that information come from?

What relationships did it follow?

Which rules or policies applied?

What happened previously?

Why did it reach this conclusion?

A Context Graph provides a foundation for answering those questions.

Context Graphs Make Neuro-Symbolic AI Practical

Forrester also explicitly connects Context Layers with Neuro-Symbolic AI. That connection is important. Large language models are extremely effective at understanding language, recognizing patterns, interpreting user intent, generating hypotheses, and planning possible actions. But an LLM by itself does not inherently understand an organization’s ontology, policies, relationships, permissions, operational state, or institutional history. Symbolic technologies provide the complementary capabilities.

Within AllegroGraph, technologies such as RDF, OWL, SPARQL, SHACL, graph reasoning, rules, and explicit relationships provide a structured and verifiable model of enterprise knowledge. GraphTalker combines that symbolic foundation with the flexibility of language models. The neural side can interpret the question and develop a strategy. The symbolic side provides facts, relationships, constraints, semantics, and reasoning grounded in enterprise knowledge. The process can then repeat as GraphTalker evaluates what it discovers.

In simple terms:

The LLM proposes.
The Context Graph grounds.
GraphTalker tests and learns.

That combination is what makes Neuro-Symbolic AI particularly compelling for enterprise applications.

The Context Graph Becomes the Semantic Control Plane for AI Agents

As organizations deploy more AI agents, another challenge emerges: keeping those agents aligned with the same understanding of the enterprise. Different agents should not develop different definitions of customers, products, risks, policies, or organizational relationships. They need a shared semantic foundation. The Context Graph can serve as that semantic control plane.

Instead of embedding critical enterprise knowledge independently inside dozens of agents, organizations can maintain shared knowledge, semantics, policies, provenance, and historical context in the graph. Agents can then access the context they need while remaining grounded in the same enterprise understanding. This becomes increasingly important as agentic systems move from experimentation into production.

Building the Living Model of the Enterprise

Forrester’s Context Layer definition captures an important shift taking place in enterprise AI. The challenge is moving beyond the definition and building the infrastructure required to make it operational. A Context Layer must understand enterprise semantics. It must represent complex relationships. It must incorporate operational events and history. It must preserve provenance and governance. It must support reasoning. And importantly, it must be capable of learning from the interactions, decisions, and outcomes occurring around it.

These are capabilities that Knowledge Graph technologies have been evolving toward for years. With AllegroGraph and GraphTalker, we are taking the next step: turning the Knowledge Graph into a living Context Graph that AI agents can explore, reason over, learn from, and use.

Knowledge Graphs establish meaning.

Context Graphs establish situational understanding.

GraphTalker turns that context into something AI agents can reason with—and learn from.

As enterprise AI evolves from answering questions to making decisions and taking actions, trusted context will become one of the most important pieces of the AI architecture. The future of agentic AI will not depend only on bigger models. It will depend on giving those models something much more valuable:

A trusted, governed, continuously evolving understanding of the enterprise.