Context Graph (Gemini)
A "context" refers to the information, background, and data environment an LLM or AI agent uses to understand a prompt and generate a response.
A "context graph" is a specialized, dynamic record that organizes information such as data, documents, decisions, and policies into a structured network that can be queried.
"Business Context": Refers to the web of internal wikis, Slack threads, decision precedents, and other local knowledge that helps managers and employees do their jobs.
A context graph contains elements such as:
Nodes: Discrete entities (e.g., a specific customer, a software file, an employee, or an invoice).
Edges: The relationships between nodes (e.g., "author of," "depends on," or "escalated to").
Temporal Context: Timestamps track when a relationship began, when it changed, or whether a policy is currently valid.
Decision Traces: Operational data that records precedents such as why a past decision was made, what policy exceptions were granted, and who approved them.
Why are Context Graphs Important
Without context, an AI agent operates and applies rules rigidly. By querying a context graph, an AI agent can:
Understand the "Why": It can learn from past company precedents rather than just relying on generic training data.
Maintain Compliance: It respects historical exceptions, security permissions, and governance rules automatically.
Avoid Hallucinations: It roots its responses directly in the company's verified institutional history.
Core Architecture
Traditional RAG: Documents are broken into flat text chunks and converted into vector embeddings. Retrieval is based on mathematical proximity to the user's query.
Context Graph: Data is structured into nodes (entities, people, documents) and edges (relationships, rules, and temporal states). Retrieval involves deterministic multi-hop traversing instead of semantic guesswork.
Retrieval & Reasoning
Companies combine RAG (Retrieval-Augmented Generation) and Context Graphs (Knowledge Graphs) to overcome the limitations of using either system alone.
Using them together often referred to as GraphRAG provides structural reasoning and explicit connections that standard semantic RAG cannot achieve.
Why Combine Both Technologies?
Advanced Multi-Hop Reasoning: Context graphs allow AI to trace connections across multiple steps (e.g., Customer A is linked to Project B, which uses Technology C, provided by Vendor D). Standard vector RAG struggles with this.
Governed and Accurate Context: Context graphs apply business rules, hierarchies, and security policies. They ensure the AI only uses verifiable, temporally valid, and permitted information.
Improved Explainability: Because answers follow a traceable map of nodes and relationships, the AI can cite the exact concepts and connections it used, rather than relying on an unexplainable LLM "guess".
How They Work Together
RAG acts as the broad search layer, scanning massive amounts of unstructured data (like PDFs, emails, or chat logs) using semantic similarity.
Context Graphs act as the structural framework, organizing core entities, explicit business logic, and definitions.
The Result: The system retrieves relevant document chunks from the RAG pipeline, but uses the context graph to map how the entities in those documents relate to the user's query, resulting in precise, context-rich answers.
Note: GraphRAG does not seem to perform the services of a CMDM, asset management, provisioning tool.
Context Graph: Data is structured into nodes (entities, people, documents) and edges (relationships, rules, and temporal states). Retrieval involves deterministic multi-hop traversing instead of semantic guesswork.