What is a context graph?
A context graph is a company's knowledge graph. It connects records, relationships, human judgment, decisions and outcomes, so the reason behind a number sits next to the number. When a team asks a question, AI can start from what the business already knows instead of asking people to explain the background again.
What it holds
Most companies keep each part of a decision in a different place. The finance system has the cost. The CRM has the customer. A manager's reasoning lives in an email thread. The result shows up months later in a report nobody links back. A context graph joins those pieces:
- Records from the systems the company chooses, structured and unstructured
- Relationships between people, accounts, contracts, workflows and teams
- Human judgment, such as a correction, an exception or the reason an option was declined
- Decisions and the evidence each one used, kept as Decision Traces
- Outcomes measured afterward, including failed and inconclusive ones
How Nodes uses it
Nodes Engine builds the context graph from authorized information across data silos and relevant external sources. Engine then uses it to investigate a question, compare options and carry out approved work. The company chooses the sources and how often they are checked. It also defines what each person and agent may read, recommend and change.
Shared does not mean open to everyone
A finance workflow and a hiring workflow can draw on the same graph without either one seeing every record. Each person and agent uses only what its permissions allow. Connecting information does not widen anyone's access or authority.
Why it outlasts the model
Models, LLMs and agents will change. The company keeps its context graph, its business rules and its decision history, so a new model can use what the business already knows. A replacement model is still evaluated for task quality, latency, risk and total AI and human cost before it takes over the work.
What it does not prove
Storing a result is different from learning from it. A lesson from one decision needs evaluation before it is reused in another, and a measured result does not by itself prove the change caused it. Keeping the source, the date it was true and where a lesson applies lets a reviewer check.
Related terms
- Decision Traces keep the record of each decision.
- Approval controls decide which actions wait for a person.
- Workflow orchestration moves the approved work across systems.