Context graph
A knowledge graph extended for AI agents. Besides things and relationships, it stores decisions, policies, exceptions, precedents, and outcomes as connected nodes, each marked with when it was valid. The point is the "why" behind a decision, not just the "what," so an agent can look up how a similar case was handled and whether that rule still holds.
In plain English
A knowledge graph is a map of facts: things and how they connect. A context graph adds the sticky notes: “we made an exception for this client in March,” “this discount policy ended in June,” “last time this happened we did X.” Then it dates every note so the AI knows which ones still apply.1
In the real world
- A support agent that knows a refund was a one-time override, not standing policy, because the graph says so.1
- An approvals assistant that cites the precedent it followed.
- An approvals assistant that says “approved under the March exception, which ended in June.” That sentence is a context graph talking.
- Vendor decks. This is where you’ll meet the term first.
When to use it
- An agent has to remember decisions across sessions, not just facts.
- Someone will ask “why did the AI do that,” and you need an answer you can show.1
When not to use it
- You don’t have a knowledge graph yet. This builds on one.
- Your agent does one-off tasks with no history to consult. Plain context engineering is enough.
Good, bad, ugly
- Good: the AI can tell a rule from an exception and a current fact from an expired one.
- Bad: someone has to record decisions as they happen, with dates. That is a process change, not a software purchase.
- Ugly: the term is six months old and every graph vendor has adopted it. Read past the label.
Key facts
- Adds to a knowledge graph
- Decisions, policies, exceptions, and outcomes, each with the dates it applied
- Status
- Early. The consultancy Thoughtworks listed it in April 2026 as worth a look, not yet proven.
- Built for
- Agents that need to remember past decisions, and teams that need a record of why the agent did what it did
Common questions
What's the difference from a knowledge graph?
A knowledge graph says what is true. A context graph adds why it was decided, by whom, under which policy, and for how long. Your normal systems record what happened; this records why.
Why does the time part matter?
Facts expire. A price, a policy, an owner changes. In a context graph an old fact is marked no longer valid instead of overwritten, so an agent can tell current rules from past ones.
Should a small business care?
Not yet. Know the term so you recognize it in a vendor pitch. The sources describing it are mostly vendors and one analyst radar.
References
- Context graphThoughtworks Technology Radar · 2026-04-15View source
- Graphs, knowledge graphs, and context graphs: Which one do you need?Neo4jView source