Skip to main content
A causal graph is a diagram of the variables in your domain and the causal edges between them. Nodes are things you can measure or reason about (rates, prices, yes/no states); edges are the causal links between them. Nora uses the graph in two ways:
  1. Verification approved (live) edges are folded into grounding checks, so answers that contradict the graph get caught.
  2. Retrieval the graph supplies causal facts your documents alone don’t state, without duplicating chunks retrieval already found.

​
Graph shape

  • Nodes are variables numeric, count, or yes/no, plus non-measurable concept nodes.
  • Edges are causal an arrow from A to B means “changes in A cause changes in B”.
  • The graph is a DAG (directed acyclic). Adding an edge that would create a cycle is rejected on the spot.

​
Ways to build one

Open Causal Graph in the sidebar. An empty graph offers three starting points:
  • Add by hand author variables and edges yourself. Best when you’re the domain expert. See Variables and Edges.
  • From a folder an LLM bootstraps a graph from a Documents folder.
  • From a refinery seed the ontology from a refinery’s blocks.
You can also open the AI panel (Create with AI) and describe the model you want in chat. See Bootstrap from documents. A workspace can hold multiple graphs: switch, create, rename, or delete them from the graph selector in the toolbar.

​
Proposed vs. approved (live)

Every node and edge carries a review state:
  • Proposed staged for review. Not used downstream.
  • Approved (live) treated as live by verification and retrieval.
Hand-built items start approved; LLM-built items arrive as proposals. When there are pending proposals, a Show proposals · N toggle appears in the toolbar. Turn it on and the canvas highlights only the proposals; use the neighbouring Approve all button to accept every pending proposal at once. Each item’s Review tab also lets you approve/demote individually and controls whether LLM re-runs may overwrite it (lock it if you’re sure). See the review controls in Variables and Edges.

​
Learn more

Variables

The nodes: what you measure, and their review state.

Edges

Direction, sign, strength, and review.

Bootstrap from documents

Let an LLM propose the graph; you approve.

Relationship analysis

Pick a cause and an effect; the graph tells you what to watch out for.

Verify agent answers

How live edges gate what agents say.

How agents pick up a graph

Graphs connect through data sources: no per-Agent attach step.

​
What causal graphs are not

  • Not a replacement for knowledge. Keep your docs: the graph’s evidence chunks come from them.
  • Not the reasoner. Your Agent still uses an LLM; the graph is a check and a source of causal facts.
  • Not a general knowledge graph. It’s specifically for causal relationships.