Terminology Graph Explorer

The Terminology Graph Explorer is an interactive map of the domain-specific terminology inside a knowledge base. It shows which concepts appear across your documents and how they co-occur — which pairs of terms tend to be discussed together, and how strongly.

The graph is useful in two distinct ways: as a research tool you can use directly to navigate your corpus from a concept perspective, and as a foundation the agent draws on when it needs to understand what a term means in your specific domain.

What the graph represents

Each node is a term extracted from your documents — a concept, method, entity, or phrase that appears enough times to be meaningful. The connections between nodes are co-occurrence edges: two terms are connected if they appear in the same semantic chunk across your corpus. The thicker an edge, the more papers in which those two terms appear together in a single semantically cohesive chunk.

Nodes are sized by the number of chunks they appear in, and the information score (shown in the info sidebar) reflects how information-rich the passages containing that term tend to be regarding the term — a high score means the term is substantively discussed.

Building the graph

The terminology graph is built per knowledge base, not per project. It requires at least 10 documents to be indexed in the KB before it can be built.

To build or rebuild it:

1. Open the KB from the Knowledge Bases view.
2. Scroll to the Terminology Graph section at the bottom of the KB page.
3. Click Rebuild.

Building the graph is free of charge and runs in the background. You can watch the progress bar fill from 0% to 100%. Once it reaches Completed, the Explore button becomes available.

Rebuilds are throttled: once a build completes, you can trigger a new rebuild after four hours.

The graph view

The graph shows a local neighbourhood around a root term — not the entire graph at once, which would be too dense to navigate. From a root, you see:

• Root node — the term you navigated to, displayed in a dark node at the centre.
• First-hop neighbours — terms that directly co-occur with the root, shown in a blue-tinted ring around it.
• Second-hop neighbours — terms one step further out, shown in a lighter colour.

Edges are drawn as curves between nodes. Edge thickness scales with the number of papers in which the two terms co-occur in at least one semantic chunk.

Searching and navigating

Use the search box in the top bar to jump to any term. As you type, autocomplete suggestions appear, showing the term's name alongside its paper count and chunk count. Select a suggestion to navigate there — that term becomes the new root and the graph redraws around it.

Double-clicking any node also navigates to it, making that node the new root. This is the primary way to move through the graph by following concept chains.

The explorer maintains a navigation history. If you navigated through several terms, you can go back and forward using the browser back/forward buttons.

Keyboard shortcuts also work when the search box is not focused:

• + / = — zoom in
• - — zoom out
• z — zoom to the currently selected node

Inspecting nodes and edges

Click a node to select it. The Info tab on the right sidebar shows:

• The term's display form
• Papers — number of documents in which this term appears
• Chunks — number of semantic chunks containing this term
• Info score — average information density regarding this term in passages containing it. A higher score means the term is discussed in more substantive contexts.

Click an edge to select it. The Info tab on the right sidebar shows the co-occurrence relationship:

• Both term names
• Papers — number of documents where both terms co-occur
• Chunks — number of chunks where both terms appear together

Viewing supporting text

The Chunks tab in the right sidebar shows the actual text passages from your documents that are associated with the current selection. Each entry is a citation card linking back to the source document at the exact passage — the same citation format used on mindmap nodes, similarly you can add labels to these citation items for quick reference.

• When a node is selected: all passages containing that term, ranked by information score - the first entries are the most informative regarding that term.
• When an edge is selected: passages where both terms appear together.

How the agent uses the terminology graph

When the agent runs a research workflow, it doesn't just query your documents by keyword or embedding similarity. It uses the terminology graph to resolve ambiguous terms, and find related concepts.

See also

• Knowledge Bases — managing KBs, adding documents.
• RAG & Retrieval — how indexing extracts terminology and how the agent retrieves from KBs.

What's next

• AI Verification — verifying individual mindmap nodes against their citations.