Glossaries

Extract glossary terms from documents

Building a glossary by hand is slow when the vocabulary already lives in your policies, specs, and onboarding guides. The Glossary Extractor reads a document and suggests business terms from it, each with a description, a likelihood, and a one-line reason. You review the suggestions and add only the ones you want.

Treat the suggestions as a starting point. You decide what belongs in the glossary.

In the app: Business Context › Glossaries › your glossary › Glossary Extractor

Before you start

Choose a source

  1. Open the glossary and click Glossary Extractor.
  2. Pick a source with the selector at the top of the dialog:
    • Upload: choose a file from your computer. It accepts the same file types as Processes, such as Word, PowerPoint, PDF, Excel, Markdown, and plain text, up to the size shown in the dialog.
    • Paste: paste text into the box. Useful for one section of a longer document.
    • Process document: pick a document you’ve already uploaded to Processes. Only documents with the status Ready are listed. This option appears only when Processes is set up for your environment.
  3. Optionally add Extra instructions (optional) to steer the agent, for example “focus on finance terms, ignore product names”. The limit is 2,000 characters.
  4. Click Extract Terms.

The dialog shows Interpreting document… while the agent reads. A long document is read in several parts, so give it a moment.

[screenshot]
The Glossary Extractor dialog with the Upload, Paste, and Process document sources and the extra instructions box.

Review the suggestions

The results appear under Review extracted terms. A summary line tells you how many new terms were found and how many already exist, for example “8 new terms found, 2 already exist — read in 3 parts.”

Each row shows:

Column What it tells you
Checkbox Whether the term will be added.
Likelihood Very Likely, Likely, Maybe, or Possible: how confident the agent is that this belongs in a governed glossary.
Term The suggested name. Marked Already exists when any glossary in your organization already has a term with that name.
Description A suggested definition, with the agent’s reason underneath.

New terms rated Very Likely or Likely start ticked. Maybe and Possible start unticked, so you opt them in on purpose. Terms that already exist can’t be ticked. Use the header checkbox to tick or clear every selectable row.

[screenshot]
The review grid with likelihood badges, one Already exists row, and the confident rows pre-ticked.

The likelihood is the agent’s judgement, not a fact. A Very Likely row can still be wrong for your glossary, and a Possible row can be exactly what you need.

Note: For a very long document, only the first portion is read, and the dialog warns you that terms further in may be missing. Paste the later sections in separately to cover them.

If the agent finds nothing that looks like a business term, it says so. Try another source or change your instructions.

Add the terms you want

Nothing is saved while you review. Click Add N Terms to write the ticked rows to the glossary. Any that turn out to duplicate an existing term are skipped, and the confirmation tells you how many were added and skipped.

Each new term is created with the status Draft and the suggested description as its definition. Open each one to check the definition, then add owners, synonyms, and relationships as you would for a term you defined by hand.