I’m building Terraphim around a small frustration in knowledge work: we change our words, then lose track of the ideas behind them.

A project brief calls something a risk. A colleague asks about their concerns. A literal search can miss the useful decision sitting beside the word neither person remembered to type.

We made an 80-second demonstration of a different flow. It starts with choosing a word in an editor and ends with finding a decision from the command line. The connection is an explicit, exportable knowledge graph.

Try the editor and download the actual files.

Start with the sentence

The example is a public community garden brief. Timber might arrive late, rain might delay planting, and volunteers might cancel. Each risk has a decision beside it: keep a backup supplier, reserve a covered area, or keep a reserve list.

In Terraphim Editor, I add “Concern” as an alternative to “Risk”. I can swap between the two without losing the other wording. For the final clean document, I keep “Risk”. The word “Concern” is absent.

That last detail matters. Otherwise a successful search for “concern” could simply be a literal match dressed up as a knowledge-graph demonstration.

Make the relationship inspectable

The editor’s Knowledge graph export shows the authored relationship:

synonyms:: Risk, Concern

Those alternatives refer to the same idea in this document. I download the graph as a Terraphim thesaurus and export the clean text. This is an explicit sharing step; creating an alternative does not silently change every role or every other application’s graph.

The downloads are part of the demonstration. You can inspect the graph, reopen the Markdown with alternatives, and compare it with the clean brief.

Search with the meaning you just recorded

I pass that exported graph to terraphim-grep:

terraphim-grep concern --thesaurus knowledge-graph.json --paths corpus -C 2 --search-only --json

It finds all three “Risk” passages and the decisions next to them, even though “Concern” does not appear in the clean text.

The released-binary verification also includes a baseline. With the editor’s default role, which does not contain this authored synonym, the query returns zero passages. Supplying the exported graph produces three. That makes the effect of sharing the synonym visible.

Decide whether you need a generated answer

Search alone gives the source passages and their locations. Replace --search-only with --answer to request synthesis from your configured model provider.

The paired results retrieve the same three passages. The generated answer adds a summary with references; it does not replace the evidence. It also should not turn a decision into a completed action. “Keep a backup supplier” is a recorded decision, not proof that someone has already contacted one.

Editing and local retrieval run locally. Optional synthesis sends the retrieved passages to the configured provider. A fresh answer may use different wording.

Try the evidence, not just the video

The tutorial includes the graph, clean brief, annotated Markdown, actual results with and without synthesis, captions, transcript, and a ZIP with reproduction instructions. It distinguishes the original recording’s candidate builds from the separate verification with Editor v0.1.0-rc.1 and terraphim-grep v1.21.19.

This is one focused browser-to-search workflow. It is not a claim that every editor host or feature is finished.

“Grammarly for knowledge work” is the ambition: help with the wording now, and preserve enough meaning to make the work easier to find later.

Which pairs of words cause missed searches in your project notes?

Demo and reproducible downloads: https://terraphim.ai/how-tos/synonym-search/

Disclosure: I work on Terraphim. The example is public sample input; the graph exports and search results are actual outputs. The video uses synthetic British female narration. This article was prepared with AI assistance and checked against the actual outputs.