Rigorous synthesis, without the spreadsheet grind

Notes, highlights, personas, journeys, user tests, tables, and files — one linked evidence layer.

Upload PDFs, docs, spreadsheets, interview recordings, and links — Crux parses and indexes them all.

Highlight passages in source documents and code them with a shared label library.

Point the Agent at a source and get labelled highlights, an analysis frame, and the links between them.

Structured, editable tables with stable cell references — built by hand, from CSV, or from Ask.

Plan, run, and synthesise user tests — with AI drafting the guide and the readout.

One ⌘K search across research items, experience entities, files, and highlights.

Group everything that came out of one meeting, workshop, deliverable, or audit — evidence and product structure alike.

Say what each step of a piece of work has to leave behind — then link or create the artefact that fills it.

Dated outcomes with an owner, a priority, and a status — with the activities and to-dos that deliver them underneath.

See the whole plan as an outline, a month calendar, a gantt timeline, or the people doing the work.

Every version can say why it changed, what prompted it, and which source is behind it.

Context-aware answers from your own graphs — with citations, not vibes.

Press Start and the Research and Experience agents seed both graphs from your sources.

Pick Fast, Sharp, or Capable per project — a dial for how hard the Agent thinks.

Turn any answer into notes, evidence, personas, journeys, tables, to-dos, or page structures.

Break a goal down into milestones, activities, and to-dos — reviewed in a preview before any of it is saved.

AI drafts, you decide — answers and generations never overwrite your work.

Score a live page against a cited usability benchmark — findings land in the research graph.

Pull evidence from Google Drive, Dovetail, and your own website into the research graph.

Private by default; share projects with your organisation when the team needs them.

PII auto-redaction, prompt-injection blocking, no model training, strict tenant isolation.