01 · Capture

Record the work. Not a video of the work.

A screen recording cannot be searched, cited, or replayed. Cairn records structure — the elements, the labels, the order — and generates everything else from it.

How it works

One pass through the task, and the documentation exists.

  1. 01Click through the task onceBrowser extension, desktop app, or dictate it aloud. No script, no retakes, no editing timeline.
  2. 02Cairn records structure, not pixelsElement roles, accessible names, ranked selector candidates, per-step URLs — and a screenshot taken before the click mutates the page.
  3. 03Redaction runs before anyone sees itEnforced policy rules apply at ingest. The field label is kept; the value typed into it never leaves the machine.
  4. 04You get a guide, not a to-doSteps, ordering and draft prose arrive already written. You edit rather than author.
In the product

The extension, mid-capture

Steps accumulate on the left as you work. Each one carries its screenshot with the click target already highlighted — no annotation pass afterwards.

capture-extension.pngExtension mid-capture: step list left, screenshot with highlighted click target right
The detail

What makes the capture good enough to replay later.

Three capture surfacesA Manifest V3 browser extension, a Tauri 2 desktop app for native software, and voice dictation for workflows that are not on a screen at all.
Screenshots timed to the intentTaken before the click applies, so the button being described is still visible in the image. Most tools capture after, and show you the result instead of the action.
Labels, never values"Card number" is recorded. The number is not. This is an architectural rule with tests behind it, not a setting someone can flip during onboarding.
Four tiers of redactionManual blur; assisted Smart Blur with a durable decision log; enforced policy redaction that runs at ingest before any human sees the image; and custom admin regex or selector rules.
Strict originals for EnterpriseEncrypt every screenshot original at rest, never serve it, and permanently delete it once a redacted derivative exists.
Ranked selector candidatesThe reason a captured step can later be replayed on a live page rather than merely illustrated. This is the data the walkthrough runs on.
Where this sits

Capture is where the loop starts — and where the returning work lands.

01Capture
02Knowledge
03Engage
04AI agent

Capture flows to Knowledge, to Engage, to the AI agent; a miss becomes a work order and a resolution becomes a draft, feeding back into Knowledge.

Honest limits

Capture is only as good as the application being captured.

You will find this section on every feature page. If a constraint would annoy you in week two, we would rather you knew in minute one.

  • An application that renders everything to canvas with no accessible names gives Cairn very little to work with. You will still get a guide with screenshots; the interactive walkthrough will degrade to static.
  • Voice capture produces prose, not element data. It documents a workflow well and cannot be replayed as a walkthrough.
  • Desktop capture is Windows and macOS. Linux is not built, and we would rather say so than list it as "coming soon".