What you get
Each run produces a folder of results. You can browse it, download it as a zip, or publish it.
Full-page screenshots
Every screen and state, including content hidden inside scrolling panels, lazy-loaded lists and pop-ups.
Annotations
Numbered callouts, boxes and notes positioned on real elements. Personal data is blurred.
Page models
Each page's layout, fields, labels, tables, actions, tabs, messages and design details, as structured data.
User flows
Step-by-step journeys through the app, with UX notes for each step.
Behaviour notes
What each action does: the change it causes on screen and the request it sends to the app.
A report
A written report shaped by your goal, plus a browsable HTML version with the screenshots built in.
How it works
The code does the precise, repeatable work. The AI decides what to explore and writes the findings.
Write a brief
A short job description: which app, which areas are in scope, the goal (documentation, QA, design review, competitor review and so on) and which outputs you want.
Sign in safely
The tool logs in with a dedicated account kept on the server. The AI never sees the password. If a site asks for a second factor or a CAPTCHA, a person can finish that step through a one-time link.
Check before exploring
A preflight confirms the login worked and scans each page for hazards, such as sign-out links, account switchers, delete links and toggles that save as soon as you click them. Those are marked off-limits before any clicking starts.
Explore and capture
It maps the navigation, visits each page in scope, opens tabs, menus and dialogs, and captures each state. Pacing is deliberately human-like so the app isn't overloaded.
Check its own work
Every capture gets an automatic quality check, and every callout is checked against the element it points at. Problems are fixed before the report is written.
Write it up
Results are assembled into the report and documents your goal asked for, ready to download or publish.
Ways to use it
There are three ways in. They all use the same engine and the same safety rules.
Ask Claude
Connected as a tool to Claude. You describe what you want documented in plain language, and Claude drives the session step by step. This is the best way to get rich, written documentation.
Background jobs
Long crawls of a whole module run as a job. You start it, check on it later, and download the results when it finishes.
Command line
A survey that doesn't need AI, plus replay, compare, export and publish commands for developers and scheduled tasks.
Safety by design
The tool is read-only by default. It's built to look at an app, not to change it.
Inside the app
- Before every click, it checks the button's label and the element itself. Save, delete, submit, sign-out and account-switch actions are refused.
- A second, network-level guard blocks any request that would create, change or delete data.
- The browser is locked to the domains named in the brief.
- Editing can only be turned on for sandbox or test environments, with explicit allow and deny lists.
Around the data
- Credentials stay on the server and are tied to the sites they belong to. They never pass through the AI.
- Saved sign-ins are encrypted at rest.
- Personal data in screenshots is blurred.
- Access to the service needs authorisation, and download links are signed and expire.
- Old runs can be cleaned up automatically on a schedule.
Document types
There are ready-made instructions for 14 kinds of output. Each one sets what to capture and the quality bar for the write-up.
Quality checks and learning
Automatic capture QA
Flags blank pages, unexpected login screens, duplicate shots and callouts that miss their target.
Accessibility
Runs WCAG 2 A/AA checks on each page and includes the results in the report.
Replay
Saved capture plans can be re-run later to refresh screenshots after a release.
Compare runs
Two runs can be compared to show which pages, fields, columns and actions changed.
App profiles
The tool remembers each app's quirks and hazards, so later runs are faster and safer.
Test export
Recorded flows can be exported as a starting point for automated browser tests.
FAQ
Can it change data in the app I'm documenting?
Not by default. Two independent guards, one on clicks and one on network requests, stop anything that would save, submit or delete. Editing can only be switched on for sandbox or test environments.
Does the AI see my passwords?
No. Credentials are stored on the server and used only by the sign-in code. The AI is told whether the login worked, never what the credentials were.
What about apps with two-factor sign-in?
If a site asks for a code or a CAPTCHA, the run pauses and sends a one-time link. A person opens it, completes the step in a live view, and the run carries on.
How long are results kept?
Results stay on the server until they're cleaned up under the retention policy. Important runs can be pinned so they're kept.
How do I get access?
The service is for authorised users only. Contact your administrator to be connected.