How does Drafty decide which pages to include?

From what the site already publishes: its robots.txt, sitemap, and homepage links. Repeating templates collapse to one example (one recipe stands in for three hundred), and the canvas caps at twenty screens. You can preview the list first, or hand-pick the URLs yourself.

Does it crawl the whole site?

No. Screens are chosen before anything is rendered, so a content-heavy site with thousands of pages still becomes a canvas of at most twenty screens. There's no spidering and nothing recursive.

Can I keep a site canvas up to date?

Yes — it can re-shoot its own screens on a schedule, exactly like any self-refreshing canvas. The same pages are captured again and the canvas updates in place, with history. Useful for watching a competitor's site or a staging deploy over time.

Is the canvas live, or a snapshot?

A snapshot — every screen shows when it was captured, and links quietly to the live page so you (or your agent) can check the current state. Comments attach to the snapshot, which is what makes them precise.

Does it work on staging or password-protected sites?

Public and basic-auth URLs work today. Sites behind a real login aren't supported yet — that's on the roadmap. Your agent renders the screens locally on your machine, so a private staging URL never transits a third-party service.

How is this different from commenting on a normal canvas?

Same commenting, different anchor: on a site canvas you're pointing at a precise spot on a screenshot (Drafty records the exact position within the image), rather than at a live element. Your agent sees the picture plus the point — which beats describing the bug in words either way.

Updated June 11, 2026

Can I annotate a live website?

Quick answer

Yes. Say "drafty present <url>" to your agent and Drafty builds a site board: the site's main screens (found from its own sitemap, no crawling), each captured at desktop and phone width, published as an annotatable canvas. Tap the exact spot that looks wrong and type — your agent gets the image, the precise point, and the page's URL. Works on your own staging deploy or any public site; nothing is installed on the site itself.

Yes — and you don't need to own the site, install anything on it, or ask anyone's permission. Say drafty present plus a URL to your agent, and Drafty turns the site into a site board: a canvas of its main screens, every one of them commentable.

What you get

The agent reads what the site already publishes — its robots.txt, sitemap, and homepage links — picks the screens that matter (repeating templates collapse to one example, capped at twenty), and captures each one at desktop and phone width, with a timestamp. The result is published as a normal Drafty canvas: labeled sections, one screen after another.

Nothing is crawled. Screens are chosen before anything is rendered, so a site with ten thousand pages still becomes a canvas of at most twenty screens.

Commenting on a screen

Tap the exact spot that looks wrong — a wrapped nav, a broken card, a price that's changed — and type. Comments on a screen take point anchors: Drafty records the precise position within the image, so your agent sees the picture, the point you marked, and the URL of the real page it came from. From there the normal loop runs: the agent looks, researches, proposes, replies on the thread.

What it's for

Staging QA — present your own staging deploy and walk the screens at both widths.
Redesign kickoff — capture the current site, annotate everything that bugs you, hand the thread to the agent building the new one.
Competitor research — capture a rival's site and ask your agent what they're doing that you aren't.
Client review — send someone a canvas of their own site; they tap and type, no tools or accounts needed.

Keeping it fresh

The canvas is a dated snapshot — every screen shows when it was captured. To watch a site over time, put it on a refresh schedule: it re-shoots the same screens on cadence and updates in place, with history. Same machinery as a self-refreshing canvas.

For the full story — and everything else it's good for — see Stop screenshotting other people's websites.

Get started

Frequently asked

How does Drafty decide which pages to include?: From what the site already publishes: its robots.txt, sitemap, and homepage links. Repeating templates collapse to one example (one recipe stands in for three hundred), and the canvas caps at twenty screens. You can preview the list first, or hand-pick the URLs yourself.
Does it crawl the whole site?: No. Screens are chosen before anything is rendered, so a content-heavy site with thousands of pages still becomes a canvas of at most twenty screens. There's no spidering and nothing recursive.
Can I keep a site canvas up to date?: Yes — it can re-shoot its own screens on a schedule, exactly like any self-refreshing canvas. The same pages are captured again and the canvas updates in place, with history. Useful for watching a competitor's site or a staging deploy over time.
Is the canvas live, or a snapshot?: A snapshot — every screen shows when it was captured, and links quietly to the live page so you (or your agent) can check the current state. Comments attach to the snapshot, which is what makes them precise.
Does it work on staging or password-protected sites?: Public and basic-auth URLs work today. Sites behind a real login aren't supported yet — that's on the roadmap. Your agent renders the screens locally on your machine, so a private staging URL never transits a third-party service.
How is this different from commenting on a normal canvas?: Same commenting, different anchor: on a site canvas you're pointing at a precise spot on a screenshot (Drafty records the exact position within the image), rather than at a live element. Your agent sees the picture plus the point — which beats describing the bug in words either way.