Resources
What people actually build with this
Four situations that come up again and again, written out in full: what goes wrong without a browser, what the API does about it, and which endpoints do the work.
For teams building agents
Give a model the page, not the page's source code
Your agent needs to read a supplier's product page, a competitor's pricing table, a customer's help centre. You fetch the URL and get back a document that is 95% framework: hydration payloads, inline styles, cookie banners, three navigation menus. The model spends its context window reading markup, and it still hallucinates the price, because the price arrived after the JavaScript ran and the fetch never ran any.
One POST renders the page in a real headless Chromium, waits for the content to actually exist, and hands back Markdown that reads like the page looks. Headings, tables, lists, links and alt text survive. Scripts, styles and hydration artefacts do not.
Where it shows up
- Ingestion for a RAG index that stays close to what a human reader would have seen.
- A research agent that can open a link its user pasted, rather than guessing from the URL.
- A product-catalogue sync that reads the live page instead of a stale feed.
For teams watching something
Find out that a page changed before your customer tells you
Something on the open web matters to your business and nobody owns watching it: a competitor's pricing page, a regulator's guidance note, a partner's API status page, your own checkout after a deploy. The usual answer is a cron job that diffs raw HTML, which means it fires every time a build hash or an ad slot changes, and everyone learns to ignore it.
A monitor captures the page on a schedule and keeps the history. Because the comparison runs on the extracted content and the rendered image rather than on the markup, a rotated asset URL is not a change and a reworded refund policy is. When it does change, a signed webhook arrives with the run attached.
Where it shows up
- Competitive pricing alerts that go to a channel people still read.
- Compliance and policy watch on documents you do not control.
- Post-deploy visual checks on the handful of pages that must never break.
For teams that need a record
Keep proof of what a page said on the day it said it
A dispute, an audit, a takedown, a claim about what your terms used to be. A saved copy of the HTML is not persuasive, and neither is a description. What settles it is an image of the page as a browser rendered it, with the date attached — and the moment you need one is always after the page has changed.
Every capture can return an image or a PDF of what the browser saw, alongside the extracted text. Captures are retained on your plan's schedule and can be pulled back from the dashboard or the API, so the record is created as a side effect of the work you were already doing.
Where it shows up
- Archival of terms, disclosures and public filings on a schedule.
- Evidence capture inside a support or trust-and-safety workflow.
- Client-ready PDF reports generated from live pages.
For teams doing this to a whole site
Work through a site without writing a job runner
One page is a request. Nine hundred pages is infrastructure: a queue, retries that do not double-charge, somewhere to put partial results, a way to answer "is it done yet", and a plan for the run that dies at page 600 because a Lambda timed out. Most of that is not the problem you set out to solve.
Point a crawl at a seed URL with the origins and path patterns it may follow, or hand a batch an explicit list. The job is durable: it survives a worker dying, each child reports its own result, quota is charged when a child starts rather than when the parent is created, and an idempotency key means a retried submission is the same run and not a second one.
Where it shows up
- First-time ingestion of a documentation site or knowledge base.
- Periodic refresh of a catalogue you already index.
- Bulk migration work where you need every page's content and a picture of it.
02 / 02 · Start here
Three ways in
Whichever of the four above is yours, it starts the same way.
Give your product a browser.
Get clean web content and visual proof into your workflow in minutes.