Dawn Can Now Read the Live Web Natively

Dawn agents can now read public web pages natively, including pages whose useful content appears only after JavaScript runs. Give an agent a URL and ask it to inspect, compare, summarise, or extract the page. For standard public-page use cases, the native scraper is usually all that is needed; FourA remains the recommended option when the job requires proxies or dedicated scraping infrastructure.

Until now an agent could find a page and then struggle to read it. Search returns a snippet. Fetching the URL directly returns whatever the server sends before any JavaScript runs, which on a modern site is often an empty shell with the real content nowhere in it.

Start With a URL and a Question

You do not need to choose a scraping mode or explain how the site is built. Paste a public URL into the conversation and describe the information you need:

Read this pricing page and compare the plans, limits, and annual discounts.

Extract the release notes from these three product pages and build a dated comparison.

Check this public status page and summarise the current incidents and affected regions.

Dawn first tries the quickest appropriate way to retrieve the page. When the useful content depends on JavaScript, it can render the page in an isolated browser and extract the resulting text. The agent receives readable page content rather than raw markup, scripts, cookie banners, and repeated navigation.

That makes the feature useful across ordinary articles, documentation, public product pages, status pages, and many modern application-style sites. It also means the same request can work across a mixture of static and JavaScript-rendered sources without the user having to diagnose each one.

Turn Pages Into Useful Material

Once the content is available, it joins the rest of the task. Ask Dawn to compare several vendors against a purchasing brief, collect public evidence for a research note, extract a structured list into a spreadsheet, or use current web information alongside workspace files and connected sources. A product agent can compare competitor release notes with the roadmap. An operations agent can combine a public service notice with internal incident context. A marketing agent can inspect public landing pages before drafting a campaign review.

Native web data can also feed a Live Dashboard. A dashboard can refresh a focused public dataset on the server instead of freezing copied page text into the first version. The published view still follows Dawn’s dataset and sharing rules.

Native Scraping or FourA

The native scraper is the simplest starting point for standard public pages. It is built into Dawn, needs no additional connection, and handles both ordinary HTML and many JavaScript-rendered sites.

FourA continues to be the recommended option when a scraping use case requires proxy routing or benefits from a dedicated scraping service. Teams can keep using FourA for those workloads while relying on native scraping for everyday page reading and extraction. The two options are complementary: start with the built-in path, and use FourA when the site or operating requirements call for it.

Focus the Extraction

For long or noisy pages, tell the agent what matters. Asking for the pricing table, the article body, a named section, or the first 20 results gives the run a clear target and keeps the output useful.

The agent can work with bounded sections of a page rather than pulling an unlimited response into the conversation. This is especially helpful for documentation sites, long reports, and pages with repeated listings. If a page contains more information than one task needs, narrow the request just as you would when asking someone to review a large document.

When the source matters, ask Dawn to include the page URLs in the result. Web content can change, so a report that depends on current public information should make its sources and retrieval context easy to check.

Public Pages, Not a Remote Desktop

This release is for retrieving and extracting public web content. It is not an interactive browser session and does not sign in to websites, click through private application flows, fill forms, maintain a shopping cart, or take over an authenticated browser.

Pages that require a login, CAPTCHA, geographic challenge, or interactive consent may not be readable. Sites can also block automated retrieval or change their rendering in ways that affect extraction. When a provider offers a supported Dawn plugin, that remains the better option for private account data and durable provider actions.

Dawn also treats page content as untrusted material. A page can inform the task, but text found on a website does not become an instruction that overrides the user’s request or the agent’s permissions. Requests remain subject to limits on destinations, redirects, response size, and run budgets.

Use It Wherever Current Public Context Matters

For standard use, there is nothing new to connect. Ask an agent to read a public URL in a normal conversation, use it as part of a research task, or include it when creating a report or dashboard. If the task requires proxies, connect FourA and use its dedicated scraping capabilities instead.

For broader discovery, the /web-research command can still find relevant sources and turn the results into durable workspace memory. Native web reading complements that flow by letting the agent inspect the pages it finds, including modern pages that do not reveal their useful content in the initial HTML.

Dawn can now move from finding a page to actually working with it. That gives teams a direct path from the live web to a comparison, report, file, memory page, or dashboard—without adding another connection first.