What changed

A bell in the top bar collects what needed you: a monitor down, a certificate running out, a domain expiring, an error spike, links that broke, a drop against last week. It fills whether or not an alert channel was there to receive the same news, so a webhook that quietly stopped working no longer means you hear nothing. Read, dismiss or keep them; the whole history is one page.

Moved a site, fixed the thing that was failing, or added the monitor against the wrong URL? A monitor can now throw away every probe it has taken, and with them its uptime figures and its outages, here and on the public status page. It asks first and says what goes, then leaves a note on the timeline saying who did it.

Site keys, hostnames, snippets and stack traces were drawn in a coding typeface, which made the product look like a terminal for no benefit. They are proportional now, with the one thing a monospace was actually earning its place for kept: a slashed zero and letterforms that cannot confuse 1 with l. The documentation follows.

Screens that have no date range — Settings, Ask, Alerts, Realtime — used to drop the picker and the export button the instant you arrived. They fade now, and only when they actually change hands: arrive on a screen that has them and they are simply there.

Every monitor on a status page carries a 1h / 24h / 7d / 30d / 90d picker. One bar is five minutes on 1h, an hour on 24h, two hours on 7d and a day on the two long ones, and each monitor keeps its own choice.

Green is every probe up, red is every probe down, yellow is a span that was partly both, grey is a span nothing was probed in. Pointing at a bar names the span in the site's time zone and how many probes ran and failed in it.

Point status.yoursite.com at us with one CNAME and the status page is served there, certificate included. The published dashboard takes a domain of its own the same way. Both are under Sharing in Settings.

A site can be cited by an assistant without a single fetch being recorded, because assistants read pages from their own servers where the script never runs. Crawlers now opens on AI answers when that is the case and says so, instead of looking empty.

Asked about revenue on a plan that includes it but with no sale reported yet, the copilot said revenue was a Pro feature. It now says what is actually true: nothing has been reported.

The rail reads Know (Overview, Realtime, Sessions, Events, Revenue, Features), Protect (Uptime, Errors, Performance, Links), Grow (Search, SEO, Journeys, Goals, Funnels, Retention, Heatmaps, Crawlers) and Understand (Timeline, Alerts). Ask sits above them because it answers all four.

Up to five monitors per site on Pro (one on Hobby, twenty on Business): HTTP, a keyword that has to be on the page, or a host and port. Each runs on its own interval, down to every minute on Pro; two failures in a row is an outage, and an alert goes out when it starts and when it ends. Once a day the certificate, the registration (from the registry's own record), the DNS answer, robots.txt, the sitemap and the homepage are read too.

A page at tracing.tools/status/your-name with every monitor you pick, ninety days of daily uptime and the past incidents. Served by us, so it is up when your site is not. Set it up under Sharing in Settings.

A daily crawl starts at the homepage and the sitemap, follows every internal link and checks every external one. Each broken link is listed with the pages that point at it, whether an indexing crawler fetched it, and the sessions the path had in the last month, so the list is in the order to fix them. Links that broke since the last crawl get an email.

Every page the crawl fetched with what keeps it from ranking: noindex, missing or duplicate titles, canonicals pointing elsewhere, pages the sitemap forgot, slow answers. Joined to the pageviews and, when Search Console is connected, the clicks.

Deploys (a new build is a change in the scripts the homepage loads), homepage and SEO changes, robots.txt and sitemap edits, DNS and certificate changes, outages and recoveries, broken links, and anomalies in the site's own numbers, all in one feed. Add a note for anything you did yourself. Changes are drawn as lines on the Overview chart, so a dip has its cause next to it.

One message per thing that needs a person: a monitor down and back up, certificate and domain expiry at set days out, error spikes, links that broke, and a visitors, conversions or revenue drop against the week before. Add as many channels as you like (an inbox, a JSON webhook, a Discord channel), send a test to each, switch any kind off per site. Every alert sent is on the record.

The four sections fold and stay folded, the rail scrolls behind a hairline that only shows on hover, and every new screen has its own loading shape instead of a spinner.

Three new tools: uptime and certificate state, the timeline, and the crawl's findings, so 'was the site down yesterday' and 'what changed before conversions dropped' have answers.

Lists that used to show everything at once are paged and sortable by their columns: broken links, pages, the timeline, alerts, errors, events, crawlers, search queries and pages, features, and the performance pages.

The tracker sends a short text digest of each page family once a day; the copilot reads them and names the features of your site, from the search box to the checkout. The Features screen then counts who uses each one, which pages they landed on first and how many of everyone who landed there got to it, with no events to define. One click turns a feature into a goal or a funnel.

Uncaught exceptions and unhandled promise rejections are reported with their message, file, line and stack, grouped by cause, and broken down by page, browser, system and country. Errors raised by a visitor's browser extension are dropped before they leave the page. tracing.error(err) reports one you caught yourself.

Every click as heat over the page itself, a ranked list of the elements behind them, and how far down the page visitors scroll, drawn as bands. Per device, since a phone and a desktop are two different pages.

An empty Goals or Funnels screen now points at one button. It reads the last thirty days of events, pages and routes, plus any detected features, and writes the obvious conversions and funnels. Everything it adds is editable.

This page, and a What's new section in every site's Settings that shows the version you are on.

New opt-outs on the script tag: data-errors, data-heatmaps and data-features. The script is 5.4 kB gzipped; the marketing pages and the docs say so.

A question bar over your own traffic and revenue, with inline autocomplete and answers grounded in the numbers on the screen.

Visitors, pageviews, sources, pages, technology and a realtime globe with the people on the site right now. Every plan.

Custom events with properties, named conversions, step funnels with drop-offs, cohorts by first visit, and sales tied to the page, source and campaign that earned them.

Core Web Vitals as real visitors saw them, and a directory of the bots and AI crawlers reading your pages.

Search Console and AdSense joined to the traffic that earned them, a REST API with exports, and your analytics inside any MCP client.