Changelog

Our payment provider is now Polar, the merchant of record: it collects the tax that applies where you are and issues the invoice, and the card form opens inside the app rather than on another site, in onboarding and in Billing. Plan changes are made on your own Billing screen with the price shown before you confirm; an upgrade applies at once and a downgrade waits for the period's end. Every invoice is a download away, and replacing the card or cancelling is one press into Polar's customer portal. The 7-day trial still takes a card and charges nothing until day seven. If you paid with Stripe before this, your plan keeps running; when it is time, Billing asks you to choose it again once, and the days you already paid for carry over.

The tracker, the one file that runs in your visitors' browsers, is now released under the MIT licence, so a company that cannot ship AGPL code on its own pages can use it without a question to legal. The server and the rest of the product stay AGPL-3.0, exactly as before.

The last onboarding step used to wait for a first event and say nothing until one came. It now reports the tag itself: the moment your script loads anywhere, the step says where it was seen and whether events from there will be counted. A tag loaded on a host outside the site's allowlist gets a warning and the fix, adding that host, instead of a wait that never ends. The overview's install modal carries the same badge.

Press See all on the revenue card and every payment in the selected range is listed, newest first: charges, refunds and disputes, each with its amount in your reporting currency and a mark on the ones that could be tied to a visit. The list follows the range you have picked and loads more as you scroll.

The sites screen stopped being a list of names. Each card now carries the site's all-time visitors and page views, its net revenue when a revenue provider is connected, and a small sparkline of weekly visitors, so you can read how every site is doing before you open it. The figures refresh every few minutes.

A new card collects the visits that AI assistants send you: ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, DeepSeek and more, whether they arrive as a referrer or as the utm tag an assistant's app attaches to its links. The title shows their share of all visitors and See all lists each assistant. Devices, browsers and operating systems moved into one card to make the room, with Devices first, and the public share board gained the same card. Google's AI Overviews cannot be told apart from Google search, so they stay counted as Google.

The share dialog gained the networks: post the overview image to X or Bluesky with a caption ready to edit, or to LinkedIn, and on a phone the system share sheet takes it from there. The colour swatches and the dithered chart area were refined on the way.

A site now carries a reporting timezone of its own: the dashboard, the weekly digest, the MCP tools and the API read in it, so a team spread across zones sees the same day. Underneath, the smallest unit of the charts became fifteen minutes, which is what lets a zone half an hour off UTC, India or Newfoundland among them, read its real hours instead of the nearest whole one.

Press Share on the overview and the numbers on screen become a 1200×630 image: the site, the range, the figures you pick and the visitors chart, drawn in your browser from exactly what the cards show. Light or dark, six chart colours, and the site name can stay off it. Download the PNG or copy it straight to the clipboard. An active filter is always printed on the picture, so a slice never passes for the whole site.

Click a source, country, city or device in any breakdown and the whole overview follows it: every card is redrawn for the visits that match, and the chips ride the tab bar until you remove them. Top pages also gained entry and exit rankings, so you can see where visits begin and where they end.

The funnel editor became a two-pane board: steps stand as connected nodes you can insert between, edit in place and remove, with page and event suggestions drawn from your own traffic. Every funnel also gained a report range, from today back to 90 days, instead of the fixed last month.

A weekly digest of your sites' numbers, a note when the month's events are running out, and notification preferences with one-click unsubscribe on every mail you may refuse.

The open-source build became a package that actually boots on your own infrastructure: its own migration stream, SMTP and admin bootstrap wired in, and the full dashboard running without the hosted service.

The product has a manual: installation guides per framework, features explained the way they actually behave, and the MCP and API pages agents read. It lives at /docs, and the screens that need it link straight into it.

Ask in plain words, “why did signups drop last week”, and the answer streams back computed from your own numbers, through the same read API the dashboard itself uses.

npx getopen init detects your framework, writes the snippet where it belongs and walks the rest interactively, before an account even exists on the machine.

Connect Claude, Cursor or any MCP client to your sites: the dashboard's reads are the tools, authorized per site with OAuth, so an agent pulls traffic, funnels and revenue without an export. Setup lives in site settings and the docs.

Publish a card of your stats and put it anywhere an iframe goes, in either theme, with a live preview while you shape it. The footer of our own site runs one.

“Online now” became a server-owned fact: a visitor counts as here for five minutes past their last signal of any kind, a hidden tab drops out instantly and returns instantly, and the realtime board and the globe can no longer disagree about who is on the site.

Stripe payments now join the visits that earned them. Every transaction carries its journey (the pages, the referrer, the campaign), and the overview gained a revenue card that reads in your reporting currency.

Bring your history with you: upload an export from your previous analytics provider and it lands as first-class data: old ranges keep their numbers, blended ranges say exactly which parts are imported.