Stuff I've made

What stays in your browser

Everything the page is made of is stored by the browser on your own machine, and none of it is sent anywhere:

The page itself

Netlinks and their categories, widget contents (scratch pad text, task lists, RSS feed addresses), the greeting name, colours, scan lines, sounds, the tab's title and favicon, your chosen search engine and where you told the weather widget you are.

A wallpaper you upload

Kept in the browser's IndexedDB rather than uploaded anywhere, because an image or video can run to tens of megabytes.

Your most-visited sites

Read once, on the first run, to fill the netlinks grid so it doesn't start empty. The browser answers that question locally and the answer never leaves it.

A backup, only if you ask for one

The settings panel can write a snapshot of your settings or your netlinks into your browser account, so a new machine can restore them. It is written when you press the button and at no other time, and it goes to your browser account - Google's or Mozilla's - not to avi12.

Where you are

The weather widget starts out following this device, and it works out where that is when the page loads, without asking you anything. Which of two ways depends on what your browser is willing to say:

The device answers

The position is rounded to four decimal places, and those coordinates go to bigdatacloud.net, which turns them into the name of a place so the widget can say where it is.

The device won't, or can't

A desktop with no radios to be placed by, a Firefox whose position provider can't reach anything, a browser you told no - then ipwho.is is asked instead. It goes by the connection, so what it sees is your IP address, and it answers with a town and that town's coordinates. City-level and no better, which is all the weather ever needed.

Either way, the coordinates then go to open-meteo.com, which answers with a temperature and a number standing for the sky. That is the forecast, and it is the whole of what the weather widget fetches.

None of the three asks for an account, is given a key, or is told anything else about you. The widget shows you which of the first two placed you, so a VPN moving the second one is visible rather than silent.

Pin a place instead and the first two are never asked: you type the name and the coordinates yourself, nothing is looked up, and the device is never read. The coordinates you typed still go to open-meteo.com, because a forecast for a place is a request naming it. Pin Night City and even that stops - its sky is invented on your own machine, and nothing is sent anywhere at all.

The location permission is asked for at install because Chromium refuses to make it an optional one - and, having it, the page does use it on load rather than waiting to be told to. Pinning a place is how you turn that off.

What leaves your browser, and where it goes

Searches

Go to the engine you picked in the search bar, or to your browser's own default engine - the same trip the address bar makes. They are not logged or counted anywhere in between.

Prompts

A prompt goes only to the destination you sent it to - Copilot, ChatGPT, Claude, Perplexity, Google AI Mode or Ask Brave - and only when you press send. Two of them won't take a prompt in the address, so for Claude and Copilot a small script runs on that site to write the prompt in and press send. It carries the prompt and nothing else.

Addresses fetched through a proxy

An arbitrary web page doesn't let a browser extension read it, so three things are fetched through a public CORS proxy - api.allorigins.win, cors.redoc.ly or api.codetabs.com, whichever answers first: an RSS feed you added, the title of a netlink when you ask the form to fill it in, and a wallpaper address you pasted and asked to keep. Those proxies see the address being fetched. Nothing else goes through them.

The forecast

The weather widget is on out of the box, so the first time the page opens, your coordinates go to open-meteo.com and come back as a temperature and a sky. It is an open API - no account, no key - and the request carries no cookies, because a fetch to another site carries none unless it is asked to and this one isn't. It is asked by coordinate and never by name. It is also the only source: there is nothing to allow first and nothing to fall back to.

The wallpapers it ships with

The four shipped backgrounds are photographs hosted at images.pexels.com, so your browser loads one of them as an ordinary image the first time the page opens. Pexels sees what any image host sees. A wallpaper of your own is stored on your machine instead.

Your name

No browser API hands over a name, so "Use Google account" asks Google. You sign in at Google's own account chooser, the one-time code that comes back is exchanged at Google's token endpoint for an ID token, the token is read for the first name in it, and the access token is handed straight back to Google's revoke endpoint. Only openid and profile are asked for - no mail, no contacts, no files. Neither token is written anywhere. Nothing but the name is kept, and the name sits in a text field you can overwrite with anything you like.

Copilot Journeys and Tips

On Microsoft Edge for Windows 11 the page can show the Journey cards and prompt Tips Edge writes under your own profile. An extension can't read a browser profile, so a separate companion app reads those files and answers over native messaging. It reads them on your machine and hands them to the page on your machine - nothing about them is sent anywhere.

The app is the only source of both. Microsoft publishes the tip catalogue at a public address, which this used to fetch directly, and it no longer does: that address answers the same catalogue to everybody, and what belongs on the row is the one Edge cached for the profile you are actually in. So no request goes to Microsoft either.

Until you press Set it up, nothing is sent to the app at all, and the permission it needs is asked for at that moment rather than at install. Every other browser and platform never sees this section.

Edge's Tips ask about your own tabs or your own recent reading, so a card asks - separately, and only when you press it - for one of those two: your open tabs, or your history. Never both, and never the sites themselves. What it writes into the prompt is at most ten titles and addresses, with search pages and sign-in pages left out, and the destination is told it may open any of them it needs. The extension never reads what is on those pages, because doing so would mean asking you for every website you visit, which it will not do. Refusing leaves the card working with less. What you then send goes to the AI destination you chose, and nowhere else.

Counting installs

The extension reports five things to Google Analytics so there's some idea of how many people use it and what they use: a new tab opened, a search submitted, a prompt sent, the companion linked, and a control pressed. Alongside each one it sends the search engine or destination involved, the kind of card, and whether it worked - plus the extension's version, which browser it is (Chromium, Microsoft Edge or Firefox), and your browser's interface language.

"A control pressed" is every button and link on the page, and what it sends is the name the control was given in the source - "netlink_opened", "widget_toggled" - never the label you see. That distinction is the whole point: your netlinks are named after sites you chose and your categories after words you typed, so a report that described a button by what it says would be carrying your data out. Nothing derived from the page's text is ever sent, and a control with no name in the source sends nothing at all.

It counts against a random identifier generated on first use, which says "the same browser as last time" and nothing else. Your prompts, searches, netlinks, addresses and name are refused outright by the reporter - it drops anything named like that and any value that looks like a URL - and the advertising flags are set to denied, so none of it can become advertising data. Nothing is reported from a development or sideloaded build.

On Firefox, data collection is declared as a permission: the install screen lists location as something the extension uses, because the weather widget does, and lists technical and interaction data as a tick you can clear. Clearing it stops the reporting entirely.

What avi12 receives

Nothing. avi12 runs no servers for this extension and receives no data from it: no settings, no netlinks, no prompts, no searches, no location, no account, no crash reports, no ads. The only third parties in the path are the ones named above, each for the one feature that needs it.

Children

Cyberpunk 2077 New Tab is not directed at children and collects nothing about anyone's identity.

Questions

Email contact@avi12.com, or open an issue at github.com/avi12/cyberpunk-2077-new-tab.