The whole capture loop, in your browser.
Startwell Capture for Chrome is a free extension. Screenshots with your notes, voice notes and walkthroughs of a tab are filed by topic and handed to an AI agent in one copy. No account and no desktop app needed. Everything stays in your browser.
Coming to the Chrome Web Store. Until then, it loads from a folder in about a minute. Chrome 116 or later.


- Mark it up: arrow, box, pen, label, numbered pin, blur, crop.
- What it is: a bug, an idea, a question or praise.
- Pins number themselves, each with its own note.
- Save, take another: several shots, one piece of feedback.
What it does
Feedback
A screenshot of the page to mark up: arrows, boxes, numbered pins with notes, blur, crop. Then say what kind of feedback it is.
⌥⇧FAlt+Shift+FAlt+Shift+F
Screenshots, with a note
The whole page, or drag over part of it and say what is wrong in a line.
⌥⇧SAlt+Shift+SAlt+Shift+S
Voice notes
Talk, and your words appear as you speak. Turned into text in the browser, on your computer.
⌥⇧VAlt+Shift+VAlt+Shift+V
Walkthroughs of a tab
Show something once, talking as you go: the video, your voice and a picture of every step.
⌥⇧UAlt+Shift+UAlt+Shift+U
Links and selected text
Save the page, a link or the text you selected from the right-click menu, with where it came from.
Topics and Focus
Focus is the topic you are on. Everything you capture goes there until you change it. The toolbar icon opens the side panel.
Hand it to an agent
Copy one prompt that says what you recorded and what to do, or download a pack: README, words, pictures, steps.
Change the keys at chrome://extensions/shortcuts.
Feedback with the picture marked up
The quickest way to tell an agent what is wrong on a page: show it, mark it, say why.
Open the editor
Choose Feedback from the toolbar button, or right-click the page and choose Give feedback on this page. A screenshot of the tab opens in the editor beside it.
Mark it up
Arrows, boxes, a pen and labels, in a few colours. Numbered pins put a note on the exact spot. Crop to the part that matters. Undo is ⌘Z or Ctrl+Z.
Nothing private by accident
Password and card fields and email addresses on the page are found and hidden before you start. Drag the blur over anything else. Hidden parts are grey blocks in the saved picture; what was under them is not kept.
Say what it is, and save
Bug, idea, question or praise, and a note on what you expected. It goes to your Focus topic with the page, the browser and the window size. Save, take another adds the next shot to the same piece of feedback.
Hand it off
Pick what the agent should do (fix it, write a bug report, make tickets, sum it up) and copy one prompt, or download the pack: a README, feedback.md with every note and pin, and the marked-up pictures. Switch on Errors and the console errors and failed requests the page reported go with it.

## What I want Fix what I marked on these screenshots in the code you are working in. 1. Look at each screenshot. The numbered pins point at exactly what each note is about; arrows and boxes show the area. 2. Use the page address, the errors the page reported and the failed requests to find the code behind it. 3. Reproduce each bug before changing anything. If you can't, say so instead of guessing. 4. Keep ideas and questions apart from bugs, and don't quietly drop them.
Walkthroughs, in the tab
Show it once, talking as you go. The agent gets every step.
Start it in the tab you want to show
⌥⇧UAlt+Shift+UAlt+Shift+U, or Walkthrough from the toolbar button. The first time, Chrome asks to let it record the tab and see the pages it visits.
The video is of the tab you started in; pictures and clicks follow you to other tabs in that window. Chrome doesn’t let extensions record some pages, such as chrome:// pages and the Chrome Web Store. A walkthrough won’t start on one and says so, and a tab like that you switch to during a walkthrough is noted as a visit, with no picture.
Click through and talk
A small bar on the page shows the time, with pause and stop. Each click is written down by the name of the button or link, and a picture is taken after each click and each new page, on every page the tab goes to.
What it never keeps
What you type. A password field is noted as “password field” and nothing more. Password and card fields, and email addresses on the page, are hidden in the pictures before they are stored.
Stop, and it is a pack
The video, your words, the steps and the pictures, laid out the way the desktop app lays them out, so an agent reads one the same as the other.
- 2026-11-05 1422 Checkout shows dollars/
- README.mdstart here: what happened, what I did
- transcript.mdevery word, with the clicks between
- steps.mdeach step: page, words, what was clicked
- steps/001.png … 006.pnga ring at each click
- video.webmthe tab, as recorded
- audio.wavyour voice
- manifest.jsonthe same, as JSON
_00:12 · Clicked button "Pay $37.45"_ ### Step 3 · 00:13 · Checkout · Demo Shop  - **URL:** http://demoshop.example/checkout - **Captured:** click (after it settled) - **Said:** And here it says dollars, but the basket was in pounds.
An excerpt written for this page, in the shape the pack uses. Yours carry your own words.
It stays in your browser
Captures live in your Chrome profile, on this computer. Startwell Ventures never receives them, and there is no account. It records nothing until you start a capture.
The one download it makes is the speech model you choose, once, from Hugging Face. That sends none of your captures. There are no analytics and no tracking.
Kept in the browser
- Screenshots, voice notes and walkthroughs
- Speech to text
- Your notes, links and topics
- Hiding passwords, card fields and emails
- Packs and hand-off prompts
Leaves only when you
- copy a capture or a prompt and paste it
- download a pack
- pair with the desktop app, which then hears about the page in front, over a connection that stays on this computer
Add it today
It isn’t on the Chrome Web Store yet; the listing is on its way. Until then Chrome loads it from a folder, which takes about a minute.
Download it
Download the zip and unzip it somewhere it can stay, such as your Documents folder.
Download the zipOpen Chrome’s extensions page
Paste this into the address bar, then switch on Developer mode at the top right.
chrome://extensions
Load the folder
Click Load unpacked and choose the folder you unzipped (the one with manifest.json in it).
Pin it, and say hello
Pin Startwell Capture from the puzzle-piece menu. Its welcome page opens: allow the microphone and choose a speech model. It downloads in the background, so carry on.
A copy loaded from a folder doesn’t update itself, and the store version will be a separate install that starts empty. Download packs of anything you want to keep before you switch.

With the desktop app
If Startwell Capture also runs on this computer, pair the two. The desktop app then knows the page you are on and which tabs are open, so its own captures say where they came from, and it notices a call in a browser tab. On Linux with Wayland, the desktop app often can’t read page addresses itself, so this is how it gets them.
They talk over a connection that never leaves this computer. Private windows are never sent. Your captures in the extension stay in the extension.
In the desktop app
Open Settings, Browser add-on, and click Show a code next to Pair a browser.
In Chrome
Open the side panel’s Settings, type the code under Desktop app and click Pair. Chrome asks to let it read your tabs and reach the app; that is only asked for here.
Done
The desktop app shows Connected. Once for each browser. Unpair from either side.
Questions
Do I need the desktop app?
No. The extension works on its own, with no account. If you also use Startwell Capture on your computer, you can pair them: see above.
What can it capture that the desktop app can’t, and the other way round?
It sees what is in Chrome, so it works on any computer Chrome runs on, including ones where the desktop app can’t read the screen. It can’t see other apps, record meetings, or hear the other side of a call. For those, use the desktop app.
Which speech model should I pick?
The welcome page picks the one that suits your computer, and you can change it in Settings. Tiny (about 100 MB) is fastest; bigger models are more accurate and slower. English-only models do better at English. A model downloads once from Hugging Face and then works offline.
Does it work in Edge, Brave or Arc?
It is built and tested in Chrome. Other browsers built on Chromium load extensions from a folder the same way, but we haven’t tried them yet. Tell us if you do.
Where are my captures, and how do I delete them?
In your Chrome profile, on this computer. Delete a capture or a topic in the side panel, or remove the extension to delete everything it stored. Download a pack first for anything you want to keep.
Does it run in incognito windows?
No. It doesn’t run there, and a paired desktop app is never told about a private window.
Is it free?
Yes. No account, no licence key, no trial, no analytics.
Show it, say it, hand it over.
Free, from Startwell Ventures. Everything stays in your browser.