Record the screen. The controls tell you the truth.
Screen recording on Windows, and the picker below is a working one. Turn on system audio and read what the switch tells you. Press Start, then try to throw the take away without meaning it.
Record
In the app a Source list sits under these three and fills with the monitors and windows actually on your machine. A web page cannot see them, so this demo stops at the kind.
Start will record the whole screen, video only.
Nothing is captured. This is a replica of the panel, not a recorder.
The app also puts a webcam bubble toggle, a microphone device select and a Cancel button on this panel. None of the three can do its real job in a browser tab, so none of them is drawn here.
Record
A pill that
stays out of
the shot.
Timer, pause, webcam bubble and a hold-to-confirm discard, in one strip of chrome you drag wherever it is out of the way. The timer sits in a column sized for its widest reading, in tabular figures, so the buttons beside it hold still while the digits change. Paused turns the dot amber and stops it pulsing, so colour carries the state on its own for a reader who asked for less motion.
Trim by dragging two handles or by typing an exact time, with the kept length reading back as you go. The first answer is Save as a new card, which leaves the original take where it was. Replace original is offered second, and it is the one button on that bar that cannot be undone.
- Tabular timer
- Non-destructive trim
- Crash recovery
Two switches, not one
Before a take starts you pick what to record: the whole screen, a single window, or a dragged region, chosen from a list of the sources actually on the machine. Then two audio toggles that have nothing to do with each other. System audio is one switch. The microphone is another, with its own device select beside it. Turning on one has never turned on the other.
The webcam bubble opens the moment you toggle it, so you can drag it where you want it before anything is being recorded. It captures nothing until recording starts, and the preview mirroring beside it is preview only: the track written to disk is never mirrored. A 3-2-1 countdown is optional.
Every default here comes from Settings, and the source kind and the microphone you chose are remembered per machine, so the second take starts where the first one did.
Recording runs at a fixed 30 frames a second.
The warning is printed on the switch
System audio captures everything the machine is playing, which on a call is every other person on it. Their meeting app will not show a recording indicator, because the capture happens at the WASAPI layer, underneath anything a meeting client can observe. So the people on that call are positively assured nobody is recording, at the moment somebody is.
That is a fact about how Windows works, and no amount of interface design makes it go away. What interface design can do is put it in front of the person who is about to cause it. The sentence is on the toggle that turns system audio on, not in a settings page, not in a help article, and not in a tooltip you have to go looking for.
On the label, always visible:
System audio - also records others on a call
And the control’s full title, in the words it ships with:
Captures everything this machine plays. On a call that includes everyone else's voice, and their meeting app will NOT show a recording indicator. In many countries and US states, recording someone without telling them is a criminal offence - tell them first.
The warning is listed in the repository’s do-not-delete contract, which is the mechanism that stops a future redesign from tidying it into a tooltip. A legal warning that a reader has to hover to find is a legal warning that was written for the people who wrote it.
Quoted with a plain hyphen where the shipped string uses an em dash. Nothing else in either quotation is changed.
Discard asks for six hundred milliseconds, because nothing is behind it
Stop keeps the take. Discard destroys it, and there is no recycle bin behind that, so the button asks to be meant: a hold of roughly six hundred milliseconds with a pointer, or two presses inside three seconds from the keyboard.
Space pauses, and only while the pill has focus, so a space typed into the thing you are recording goes where you sent it. Escape does nothing at all, deliberately: a stray Escape must not be able to end a take.
Both guards run in the pill at the top of this page. Hold its discard button and watch the fill; let go early and nothing happens.
A crash does not take the take with it
If the app is interrupted while it is recording, the take is rescued and arrives on the board as a card carrying a Recovered badge and a duration of 0:00. The file plays fine. Nothing at recovery time knew how long it was, so the badge does not invent a number, and the badge’s own title says exactly that:
Rescued after the app was interrupted. The file is playable; its length was not recorded, so the badge above reads 0:00.
Recovered and Trimmed are data on the card rather than decoration on it. They survive a backup and a restore, so a card that came back from a crash still says so on a machine that was rebuilt around it.
One button on this bar cannot be undone
0:04.5 → 0:38.0 (0:33.5 of 1:24.0)
An exact time in either field: 4.5, 1:02.5 or 62.5. Arrow keys move a handle by a tenth of a second.
Preview, GIF, Annotate here and Burn in sit on the same bar in the app.
Trim has two range handles you can drag or arrow through, two text fields that accept an exact time written the way you think of it, a live readout, and Preview. When you confirm, the primary answer is the non-destructive one: save the trim as a new card, and leave the take you recorded alone. Replacing the original is offered second, and it is the only button on that bar you cannot undo.
A trim moves the annotations and the markers with the footage, so a mark that sat nine seconds in is still on the same frame afterwards rather than nine seconds from a new beginning.
GIF export sits on the same bar. So does Burn in, which writes a separate annotated MP4 with the shapes baked in and the effects applied, and leaves the card pointing at the original file: burning in adds a copy, it does not overwrite what you recorded.
That is what Burn in is for. On a recording, redact, blur and pixelate are drawn over the film rather than into it, and the file underneath still contains everything they cover. Burn in is the step that changes the pixels, and the markdown export says which of the two you are looking at.
Mouse positions, clicks and window focus. The steps are read out of that.
That track is a file the recorder writes beside the video while the take runs. There are no keystrokes in it, not the characters and not the key codes, and no window titles. That is a design decision rather than a gap, and it is pinned by a test in the recorder itself.
The track is what writes the Steps to reproduce section of the card’s markdown. It stops at twenty five steps, and past that the export says so in its own words: ...and N more interactions, not listed.
A list that quietly truncated would be worse than a short one.
Nothing above involves a model. The track is a list of positions and times the recorder writes to disk, and the export reads it back and numbers it.
AI · coming soon
One recording in.
Numbered steps out.
Every click will become a step: a full resolution frame, a caption written for you, and the cursor already circled where it landed. You will hand the result to somebody who has to follow it rather than a video they have to scrub.
- Every click in the interaction track will bring its own frame and caption
- The steps will export as a document, or stay as a walkthrough
- Frames will be stills, so a redaction will remove the pixels it covers
A recording is a file on your computer
The recorder ships in the desktop app, with the library, the trim bar and the markdown export around it.
For Windows today. macOS and Linux are coming soon.
Uploading one is the Sharing switch in Settings, which is off until you turn it on, and the privacy notice names everything that ever leaves the machine.