Recording → document
A screen recording becomes a document: only the moments the screen changes, each with the exact time and with what was being said. It proves the test was run, and it feeds an AI. It runs entirely in your browser — nothing leaves your computer.
No sign-up, no install, no size limit. The video is processed inside your browser and is never uploaded to any server.
An hour-long video is not attachable anywhere: it does not fit in a ticket, it does not fit in minutes, and nobody will watch it. Claude and ChatGPT refuse video files; Gemini accepts them and samples one frame per second, blind — 3,600 images and over a million tokens, almost all of it the previous screen again. Taking screenshots by hand has the opposite flaw: it is laborious, and the time of each one is lost.
One job, two destinations
Screenshotting every step and writing the caption underneath. Or explaining to a model what happened in an hour of video. Same problem — turning time into a document — and that is why one tool solves both.
A test with no evidence does not survive an audit. Today that means Print Screen → paste into Word → write the caption, forty times. Record once and get the document ready.
A model that watches video samples one frame per second, blind. An hour becomes 3,600 images and over a million tokens — and you pay again on every question.
The bridge between the two: what you want is the same — what happened on screen, in order, with the time, and with what was said. Only the reader changes at the end: an auditor or a model.
How it works
Drag the file onto the page. It is read straight from your disk — 300 MB or 3 GB makes no difference, because nothing is uploaded anywhere.
Transcribe it right there, with Whisper running on your own machine, or drag in a .vtt / .srt caption file you already have.
Review the frames, discard the bad ones and download the document. Along with it comes the text that tells the model how to read that PDF — that is what changes the quality of the answer.
Instead of grabbing a frame every ten seconds, it scans the video and stops at the moments the screen actually changes, to within half a second. In a slide deck that means one frame per slide instead of forty of the same one.
Every screen carries the moment it happened. That is what separates a pile of screenshots from a document that works as evidence — and what lets the AI cite the right moment.
There is no server. The video, the audio and the transcript are processed in your browser — including confidential meeting recordings.
The ready-made text tells the model to cite timestamps, separate what it saw from what it heard, and admit when the information is not in the document instead of inventing it.
Privacy
Most online video tools require an upload: your file goes to someone's machine, is processed there and stays stored for a period you do not control. Walkstamp does none of that — the page is a static file, and all processing happens in your browser.
There is no account, no database, no tracking. The only external connection is downloading the transcription model, and only if you use automatic transcription. Read the full policy.
No. It opens in the browser and works. An up-to-date Chrome or Edge gives the best experience, because automatic transcription uses the graphics card through WebGPU.
There is no limit imposed by us, precisely because there is no upload. The practical limit is your computer's memory. Long videos take longer to scan, and you can stop the scan whenever you want.
MP4 (H.264), WebM and MOV in most cases. Formats the browser cannot decode — HEVC and some MKV files — will not open; those need converting to MP4 first.
It uses Whisper running locally. The fast model is reasonable and the accurate one is considerably better, at the cost of a larger download. If you already have captions from another service, drag the file in and skip this step.
It is free, and the reason is structural: your computer does the work, so serving you costs me nothing. No ads, no sign-up, no data of yours being sold — not out of generosity, but because there is no server to store anything on.
Any model that accepts attachments and reads images — Claude, ChatGPT and Gemini all work. Attach the PDF and paste the prompt the tool generates alongside it.
Three ways, each a minute long, none of them requiring you to trust me. Open the developer panel (F12), go to the network tab and process a video: you see every request the page makes — and none of them carries your file. Second: the code is a single page; right-click, view source, and it is all there to read. Third, the most direct: download the offline version, turn off the internet and use it. If it works with no network, it is not sending anything anywhere.
It takes less time than reading this page.
Open the tool