Search video
like it's text
Drop a video in and VidSeek scans it inside your own tab. It writes down everything that gets said, grabs a frame every few seconds, and builds an index out of it. Then you just ask it a question in the chat and click a timestamp to jump there.
✓ your file never leaves this device
1 match, click to jump.
What it does
[ six things, zero servers ]Ask in plain english
“find me where they argue about money.” you type something like that, it ranks every frame and every line of transcript, and you get back timestamps you can click.
It listens to the video
Whisper transcribes the audio track with timestamps, so “where she says hello” gets you to the second she actually says it.
It watches the video
CLIP grabs a frame every few seconds and turns it into a vector. questions like “a city at night” get matched against what's actually on screen.
Your GPU, not a server
It runs on WebGPU when your browser supports it, WASM when it doesn't. the first scan downloads the models and the browser caches them after that.
Nothing ever uploads
The file is read into the page and stays on your computer. there's no server sitting in the middle holding it, and closing the tab wipes everything.
No sign-up, no limits
you don't make an account, nothing counts your scans, and there's no "upgrade to pro" popup popping up halfway through.
How it works
Drop a video
MP4, WebM, MOV all work. pick a file or drag it in, it's read from your disk straight into the tab.
Let it scan
Whisper works on the audio while CLIP checks the frames. first run downloads the models (around 100 mb) and after that they're cached.
Ask, then jump
ask a question, get a few hits back with timestamps and the quote it matched. click one and the player jumps there.
Under the hood
no api keys and no database anywhere. the index just lives in memory while the tab is open, so closing it wipes everything.
Ready to find
things?
Open the studio, drop a video in and see what it finds. it all runs on your machine, so nothing gets uploaded anywhere.