every number below is from a real run

channel-miner

Read a YouTube channel instead of watching it. Runs entirely on your own computer.

videos found
0
hours of video
0
language
detecting…

Nobody was asked which language the channel speaks — it is read from the videos.

transcribed
170 / 179
lines of text
328,391
on disk
237 MB
txt/every video as readable text, with timestamps
vtt/the original captions, untouched
index.csvone row per video — id, date, length, title
segments.jsonlthe same text in ~40-second pieces, for searching
1F427wokeU8 · 2024-01-11 · a 4h02m devstream · found at 01:39:36
[01:39:30] images uh so we we do this for legal
[01:39:33] reasons um so that we we don't actually
[01:39:36] have to ship any copyrighted assets or
[01:39:40] anything like that um so everything that you see

One sentence, said once, an hour and a half into a four-hour stream from two years ago. Nobody was ever going to find that by watching.

findings.mdonly what your rubric does not already cover

Optional. Tell it what you already know, and a model on your own machine reads the whole corpus against it. Most videos produce nothing — that is the point.

./setup.sh <channel-url>

That is the only thing you supply.

github.com/anthonygozzini/channel-miner