I feel like a simple automatic capture of timestamp + url + screenshot would already be very useful. This gives you a visual memory of the things you've seen on the web. I've wanted to develop this for a while, as a browser plugin.
Being able to skim the past month or two click around the thumbnails would already be amazing. I've wanted to do that many times before to check if my memory was correct, or if a page changed since I last saw it, or figure out when I last saw something online.
You don't need a special viewer for it, as your operating system's file explorer can view the screenshots already, and you don't need to set up a crawl. Screenshots also compress well, as webp or png after crunching it.
A few years ago, in an attempt to increase productivity, I used a screen recorder that took a screenshot every 10 seconds and played it back at the end of every day. So I had a timelapse of how I was spending my time -- mostly online. It was very enlightening.
The most efficient format to store a sequence of screenshots in is video, because most of them will have heavily overlapping data.
Huh. That's a pretty nifty thing to do. Just wrote a python script to do that for me and it's running in the background right now. Shall be interesting to come back to it today in the evening. Do you recall how many seconds each screenshot would show itself in the final video (basically what was the framerate?). Currently considering about 4 frames per second but would love to get your take on it :)
It's up to preference really. I had mine set to max frame rate, then I'd use the [] keys in VLC to slow down the speed. It would take 1-2 minutes to view my day (8-12 hours).
I think at one point I used an ImageMagick script to add timestamps to them.
Originally I used a Windows thing called TimeSnapper which as a bonus lets you scrub through time and shows when there was / wasn't computer activity.
I've dreamed about this as well, basically a personalized FullStory that allows you to search and replay all of your sessions across sites.
Easy block list for sensitive things like banking, internal sites, email, etc.
I currently use the Session Buddy Chrome Plugin, which helps in some cases (I was able to find a hard to Google repo today, for example), but the historic context is largely missing.
You'd think so, but code (even web code) needs to be executed and is brittle. Some of it doesn't even work as an archive even right after saving. All my images from 10-20 years ago work perfectly today. None of my code does without some major effort.
It's a hard problem to figure out what's readable text on a page, and what isn't. Even Google has a hard time figuring that out. OCR works very well with screenshots, and is purely computation time. But the real reason is generally just having timestamps, urls, and screenshots is good enough. I usually remember about when it was, and some words in the url, and don't need the heavyweight text search setup.
Trying to parse the SPAs of today is just painful. Simpler to just render the page screenshot and OCR! Guaranteed to only index text that actually matters
Comments
I feel like a simple automatic capture of timestamp + url + screenshot would already be very useful. This gives you a visual memory of the things you've seen on the web. I've wanted to develop this for a while, as a browser plugin.
Being able to skim the past month or two click around the thumbnails would already be amazing. I've wanted to do that many times before to check if my memory was correct, or if a page changed since I last saw it, or figure out when I last saw something online.
You don't need a special viewer for it, as your operating system's file explorer can view the screenshots already, and you don't need to set up a crawl. Screenshots also compress well, as webp or png after crunching it.
A few years ago, in an attempt to increase productivity, I used a screen recorder that took a screenshot every 10 seconds and played it back at the end of every day. So I had a timelapse of how I was spending my time -- mostly online. It was very enlightening.
The most efficient format to store a sequence of screenshots in is video, because most of them will have heavily overlapping data.
Huh. That's a pretty nifty thing to do. Just wrote a python script to do that for me and it's running in the background right now. Shall be interesting to come back to it today in the evening. Do you recall how many seconds each screenshot would show itself in the final video (basically what was the framerate?). Currently considering about 4 frames per second but would love to get your take on it :)
It's up to preference really. I had mine set to max frame rate, then I'd use the [] keys in VLC to slow down the speed. It would take 1-2 minutes to view my day (8-12 hours).
I think at one point I used an ImageMagick script to add timestamps to them.
Originally I used a Windows thing called TimeSnapper which as a bonus lets you scrub through time and shows when there was / wasn't computer activity.
I did it before just saving as images and using feh to play with the playback speed. It's easy to play with so just see what works for you
I've dreamed about this as well, basically a personalized FullStory that allows you to search and replay all of your sessions across sites.
Easy block list for sensitive things like banking, internal sites, email, etc.
I currently use the Session Buddy Chrome Plugin, which helps in some cases (I was able to find a hard to Google repo today, for example), but the historic context is largely missing.
Sounds like this might be what you want if you only want screenshots + timestamps and nothing else:
orWouldn't it be more useful and take less space to use SignleFile?
You'd think so, but code (even web code) needs to be executed and is brittle. Some of it doesn't even work as an archive even right after saving. All my images from 10-20 years ago work perfectly today. None of my code does without some major effort.
This doesn't allow full text search easily, though.
PDF with an image on one page, then the plain text of the page flowed over following pages.
+ (text minus stop words)
ocr?
why go from text to image and back to text? seems wasteful and error prone...
It's a hard problem to figure out what's readable text on a page, and what isn't. Even Google has a hard time figuring that out. OCR works very well with screenshots, and is purely computation time. But the real reason is generally just having timestamps, urls, and screenshots is good enough. I usually remember about when it was, and some words in the url, and don't need the heavyweight text search setup.
Just hard with the "read more" buttons.
Trying to parse the SPAs of today is just painful. Simpler to just render the page screenshot and OCR! Guaranteed to only index text that actually matters