This project draws heavily from two other projects: one called Rewind and another called Windows Recall. However, unlike both of them, Memos allows you to have complete control over your data, avoiding the transfer of data to untrusted data centers.
In short order, this will create a large corpus of unsecured local data.
Is the user expected to secure the data independently?
Do Recall/Rewind help the user to filter recorded data for retention or deletion?
However, I still have a question about this: it seems that lots of hard disk is already encrypted. After all, I also store a large amount of personal photos, documents, bills, and other important information on my computer, and I haven’t meticulously encrypted all this data again. Should I be doing that?
Full disk encryption targets a different threat model - disk encryption protects against someone grabbing your computer.
Writing into an encrypted blob on disk adds a layer of protection against bad actors exfiltrating data by running code on the laptop.
Overall I really am amazed that this sort of thing is now possible and appreciate a privacy-aware / local compute and storage version of it!
Rewind does glitch sometimes specifically with audio recording which is extremely annoying. You go back to an area where you thought you had audio notes only to find you didn’t - even though you had audio recording turned on the whole time. It has something to do with meeting detection. Which is silly bc disk space is cheap just auto record. I do like the concept of an open source version and I will look into this.
I guess it's worse in the sense that it also records audio, but large corpus of information is already at risk on a unsecure or compromised devices
Surely you encrypt your disk rather than trying to secure this one app? I mean there’s far more valuable stuff to on your machine than anything this app could possibly store
1: https://github.com/arkohut/pensieve/commit/e81057d5bebcf9cab...
Unless they've done something very very wrong performance will be fine. This isn't doing anything where python's overhead would matter.
It's glueing together some highly optimized code written in other languages, or using python as a DSL to interface with highly optimized libraries like numpy, or generate highly optimized assembly with something like JAX, or if they're really fancy compiling a restricted subset directly to GPU shaders or something.
Python is plenty fast for most stuff, and when it isn't it has one of the best pathways towards optimization.
1. OCR model
2. Embedding model
3. VLM model (optional)
I hadn’t used it since installing it. So that came as a surprise.
I then tried using it, and couldn’t get it to find things I knew were in my history. (Basic keywords match)
So I deleted it.
I like the idea of this app. It ticks all the boxes. But I haven’t found any value on this category of app yet.
Of course, I know that such a feature might seem trivial. Some things are simply forgotten and that’s fine. But what if it is a more important clue, like a bug for the web site which only trigger in some narrow condition and hard to reproduce.
I came away quite glad I don’t suffer from a photographic memory, and while I applaud the project, I prefer to forget things.
Even if you wrote it in a diary it could still be used as evidence against you. The only untouchable place is your mind.
I’m personally fascinated by this sort of reactions.