Problem: My Downloads and Desktop folders kept filling up with cryptically named, duplicate, or unrelated files. Over time, cleaning up and finding the right file became a pain.
Solution: It uses Sentence Transformers (SBERT) to understand the content and context of files, then automatically groups them. It runs fully offline, so you can safely classify sensitive files (finance, medical, personal, etc.). On Apple Silicon, it parses, tokenizes, and categorizes a file in about ~1.2 seconds.
You can download and test it now for free: https://github.com/taranntell/fallinorg/releases/tag/1.0.0-b...
Current version: Supports .txt and .pdf files in English; I’m working on adding more formats and languages.
Looking for feedback on: Classification accuracy, speed, pricing ideas, and potential bulk operations or integrations.
I first launched a few weeks ago and have been rapidly adding features based on early feedback. Happy to answer questions and share implementation details.
Alternative solution: treat your downloads folder as ephemeral and delete everything every few weeks.
I feel like we’re entering an age where there is going to be increasingly more data in every day lives. (Just think about every chat in your ChatGPT account)
I guess one solution is to make everything searchable and try to organize everything. Or start treating things as ephemeral.
There’s probably no right answer. E.g. the difference between people who like having 50+ tabs open in Chrome, and needing features to organize and search tabs, versus people who treat tabs as ephemeral and short lived. I’m in the latter camp, but maybe just a matter of personal preference.
Has anyone coined the term “digital hoarding” yet? :)
Hazel[0] works well for this, but automatic download folder cleanup feels a lot like it should be a stock Finder feature.
You can also access the video directly here: https://fallinorg.com/assets/demo.mp4
One option would be to let users drop custom folder in settings. These folders could have representative files in them (maybe with a custom Finder tag to identify them), then you can cluster documents by similarity like you are already doing.
Nice application! I am looking forward to see it evolve.
onnxruntime has Swift bindings[1], consider using that. Or better yet use CoreML. You'll also be able to support x86 Macs with either of those.
[1]: https://github.com/microsoft/onnxruntime-swift-package-manag...
That said, it’s kind of amazing that we can run models of ~90 MB this efficiently on our devices today — the performance has been really encouraging. Appreciate the feedback!
You could improve the situation by presenting some kind of roadmap and indicating the limit of presale or stating clearly the amount, or a minimum amount, of discount on V1 offered to presale purchasers.
I haven’t yet tried the thing but it looks interesting. It also looks reminiscent of quickly implemented Whisper or GPT-3 front ends released a couple years ago. I’d like to better understand the value you’re providing over Apple Intelligence provided APIs.
Regarding Apple Intelligence: you’re right, Apple is integrating more AI features at the OS level, but from what I’ve seen, it’s still quite limited. For example, semantic search are not really handled in a way that solves the problem. Fallinorg is built to work fully offline, across any file type, and with deeper control/flexibility than what Apple currently exposes through their APIs.
Put simply: if Apple ever does this well, great — but right now, I think there’s still a lot of room for a tool that is private, offline, and purpose-built for file management.
It’s $10. You’ve got to be joking.
IMHO, a Spotlight Importer[0] would be the way to go. A quick search found the MacOS Vision OCR[1] project, which might be able to be incorporated as an importer.
In any event, whatever OCR approach you prefer, leveraging Spotlight would obviate the need for a service to index and then find screenshots.
0 - https://developer.apple.com/library/archive/documentation/Ca...
> Email us at fallinorg@proton.me with ith your refund request and reason. We’ll be happy to assist you.
Not to be a jerk, but if copy-editing for a pre-sale AI feature that operates on my local files includes such an oversight, it doesn’t inspire confidence. I know copy-editing and coding are different domains, but I’m still put off by it.
Also, spaces in folders will absolutely break my legacy scripts on `~` at a future time when I re-run them.
Demo is not that convincing. Also, I need multilanguage support, and am not clear if OCR for PDFs before running is done by this. (Which again, needs beforehand knowledge of which language the document might be in?)
But, cheap, and pay once and offline. Will keep an eye on it.
> Also, spaces in folders will absolutely break my legacy scripts on `~` at a future time when I re-run them.
Folder names and spaces: full customization is coming — you’ll be able to name and organize them however you like.
> Demo is not that convincing.
Demo: what felt missing or unconvincing? Any feedback helps improve it.
> Also, I need multilanguage support
Multi-language: on the roadmap — which languages matter most to you?
> and am not clear if OCR for PDFs before running is done by this.
currently not automatic for scanned PDFs. I’m also validating whether sorting images, videos, and audio based on their content/meaning would provide real value. What do you think?
Thank you so much for testing it and giving all this feedback!
Would be neat if it studied your existing organizational patterns and tried to fit any changes to match it.
That said, I’d love if you give it a try, here's the download link: https://github.com/taranntell/fallinorg/releases/tag/1.0.0-b...
Can you please elaborate more on how would you set up a user define category?
Onboarding would ask user about their work, research, hobby interests. LLM could generate word lists asking user if it matches their understanding. And so on.
Also open document format (CSV, TSV too).
Rather than moving similar files into folders, I can see the OS suggesting related files based on similarity to another file, or permitting search by concept rather than keyword.
Personally, I don't think I have that many PDF and text files that organizing them manually would be a pain. The organization logic also is a miss for me, since I don't really organize my documents in buckets like "Legal & Contracts", but rather I have folders like "Car" (for my car's service records, bill of sale, owner's manual, etc.) and "Mortgage" (mortgage quotes, contracts, etc) that's housed under "Apartment".
Doesn't help that most of these documents are not in English.
I’d love to learn from your use case: what are the top 5 file types you find yourself storing most often on your Mac?
Thanks for the feedback!
IMHO, the future is Knowledge Management and for that you really need text like like formats like markdown.
You should look into porting your app into a plug-in like feature for a knowledge management tool like Obsidian. That is the kind of audience that is going to latch on to this idea quicker than anyone else.
Im thinking you will get clearer feedback from that audience. Of course I am biased since I am an Obsidian user. I dont use word processing apps at all. All my writing is done in Markdown applications and final documents are either ported to PDF or HTML.
I had a similar idea about organizing files but I dont have programming skills so Im glad you are working on something like this. My idea, which I do not mind that you steal (Probably someone already thought about this) is figuring out a way to isolate information into categories based on a current selected feeling of the day, or a project you plan on working on based on some type of schedule. It would help with isolating personal project from work project data that can cause distractions.
Im not a mac user but I watched the video and I can see what its doing. It's pretty impressive. It would be cool if we (Obsidian) users had something like this to link ideas together. As it stands right now, I think you have to do all of the linking manually. A lot of people in the Obsidian community use the Zettelkasten Method of organizing information and the linking feature that is part of Obsidian is naturally good at that.
Fallinorg is great and something I’ve also thought about quite a bit. Local LLMs probably could enable lots of organizational workflows. I can’t wait to try what you have. Thanks for sharing and working on this problem.