I Remember What’s in My Screenshots — Just Not Where I Saved Them
There's a particular kind of frustration I keep running into:
I know I saved an image of this. Where is it?
Then comes the scrolling.
Opening folders.
Checking dates.
Looking through thumbnails.
Trying to remember filenames I never really chose in the first place.
Screenshots are where I notice this most. Over time, they pile up, and finding one again becomes surprisingly difficult.
Eventually, I realized something:
I often remember quite a lot about the image I'm looking for. I just don't remember anything useful about the file.
I might remember an error message.
A product name.
A piece of code.
A phrase that was somewhere on the screen.
But I don't remember the filename, the date, or exactly where I saved it.
That's why I'm building Capixe.
The idea is simple:
Find images by what you remember about them, not by how you stored them.
I'm starting with text.
The first Windows prototype lets you choose a folder, analyze the images locally, and search for words contained inside them.
Choose a folder → Analyze → Search
So if you remember a word, an error message, a product name, or something else you saw in an image, you can use that memory to find it again.
The image analysis and search happen locally, so your images don't need to be uploaded just to make them searchable.
Text search is only the starting point. Longer term, I'd like to explore finding images from meaning and context too — even when you don't remember the exact words.
But before building further, I want to find out whether the basic idea is actually useful outside my own image library.
The first prototype is available for Windows:
Capixe:
https://p1rworks24-ops.github.io/Capixe/
If you have a folder full of screenshots or other images you've accumulated over time, I'd really appreciate you trying it.
But even if you don't try it, I'm curious:
Do you actually have this problem? When you can't find an old screenshot, what do you usually remember about it?
And if you do try Capixe:
What was the first image you searched for — and did you find it?
That reaction is exactly why I do this work. The zero is a finding, it just isn't the fun kind. And your instinct on where to look next is right, complaints live where people fight the problem, not where they rate a viewer app. Reddit threads about lost screenshots will be noisier and more honest than Store reviews ever get. Small ask, I want to write this exercise up as a post at some point, mind if I quote your line about not forcing the data to fit the idea? Best one sentence description of the job I've heard.
Absolutely, feel free to quote it!
Honestly, this whole exercise has been a good reminder for me too. It’s tempting to look for evidence that supports what you’re building, but the “zero” is useful data too 😅
And I agree about Reddit — that’s probably where I should dig next. Thanks for the suggestion!
The useful job is not screenshot storage, it is retrieval from partial memory. I would test searches based on remembered text, app, color, and rough date, then measure whether users find the right image without manually tagging anything. Local-first indexing could also turn privacy into a real differentiator.
Thanks, “retrieval from partial memory” is a really useful way to frame it.
Right now I’m starting with filename and text inside images, but I’d like to expand that toward signals like app, color, rough date, and semantic meaning — so users can find something based on whatever they happen to remember, without manually tagging it first.
I also agree on privacy. My current direction is local-first rather than local-only: the core search and indexing should work locally without uploading images, while more advanced AI features could be optional for things like deeper search, automatic organization, natural-language actions, and interaction.
I think the key is making it clear when something stays local and when an optional AI feature uses an external service.
You asked "do you actually have this problem" and there's a bigger answer available than replies here. Apple Photos and Google Photos both do text search in images now, and their reviews show exactly where people hit the wall with it. What they search for, what fails, what they wish existed. Plus there's a whole genre of complaints about Windows Photos NOT having it, which is basically your market raising its hand.
I mine app reviews for a living (well, for a side project) so take this with salt, but the pattern level answer to your question is yes, the pain is real and well documented. The open question your prototype actually tests is whether local + Windows + folder based is the right wedge, and nobody's reviews can answer that one for you.
Happy to show you what a full review mining analysis looks like if it would help you scope v2. Either way, validating before building further already puts you ahead of most of this forum.
Thanks, this gave me a completely different way to think about validation. I hadn’t considered looking at Apple Photos, Google Photos, and Windows Photos reviews to understand what people actually search for and where existing solutions fall short.
Your point about the real question being whether local + Windows + folder-based is the right approach is especially helpful.
I’d definitely be interested in seeing what a full review mining analysis looks like, especially what people search for and what they’re unhappy with in existing solutions. Thanks for offering!
Glad it wasn't just noise. Here's the sample, it's on a focus app called Forest but you'll get the idea of what comes out: reviewmine.gumroad.com/l/free-sample-brief
If I were you I'd be reading Google Photos and Apple Photos reviews, the desktop search complaints are basically people writing your spec for free.
Happy to run one on a specific competitor if you tell me which.
Thanks, I see what you mean now. The sample looks interesting, especially the idea of ranking pain points by evidence count.
Microsoft Photos would probably be the most relevant competitor for what I’m building, since I’m focused on Windows and local images. I’m particularly interested in complaints around finding old screenshots/images and what users feel is missing from search.
Is that the kind of analysis you meant when you offered to run one?
Done. Ran Microsoft Photos, 100 recent Store reviews, mostly the last 3 months.
Honest finding first. I went looking for your theme specifically, complaints about finding old screenshots or search falling short, and it's basically not there. Zero organic hits. People don't complain about search in Photos, they complain that it crashes or won't open images at all, roughly a quarter of reviews say exactly that, plus anger about forced cloud and AI features nobody asked for.
Two ways to read that for Capixe. Either people don't feel the search pain enough to write about it, which is worth taking seriously as validation signal, or they don't expect search from a viewer app and vent somewhere else, Windows search threads rather than Store reviews. Both worth knowing before you build.
The reliability angle is a real wedge though. Fast, local, opens anything, doesn't die on big folders. That answers the loudest documented complaint in the category.
Full brief, free, reviewmine.gumroad.com/l/free-sample-microsoft-photos. If it's useful say so, and if it's not tell me that too, publicly, I can take it.
This is genuinely useful — especially the zero.
I was expecting at least some complaints about search, so finding none in 100 recent reviews is an important signal for me rather than a bad result.
Your point about people possibly not expecting search from Microsoft Photos also makes sense. I think my next step should be looking at places where Windows users are actively trying to solve the problem — Reddit threads, Windows communities, and searches around finding old screenshots or searching text inside images.
The complaints about forced cloud/AI and reliability are interesting too, especially since I’m trying to keep the core image analysis local.
Thanks for running this and for not forcing the data to fit my idea. The report is useful.
This comment was deleted 19 days ago
The problem you’re describing is surprisingly real: we tend to remember the content of a screenshot, but almost never the filename, folder, or date. Traditional file search is built around how we stored something, while human memory works more like “I remember seeing an error message about X.”
Starting with local OCR/text search feels like a sensible MVP because it solves a specific pain point without immediately overcomplicating the product. I also like the local-first approach, especially for screenshots that may contain sensitive information.
The interesting next step, in my opinion, is exactly what you mentioned: searching by meaning rather than exact text. Being able to say “find the screenshot where I compared these two SEO tools” or “show me the screenshot with that blue dashboard and pricing table” could make this much more powerful than a simple screenshot manager.
Curious to see how people use the first version and what they actually search for—that feedback could reveal the strongest use case for Capixe.
Thanks, this is really helpful. I also think searching by meaning could become much more powerful than exact text search.
For now, I want to keep the first version focused on local OCR/text search and see what people actually search for before deciding what to build next.
Out of curiosity, if you had this kind of search today, what types of screenshots or images do you think you’d use it for most?
The distinction between remembering what was in an image and remembering anything about the file itself is a very relatable problem. I’m curious how often people actually search for screenshots this way once they have the tool.
That’s exactly what I’m curious about too. The problem feels real, but I don’t know yet how often people will actually search this way in practice. That’s one of the main things I want to learn from the prototype before building further.