How to search your photos by memory, not just date

You remember the trip. You don’t remember the date.

You know it was the one where it rained the whole time and you didn’t care. Or the one right after your sister moved back from Chicago. Or the birthday where the cake collapsed and everyone laughed about it for a week afterward. You remember what it felt like. You do not remember whether it was March or April, or which of the fourteen thousand photos on your phone belong to it.

So you scroll. You guess a rough month and swipe backward through a wall of screenshots, receipts, and near-duplicate shots of the same sunset, hoping you recognise the moment when you hit it.

This is the actual, everyday failure of “search” in most photo apps: it searches metadata, not memory. And metadata was never the thing you were trying to find.

Why searching by date and location isn’t really searching

Every major photo app lets you filter by when and where a photo was taken. Some go further — object recognition (“dog,” “beach,” “birthday cake”), face grouping, even natural-language queries like “red umbrella.” These are genuine improvements, and for some searches they’re exactly what you need.

But none of them answer the question you’re actually asking most of the time, which isn’t “where was this” or “what objects are in this photo” — it’s “what was this, and what did it mean?”

“Beach photos from 2023” is metadata. “The beach trip where we found out the hotel had overbooked us and ended up staying somewhere better” is a memory. No amount of GPS tagging or object detection gets you from one to the other, because that context — the story — was never in the photo’s file data to begin with. It was only ever in your head, and in whatever you happened to write in a caption at the time, if anything.

What “search by memory” actually means

This is the specific gap MemoBloom’s search is built to close, and it works differently from date or tag search because it isn’t matching your query against metadata at all — it’s matching it against meaning.

Here’s the mechanism, in plain terms:

When you create a memory in MemoBloom, the on-device AI writes a narrative — a few sentences describing what happened, grounded in what it saw across your photos and whatever note you added. That narrative (plus the memory’s title) is converted into a compact numerical representation of its meaning — an embedding — using a small on-device model, downloaded once on first launch.

When you search, your query gets converted into that same kind of representation. MemoBloom then compares your query’s meaning against every memory’s meaning — not keyword-matching text, but measuring how close the two are in concept. Type “the rainy trip we didn’t mind” and it can surface a memory titled “Porto Weekend” that never uses the words “rain” or “mind,” because the narrative captured that feeling even though your search terms didn’t match it literally.

This is why you can type something loose and half-remembered — “birthday where the cake fell over,” “the day it snowed,” “trip right after the move” — and still get the right memory. You’re not searching for a filename or a tag. You’re searching for the thing you actually remember.

How to actually use it

There’s no special syntax, no filters to learn. Open the Search tab and type what you remember, in your own words:

• “the trip where everyone got sick”
• “grandma’s last Christmas”
• “the apartment before we moved”
• “the hike that went wrong”
• “the day everyone forgot the cake”

Each result shows a thumbnail, the date, and a short snippet from the memory’s narrative — enough to confirm at a glance whether it’s the one you meant before you tap in.

A few things that make results better:

Write a note when you create the memory, even a short one. The AI’s narrative — the thing search actually indexes — is richer when it has more to work with. A memory built from photos alone still gets a real narrative and is still searchable; a memory with a one-line note (“last hike before the knee surgery”) tends to produce a narrative with more of the specific, searchable texture you’ll later type into the search bar.

Search for what it meant, not just what’s in frame. “Sunset at the pier” will work, but so will “the evening we almost didn’t go out” — try describing the feeling or the context, not just the visual content, and see which surfaces the memory faster.

Give it a moment on first use. The very first search after installing the app (or after a long gap) may take a beat longer while the search model loads into memory. After that, results come back quickly.

Why this only works fully on-device

Semantic search like this usually runs in the cloud — send the query and the corpus to a server, run the comparison there, send back results. That’s how Google Photos does it, and it’s a large part of why its search is so good: it has effectively unlimited compute and a model trained on billions of images.

MemoBloom takes a different trade — the entire embedding and comparison process runs on your phone’s chip. Your query never leaves the device. Neither does the narrative it’s being compared against, or the photos underneath it. The model is smaller than what a cloud service can afford to run, which is a real trade-off — it will not be as broadly capable as Google’s. But it never needs your memories to leave your phone to work, and it works entirely offline.

For photos of a beach vacation, that trade-off might not matter to you either way. For the memory of your kid’s first steps, or your parents’ last years together, or anything else you’d rather not have processed on a company’s server — it’s the whole point. It’s the same reason MemoBloom’s On This Day feature never asks a server to decide what to show you either.

The bigger shift

Search by date solves “when was this.” Search by tag or object solves “what’s in this photo.” Neither solves the actual, recurring problem of a life with fourteen thousand photos in it: finding the memory you’re thinking of, using only the fragments you actually remember about it.

That’s a harder search problem than most apps attempt, because it means indexing meaning instead of metadata. But it’s the only kind of search that matches how memory actually works — in feelings and fragments, not filenames and folders.

MemoBloom is available free on iPhone and Android. All AI processing is on-device. No account required. No photos are uploaded or stored externally.