Launching today
BeaconFind
Find any file by what's inside it. Private, on-device AI.
12 followers
Find any file by what's inside it. Private, on-device AI.
12 followers
BeaconFind is an AI file search app for Windows, press Ctrl+K to find files on your PC by meaning, not just filename. 100% private & offline. Free 14-day trial.
running the models locally for privacy makes sense, but I'm curious about the indexing side of that tradeoff. on a drive with a couple hundred GB of documents, does BeaconFind keep a background process re-indexing as files change, or is it more of an on-demand scan? asking because if I save a new file and search for it 30 seconds later, I want to know whether it'll actually show up yet or if there's a lag before it's searchable.
@galdayan Great question, and the right one to ask.
It's a background file watcher, not a periodic scan. BeaconFind hooks into the OS file-change events for the folders you've indexed, so it isn't crawling your drive on a timer. When nothing changes, it does nothing, idle CPU is basically zero. The only heavy pass is the very first index, where it has to read and understand everything once. On a couple hundred GB that takes a while, but it's a one-time thing.
After that it's incremental and per-file. When you save a new file, the watcher sees it, waits a couple of seconds (so it's not indexing a file that's still being written), then processes just that one file. So your "save it, search 30 seconds later" case: yes, it'll be there. A normal document is searchable within a few seconds. A giant scanned PDF that needs OCR can take a little longer, but you're talking seconds, not minutes.
Two things that come up next, so I'll pre-empt them: it eases off while you're actively using the machine and pauses on battery, so indexing doesn't fight you for resources. And if the app was closed when you changed files, it catches up on the next launch by checking what changed since it last looked, so nothing quietly falls through the cracks.
Happy to go deeper on any of it.
@onat_atmaca that answers it, thanks for the detail. one more - is the index itself stored per-drive, so if I plug in an external drive with its own couple hundred GB, does that trigger its own full first-pass, or does it share some model/cache overhead with the main index so subsequent drives index faster?
@galdayan Good follow-up. It's one unified index, not per-drive. Everything you point it at, whether that's an internal disk or an external one, goes into a single encrypted index that lives on your system drive, not on the drive being scanned. So you search across all of it at once and it doesn't matter which physical disk a file sits on.
On the overhead: the models load once when the app starts and stay resident, they're not reloaded or re-downloaded per drive. So plugging in that second couple-hundred-GB drive doesn't pay the setup cost again. Its first pass is purely the read-and-embed time for the new files it hasn't seen yet. Anything whose content it already indexed elsewhere gets skipped by a content hash, so duplicates across drives are basically free.
So in practice the very first drive you index is the slowest, because that run also covers the one-time model download and warm-up. Every drive after that is just the raw cost of the new content, with everything already warm.
On-device search by content instead of filename is exactly what most people actually want but rarely get without sending data to the cloud. How does it handle non-text files — images, PDFs with scanned content, that kind of thing?
@ark_y_k The "without the cloud" bit is exactly why it's rare. Doing this understanding locally is just harder to build than firing it off to a hosted model.
Two cases:
Images get embedded into the same vector space as your text query, so "whiteboard covered in sticky notes" lands right next to the actual photo and it ranks by similarity. A local vision model also captions each one in the background, and EXIF (date, camera, GPS) comes along for free.
Scanned PDFs and screenshots are an OCR job: the text is pixels, so it reads the words back out on-device, and after that they index like any normal document.
All of it on your machine, which is the tedious part most tools skip.
@onat_atmaca Doing that locally instead of just calling out to a hosted model is a real engineering commitment — most teams would take the easy route and ship the cloud version. Embedding images into the same vector space as text queries is a clean approach; does that mean a search like "photos of my dog at the beach" would actually work against unlabeled images?
Hey Product Hunt 👋
BeaconFind is a desktop app that searches inside your files by meaning, not just the filename. Press Ctrl+K, type what you remember ("the lease with the pet clause", "invoice from the plumber"), and it finds the right file even if you saved it as scan_0043.pdf. It works across documents, images, screenshots, scanned PDFs, 70+ file types and 50+ languages.
It's a proper native app that runs a handful of small AI models locally, right on your PC. Nothing gets uploaded, there's no account, and the index is encrypted on disk. For a tool that reads everything you own, keeping it all on your machine felt like the only honest way to build it.
Honest bits up front: it's Windows-only for now, and it's paid, with a 14-day trial (€19 lifetime during launch).
I'll be here all day, so ask me anything. Happy to get into the weeds :)
Local AI products seem to be reaching the point where workflow matters more than privacy alone.
Curious—do users initially adopt BeaconFind because it's on-device, or because finding information becomes noticeably faster than existing search tools?