Search your entire library in plain language — plus faces, places, moods and the text inside documents. Every AI model runs on your own machine. Nothing is uploaded, ever.
Windows 10 / 11 · macOS coming soon · 14-day free trial, no account needed
BoxPick builds a private, on-device index of your library so you can find things the way you actually remember them.
“red bicycle against a brick wall”, “birthday cake with candles”. Semantic matching with a local CLIP model — no tags to maintain.
Faces are detected and clustered on-device. Name someone once and every photo of them is grouped — the face data never leaves your computer.
Photos with GPS are reverse-geocoded offline, so you can browse by city or region with no map service and no network calls.
Calm, festive, dramatic, cozy… a zero-shot classifier sorts your library by feel, for when you don’t have words for the shot.
On-device OCR reads receipts, screenshots, whiteboards and slides. Search the words inside the picture, not just the picture.
Four tiers — identical files, re-saved/resized copies, the same shutter press across RAW/JPEG/HEIC, and near-identical burst frames — each with a suggested keeper and a “reclaim X GB” estimate.
Resurfaces photos taken on today’s date in earlier years, entirely from your local index.
Hand-made albums that sit alongside the automatic groupings — the only things in the app you can delete, by design.
Close the app mid-scan and it picks up exactly where it left off. Optional GPU acceleration where your hardware supports it.
Pick any folders on your drives. BoxPick never moves, edits or deletes your originals.
Embeddings, faces, places, moods and OCR are computed locally and stored in a single database on your machine.
Once indexed, every search is local and immediate — on a plane, in a bunker, with the Wi-Fi off.
This isn’t a policy promise layered over a cloud service — there is no server for your photos to go to. The models are bundled with the app and run on your CPU or GPU.
The one exception: licensing. To check whether your free trial or subscription is active, the app makes an occasional call that sends only a hashed device ID and — once you subscribe — your email address. Never anything about your photos.
Cancel anytime from Stripe’s customer portal.
Two months free vs. monthly.
What happens when the trial ends? Starting a new search or indexing a new folder asks you to subscribe. Everything already indexed — browsing your people, places, moods, documents, collections, duplicates and On This Day — stays fully usable. Nothing you’ve built is ever locked away.
Install, point it at a folder, and start searching. The 14-day trial begins on first launch.
Windows 10 or 11, 64-bit · ~350 MB installer · needs disk space for the photo index (roughly 1–2 GB per 100k photos)
No. All indexing and search happens on your machine. The only network traffic the app ever makes is the licensing check, which sends a hashed device ID and (after you subscribe) your email — nothing about your photos.
Yes, completely — for indexing and every kind of search. The licensing check has an offline grace period, so a flight or an outage won’t lock you out.
A CLIP vision–language model for semantic search and moods, a compact face detector/recognizer for people, and an on-device OCR engine for document text. They’re bundled with the app and run via ONNX Runtime on CPU, with optional GPU acceleration.
Never automatically. It only reads your files. The duplicate finder’s “move to Recycle Bin” is the single action that touches a file, it’s always your explicit click, and it’s reversible from the Recycle Bin.
Not yet — JPEG, PNG, HEIC/HEIF, TIFF and WebP today. RAW support is planned; the duplicate finder is already built to pair a RAW with its out-of-camera JPEG once RAW decoding lands.
Roughly 1–2 GB per 100,000 photos, stored in one database file on your drive. Your original photos are untouched and not copied.