Bring 200,000 photos in.
Point Smriti at a folder or drive. It indexes in place, never copies, never moves your originals — and quietly catches up every time you open it. iPhone HEIC and camera RAW open everywhere.
स्मृति· remembered
Faces, search, memories and maps for the photos already on your computer or external drive. The smart parts of Google Photos, without handing your library to anyone.
Featured on LINUX Unplugged #684
Your photos stay exactly where they are. Nothing is copied, moved or renamed, even at 200,000 photos.
Faces grouped into people, places named, dates fixed, duplicates and bursts found. No upload, no account.
Search, browse a person or a trip, and find new memories and collages each time you open it, even on a plane.
Most photo apps want to be a service. Smriti is a tool: nothing runs when it's closed, there's no subscription, and the whole library travels with the drive.
Tap the button. Six people. Thirty-six photos. The clustering below runs in your browser — the real app does the same on your machine with on-device face models — no cloud round-trips.
All offline. No upload, no embedding service, no API key.
Point Smriti at a folder or drive. It indexes in place, never copies, never moves your originals — and quietly catches up every time you open it. iPhone HEIC and camera RAW open everywhere.
On-device face recognition groups every photo of that person across your whole library. No cloud round-trip, no upload — clustering and review both stay on your machine.
Every geotagged photo plotted. Click a cluster to drill into the place. Tile cache lives on your drive.
Unified search across people, albums, places, and dates. Three filters, one result. Milliseconds on 250K libraries.
"This day, N years ago" and a few collages made from your photos, waiting when you open the app. Never a notification.
Exact duplicates and bursts, side by side. Keep the best, trash the rest — Trash, not delete, so you can restore if you change your mind.
Type what you remember — beach, group photo, old car. If the optional visual-search model is installed, Smriti turns your words into a local vector query, then feeds matching photo candidates through the same filters as dates, places and people. The demo is synthetic; the constraint — pixels, not filenames — is real.
Meaning, not magic. Hover a photo to see why it matched. Strong matches sit closer to the top.
The assistant isn’t a chat window glued to a photo app. It calls Smriti’s own tools — resolve people, resolve places, resolve dates, search photos, prepare an album preview — then waits. It can search and draft albums; it does not rename, move, trash or delete your originals.
You stay in charge. Assistant is opt-in, provider-backed, scoped to the open library, and album creation needs approval.
Smriti's album-suggestion engine clusters runs of photos taken away from your home city. The real app does this on indexing; below, it does it when you click. The data is synthetic, the algorithm is honest.
Found locally. No upload, no geocoding API, no “trip suggestions powered by…”
| Cloud (Google · iCloud) | Self-hosted (Immich · PhotoPrism) | Smriti | |
|---|---|---|---|
| Photos live | On their servers | On your server | On the drive they came from |
| Original quality preserved | Compressed by default (Google's "Storage saver" caps at 16 MP) | Preserved | Preserved — files never opened for write |
| Account required | Yes | Yes (you set one up) | None |
| Server to maintain | Theirs (you pay rent) | Yours, 24 / 7 | None — it's an app |
| Metadata (faces, tags, albums) | In their database | In your server's database | On the drive with the photos |
| Survives provider going away | No | Only if your server keeps running | Yes — it's files |
| Yearly attention you owe it | Renew subscription | Updates · certs · backups | None |
| Cost over five years | $250 – $750 | NAS + power + your time | $0 |
| Works on a plane | Limited | Limited (no VPN) | Everything |
| Memory footprint when closed | n/a | ~1 – 4 GB resident | Zero |
Comparison reflects each project's default setup as of 2026. Smriti's row makes no claims it can't keep. If you are weighing a server against a desktop library, here is the longer version of that decision.
Left: your original. Right: what Google Photos' "Storage saver" is documented to do: compress to save space, and resize only when a photo is larger than 16 MP. Smriti never re-encodes anything.
Original
After Google Photos
Google Takeout hands you a pile of ZIPs and sidecar files. Hand the whole pile to Smriti. It reconnects dates and places, restores favourites and real albums, collapses export duplicates, and keeps the originals on your drive.
Select every part together. Leave them zipped.
Interrupted? Run it again. Smriti resumes safely.
Smriti writes a database at
<drive>/.photovault/
— on the same drive as the photos. Unplug it. Plug
it elsewhere. Faces, tags, albums, the whole library
travels with it. Try it below.
No library connected
No library connected
Your photos never leave your machine. No telemetry, no analytics, no "anonymous usage statistics" — in the app or on this website. Here's everything Smriti ever fetches; the privacy notes list every request in detail.
When you open the Map view. Tiles are cached on your drive after first load; subsequent panning is served from cache.
tile.openstreetmap.org
The models for faces and visual search, plus offline place names. Downloaded once, checked against published checksums, then everything runs on your machine. Skip it and the rest of the app still works.
github.com · huggingface.co
If you opt in, Smriti checks for new releases once every 24 hours. Off by default. No photo data, no telemetry, just a single HEAD-style request.
api.github.com/repos/.../releases/latest
If you accept an update, Smriti fetches the matching installer for your platform and verifies its SHA-256 against the release's SHA256SUMS file. No metadata about your install or library is sent.
github.com/.../releases/download/...
Face recognition is the one slow step. On CPU, a 90,000-photo library takes ~3 hours. If you want to skip the wait, Smriti can borrow a free Kaggle or Colab GPU for the one-time pass — and finish in minutes.
It sends only 112×112 aligned face crops. Never originals, never metadata, never filenames. When the pass is done, you disconnect. Forever, not "until the next sync."
Apache-2.0 licensed. Source on GitHub. Smart features — face recognition, visual search, offline place names — are an optional one-time download, offered once your first scan finishes.
For Windows 10 and 11. Smriti is not code-signed yet: if SmartScreen appears, choose More info → Run anyway, then approve Install anyway.
Download installer →For Apple Silicon Macs. Smriti is not notarized yet; after the first blocked launch, choose Open Anyway in Privacy & Security.
Download installer →Debian, Ubuntu, Fedora — plus a portable AppImage for other distros.
Download installer →Apache-2.0 · Free forever · No sign-up · No telemetry