videre

A local-first tool for making sense of a folder full of photos and videos.

• find and remove duplicates (identical and near-identical)
• search your photos by describing them, like "sunset over water" or "my red car",
  and narrow the results in the same query by person, place and radius, date
  range, category, media type, rating, like or tag
• recognise faces, group them automatically, and search by person, with a face
  engine that learns from your corrections and asks short yes/no questions to
  finish naming groups
• import from Google Takeout, Apple Photos, or a Lightroom catalog
• browse it all in videre gallery, a local web UI with every file, duplicate
  review, a date drill-down, face clusters and people, a map view, and automatic
  time-and-place events, and select many photos at once to rate, tag, rotate or
  move to the Trash
• fix wrong file dates from the camera's own EXIF data
• group photos by where they were taken

Free and open source under the permissive Apache 2.0 licence: no account, no
subscription, no lock-in, and every line is yours to read, audit, or build on.

Documentation: docs.videre.sh

Why videre

Most photo apps want to own your library. They import everything into their
own storage, index it in a database only they can read, then nudge you toward
their cloud. videre works the other way round: it's a lens over a folder you
already own. Point it at a directory and you get a single SQLite file describing
what's there. Stop using it and your photos are exactly as they were.

• Ask for the photo you remember, the way you remember it. "Elena on a
  beach near Lisbon, summer 2023" is one query: a description, a person, a place
  and a date range together, across photos you never tagged or sorted.
• Your photos come in from wherever they are stuck. videre import pulls a
  Google Takeout export, an Apple Photos library, or a Lightroom catalog into an
  ordinary folder, originals rather than derivatives, dates put back.
• It never touches your photos behind your back. Nothing is moved,
  renamed, copied, or re-encoded. The only changes to a file are ones you ask
  for: videre fix-dates corrects a file's date, and rotating a photo in the
  gallery updates its orientation tag.
• Nothing leaves your machine. No account, no upload, no telemetry. The
  search and face models and the offline map are downloaded once on first use;
  after that the one network call is videre search --location "Berlin", which looks a place
  name up once and remembers it.
• Free, open, and permanent. videre is open source under the permissive
  Apache 2.0 licence. It costs nothing, has no subscription or paywalled tier,
  and cannot be discontinued out from under you: read exactly what it does with
  your photos, keep the version you have forever, fork it, or build on it.
• It won't delete anything behind your back. videre dedupe prints what
  could go and stops there. You decide, and you can look through the
  candidates in a browser first.
• Naming faces is bulk work, not a chore. videre groups faces together
  itself, so you name one group of 40 photos rather than tagging 40 photos. The
  gallery learns from each correction and asks "Is this Elena?" about groups it
  has become confident about, so one click names the next group. A face you
  named never moves on its own.
• A gallery when you want one, not a server you keep. videre gallery
  starts a local web UI for browsing, reviewing duplicates, naming people,
  exploring the map and reliving events by time and place. Stop it with Ctrl-C
  and nothing keeps running.
• An unplugged drive is not deleted photos. Libraries spread over external
  drives are ordinary. videre prune only drops a row when the file is gone
  and its folder still exists, so cleaning up with a drive detached does not
  wipe everything on it.
• Long jobs are safe to interrupt. Preparing search or scanning faces can
  take hours on a large library. Ctrl-C is fine; rerunning continues from where
  it stopped.

And because it's an ordinary command-line tool over an ordinary SQLite file, an
AI assistant can drive it directly. videre mcp hands search and duplicate
review to a model with no server to set up.

Install

On macOS or Linux with Homebrew:

brew install erhangundogan/tap/videre

Or download a binary from the latest release
and put it on your PATH. No Rust toolchain needed, nothing to compile.

Or build it yourself, which needs Rust:

cargo install videre

If you previously ran cargo install videre, that copy lives in
~/.cargo/bin and usually comes first on PATH, so it will keep shadowing a
newer Homebrew install. Check with which videre, and remove it with
cargo uninstall videre.

Intel Macs are not supported. A dependency ships no prebuilt binaries for
them, so videre cannot be built there at all, including via cargo install.

macOS is the primary platform. videre also runs on Linux, with one gap: HEIC
photos and video frames are decoded using a macOS system tool, so on Linux those
files are skipped for thumbnails, search, and face detection. They are still
scanned, hashed, and de-duplicated.

Full install notes, including the extra flag needed to build on ARM64 Linux:
docs.videre.sh/start/install

Quickstart

Start here. Your photo folder is the library: run videre inside it, and
everything else reads from what this creates.

cd ~/Photos
videre scan

That builds a database at ~/Photos/.videre/hashes.db describing what you have.
It does not change your photos. Every command acts on the library it runs in, so
the rest assumes you are still in ~/Photos; from elsewhere, name it with
videre --library ~/Photos <command>.

videre dedupe                 # list which copies could go
videre dedupe review          # ...or review them visually in a browser first
videre dedupe trash           # move the copies to the trash, once you have looked
videre dedupe undo            # ...changed your mind? put the last run back
videre dedupe delete          # ...or delete them permanently, faster, no undo
videre dedupe --kind resized  # the same photo at another size (WhatsApp copies)

videre embed                  # one-time: prepares photos for search
videre search "golden gate bridge at sunset"

videre faces                  # detect and group faces
videre gallery                # name the groups in your browser
videre search --person "Alice"

# Words to search by meaning, filters, OR, NOT and grouping, in one query
videre search '"sunset over the sea" (person:Alice OR person:Bob) place:lisbon rating:>=4 -tag:screenshot' --after 2022-06-01

Queries follow the familiar Gmail and GitHub search style:
docs.videre.sh/reference/query-syntax

The first videre embed downloads about 1.5 GB of model data, and videre faces a separate 180 MB. Nothing is downloaded until you run a command that
needs it, and both are resumable.

More: docs.videre.sh/start/quickstart

Commands

scan, import, dedupe, gallery, search, embed, faces, classify,
locations, fix-dates, mark, tag, export, prune, pipeline, watch,
status, stats, config, mcp.

Every command takes --help. Full reference with every flag:
docs.videre.sh/commands

Contributing

videre is maintained by one person and its direction is set deliberately. Bug
reports are always welcome. If you hit a real problem or want a feature you
would use yourself, so are patches: open an issue first to agree the approach.
See CONTRIBUTING.md.

License

Apache License 2.0. See LICENSE.