Keeping libraries separate

A library is a directory. Everything a collection accumulates lives in the .videre/ state directory at its root, so two collections in two different directories are already completely separate: separate database, separate per-model embeddings, separate locks, separate config, and separate caches. There is nothing to configure and no shared default to get wrong.

You pick which library a command acts on the same way every time: it is the directory you run the command in, or the one named by --library <dir>. See where your data lives for the full resolution rule.

One library can hold several folders

Scanning a tree that spans more than one folder into the same library is a different thing, and often the right one: a collection split across an internal disk and an external drive is one library rooted at their common parent. This page is about collections that should not see each other.

Two collections, side by side

videre --library ~/Photos scan          # the personal library
videre --library ~/WorkShoots scan      # the work library

videre --library ~/WorkShoots dedupe    # only ever considers work photos
videre --library ~/WorkShoots search "client logo"

Or run each command from inside the library, with no flag at all:

cd ~/WorkShoots
videre scan
videre dedupe                           # still only work photos

Each library keeps its own embeddings directory and its own locks, so the two never block or contaminate each other. dedupe on one cannot propose deleting a file recorded in the other, because it cannot see it.

What each library owns

	Per library
Database (.videre/hashes.db)	separate
Per-model embeddings	separate
Locks	separate
Config (.videre/config.toml)	separate
Thumbnail and geocoding caches	separate

A command run against one library cannot read or write another’s state, and prune sweeps only the cache entries belonging to the library it runs in, so it can never delete another library’s cached thumbnails.

What is still shared

The Hugging Face model cache, at ~/.cache/huggingface/hub/, is shared by everything on the machine unless you set HF_HOME. That is a benefit, not a leak: model weights are downloaded once and reused, while each library’s own embeddings stay private to it. See caches.

A scratch library

Because a library is just a directory, an experiment is a throwaway directory:

videre --library /tmp/videre-scratch scan ~/some-folder

Nothing you do there can touch a real collection: its database, config and caches all live under /tmp/videre-scratch/.videre/, and deleting the directory removes every trace.

Caveats

Nothing warns you when you point at the wrong library. A command run in the wrong directory, or with the wrong --library, acts on whatever library is there. The scoped output always prints N of M, so a filter matching nothing in the library you meant is visible rather than silent.

A directory with no .videre/ is not yet a library. The commands that read a library (search, stats, dedupe, and the rest) report that it is not initialized rather than creating an empty one; only scan and watch bring a library into being.

Locks are keyed by the canonical library root, so a symlink or a relative path to the same library resolves to the same locks, and two libraries in different directories never share one.