Crate videre_core 

Modules

atomic_file

Atomic publication for rebuildable cache and local state files.

classify

Classifications table: one row per unique content hash (photo/screenshot/ document/meme/unknown), keyed to embeddings.hash. Zero-shot classification reuses embeddings videre embed already computed. See docs/superpowers/specs/2026-07-29-screenshot-document-classification-design.md.

db

decode_failures

Records files that a decode stage tried and could not turn into pixels, so a guaranteed failure is not re-attempted on every run.

disk

What videre is using on disk, and which part of it is what.

embeddings

Embeddings table: one row per unique content hash, keyed to file_hashes.hash.

embeddings_db

Per-model embedding databases: one SQLite file per (library, model) pair, attached to the main connection as emb.

face_cluster

face_db

face_match

Bounding-box geometry for matching XMP face regions to detected faces. Pure: no I/O, no database, so the matching rule is tested in isolation.

fix_dates_target

The fix-dates target transform, shared by [videre fix-dates](the command that applies it) and videre status (which reports how many rows would change without applying anything).

geocode

Forward geocoding: place name (e.g. “Berlin, Germany”) -> (lat, lon), via the free public Nominatim (OpenStreetMap) API, with a local cache so a repeated query never repeats the network call. See docs/superpowers/specs/2026-08-01-location-clustering-design.md.

heic

hf_cache

Where Hugging Face weights land locally, and whether they are already there.

image_decode

Orientation-aware decode of source image files.

import_location

Where an import’s source files live.

import_providers

Which import sources exist, and how to recognise them.

io_timeout

library

The explicit identity of one media library.

library_config

Library-scoped settings: the type, its built-in defaults, and the config.toml under the library’s reserved state directory.

library_db

The library database: the schema it is written with, the one entry point allowed to create it, and the reader path that never creates anything.

library_guard

The guard between user-supplied path filters and one selected library.

library_io

Originals and sidecars reach the disk only through the pinned root.

library_locks

Library-scoped advisory locks: how videre processes coordinate around one library, keyed by the library’s own state directory instead of a process-global home.

library_state

Small key-value state for library-scoped bookkeeping that is neither a pipeline run nor a setting: one row per key, integer values, table created on first use. Used today for the face recluster watermark; kept generic so the next piece of watch state does not grow its own table.

library_stats

Aggregate library statistics for dashboard-style callers. Plain queries over an open rusqlite::Connection, shared source of truth for the gallery’s stats tile, videre stats, and any other embedder. See docs/superpowers/specs/2026-07-31-dashboard-stats-backend-design.md (Pass A) for what is and isn’t in scope.

location

location_cluster

GPS coordinate clustering: groups nearby (lat, lon) points by haversine distance into named location clusters, persisted to location_clusters + file_hashes.location_cluster_id. See docs/superpowers/specs/2026-08-01-location-clustering-design.md.

marks

Per-photo marks: rating, pick, colour label, like. One row per photo, keyed by content hash so a mark follows a photo across duplicates and moves.

mime_probe

Identify a file’s real type from its leading bytes.

person

Two forms of a person’s name: one to identify them by, one to show.

person_search

pipeline_runs

Per-command pipeline run history and liveness, surfaced by videre stats and other dashboard-style callers. See docs/superpowers/specs/2026-07-31-dashboard-stats-pass-b-design.md for the full design, and in particular why track_in() below does not rely on Drop/RAII for the success/failure bookkeeping: the library-scoped command lock’s release does, backstopped by the OS releasing flock on any process death.

progress

query

Composable search predicates.

selection

Shared vocabulary for saying which files a command should work on.

semaphore

status_report

The shared status model: one producer for everything that reports pipeline state, so the CLI, stats, and the MCP tools cannot drift into telling different stories about the same library (the duplication this closes was flagged as DEBT:8).

tags

Free-form tags, stored many-to-many by content hash so they follow a photo across duplicates and moves, exactly like marks. A tag is one opaque string; a /-separated path is stored verbatim with no hierarchy semantics (that is a deferred follow-up). Mirrors the shape of videre_core::marks.

thumb_cache

vectors

f32 <-> f16 BLOB conversion and L2 normalization for stored embeddings. Storage format: little-endian f16, 2 bytes per dimension.

video_meta

Date, location, dimensions, duration and codec from a QuickTime/MP4 container, without decoding anything.

video_probe

Detect whether a QuickTime/MP4 container has a video track, without decoding anything.

work

One place that decides whether there is work, and therefore whether a model is ever loaded.

xmp_gather

Collects the non-face owned labels for one photo hash (image dimensions, resolved location name, zero-shot category) that a sidecar export needs. Faces are gathered separately via face_db::labeled_faces_by_hash. Returns plain data; the binary assembles the XMP model, because that type is binary-only.