Every file type. Every AI layer.
A single file detail surface that adapts to images, PDFs, audio, video, Office documents, and data. Six AI extractions per file, kept in a dual AI / Reviewed pattern so curators always control what publishes.
The right viewer for the right file.
A unified detail page that knows what a TIFF is, what an MP3 is, and what an EML is — and shows the right tools, fields, and extractions for each.
Image
Width, height, DPI, color space, bit depth, EXIF — preserved on import.
Native PDF.js viewer with smart text vs scanned detection.
Audio
Styled waveform player with timestamped transcript and proofreader.
Video
Native player with VTT captions, client-side thumbnails, full-viewport proofreader.
Word / Spreadsheet
Office docs and CSV / TSV with row, column, font detection.
EML / MSG with header parsing, attachment listing, body extraction.
AI says "here's a draft." Curators say "here's the truth."
Every AI extraction lives as two versions side by side — an immutable AI snapshot and an editable Reviewed version. Each field carries its own status: pending, accepted, modified, or rejected. Nothing publishes without explicit curator sign-off.
- AI Generated — immutable, read-only, what the model produced
- Reviewed — your working copy, with pristine / dirty tracking
- Per-field status: pending · accepted · modified · rejected
- Copy from AI to seed the review, then edit by hand
- Accept All for fast triage on long extractions
The metadata an archivist actually wants.
Not just an OCR dump. Six structured extractions land on every processed file — each with its own review surface.
Caption
One-sentence AI description, with an editable reviewed version.
Keywords
Auto-extracted tags with add/remove on review and live count comparison.
Transcript
Timestamped, speaker-diarized, per-entry confidence — accept or edit per line.
Entities
People, organizations, locations, dates, events, terms — confidence + source.
Locations
Geo-extracted places with type, coordinates, and confidence.
Sentiment
Overall tone (positive / neutral / negative / mixed) with breakdown bar.
Technical metadata, automatically.
Everything PREMIS expects, lifted from the file itself on upload — no manual entry, no spreadsheet round-trip.
- Image
- Width · Height · DPI · Color · EXIF
- Audio
- Duration · Codec · Bitrate · Sample · Channels
- Video
- Duration · Codec · Bitrate · Frame rate · Resolution
- Page count · Word count · Fonts · Language
- Data
- Row count · Column count · Column names
- All
- Checksum + algorithm for fixity
Upload a thousand files. Walk away.
Drag-and-drop a folder. Watch a queue with per-file progress, speed, and status. Pause, resume, retry, or remove individual files. Batch-clear completed and cancel-all when you need to.
- Status tracking: Queued · Uploading · Processing · Completed · Failed · Paused
- Per-file Pause / Resume / Retry / Remove
- Batch operations: Clear Completed · Cancel All
- Overall progress with active / completed / error counts
- Target an existing item or auto-create a new one
Re-scan, re-render, never re-catalogue.
Send the photograph back through the scanner at higher resolution? Re-export the oral history with cleaner audio? Drop the new version straight on top of the old one. Your title, your reviewed transcript, the published snapshot, the item it's attached to — all of it stays put. A small version tag tells future readers the file was refreshed.
- The new file inherits the old file's catalogue record
- Reviewed metadata, item links, and published snapshots survive the swap
- A version tag (v2, v3 …) appears in the file header so the change is visible
- Crops and annotations reset — they belong to the pixels you just replaced
See how your files relate to each other.
Switch the Media Library from grid to network and the files arrange themselves into clusters — the same survey, the same photographer, the same recurring subject. Drag a node to reposition it. Click one to make it the centre of the world and watch the rest re-orient around it. Hit Find similar on any file to open a focused star-graph of its closest matches.
- Three view modes on the same library: Grid · Table · Network
- Files cluster automatically by shared subjects, tags, and AI signals
- Pick any file to recentre the graph — its closest matches pull in
- Find similar opens a star-graph centred on the source file
Spring-clean without holding your breath.
One grid of every file in your archive. Find the test uploads, the duplicate scans, the abandoned drafts — select a hundred at a time and clear them. The system quietly protects anything that's already attached to a catalogue record, so a busy afternoon doesn't end with an apology email to your director.
- Filter by file type, processing status, or creator
- Select hundreds of files at once and act on them in batches
- Files attached to items are protected from accidental cleanup
- A confirmation step on every destructive action — no surprises
Throw a real file at it.
Upload an oral history, a manuscript, a TIFF, an EML — see the extractions land in minutes.