Video archive intelligence

Search every spoken word in your archive — without uploading a frame

Archivist Sort transcribes your video locally with Whisper-family backends and makes every spoken word searchable across your whole archive. Nothing leaves your Mac, and if a clip matters enough, you can pull it out as an experimental FCP7 XML (xmeml) handoff for your editor.

Transcript search and local video search are part of the free read path — the download, scan, review, and full plan are free forever. Executing any resulting cleanup plan through Sort is the paid step.

The problem with video archives

A video archive is only as findable as the words inside it. If a clip matters, what you remember is what someone said — but the footage is stored as a filename and a timestamp, not as a sentence. "Where's the interview where they explained the release plan?" is a search you can't run on folders.

Archivist Sort closes that gap the same way it handles everything else: locally, deterministically, and without trusting you to trust it.

Transcribe locally, never upload

Transcription is entirely on your machine using Whisper-family backendswhisper-cli, openai-whisper, or mlx-whisper. The app finds Homebrew and pipx installs automatically, so you don't manage a path or a server. No cloud, no accounts, no upload — the source files and their transcripts stay on your Mac.

Transcription is optional: Sort works fully without it. If you only need the file-cleanup and archive-organizing core, nothing about this feature demands your footage gets touched.

Transcription quality follows your backend and your hardware — the app never invents a transcript, and it is honest about when a usable backend isn't found. Settings shows exactly which binary was found, where, and why — or the specific reason none is usable.

Search every spoken word

Once your footage is transcribed, search your transcripts on the Report screen and find the exact clip where something was said. Search across the whole archive, not file by file — every spoken word becomes a search term. Duplicate hits collapse to unique clips, so the same moment appearing in two source files doesn't clutter your results.

The workflow stays review-first: nothing about search moves, deletes, or alters your source files. It's read-only on your footage until you decide to export.

Export matching clips as an experimental FCP7 XML bin

When a search finds the footage you need, export the matching source files as an experimental FCP7 XML (xmeml) bin — a handoff starting point for Resolve or Premiere, not a finished timeline.

This is a handoff starting point for your editor, not a finished edit. Plan to assemble and refine in your NLE of choice.

Why the receipts matter

Every search and every export runs under the same fail-closed rules as the rest of Archivist Sort: nothing executes without your explicit acknowledgement, nothing deletes by default, and every action is journaled with checksums. Video search doesn't get a looser standard than the cleanup engine — it gets the same one.

Read how the feature actually works end to end: Guide: find the clip where someone said something.