For studios, archives and creators

Every frame you own,
finally findable.

MetaLoom watches, listens to and reads your media — then hands it back as an archive you can actually search. On your hardware.

See what it does Run the demo
Scroll

The problem

You didn't lose the shot.
You lost the filename.

The interview is in there. So is the drone pass, the b-roll, the sponsor logo you now need cleared. Somewhere, on a drive, behind a name a camera made up.

Everything an ordinary archive knows about this file.

What MetaLoom does

Drop in a file.
Get back what is inside it.

Every asset runs through a pipeline you compose. Each step adds something the next one can use — and everything lands in one searchable index.

How MetaLoom processes an asset A media file is dispatched to four kinds of processing worker — speech, faces, scenes and fingerprints — whose results are collected into one searchable index. one file listens · speech to text looks · faces & scenes reads · text & documents matches · fingerprints one index
  1. 01

    Ingest

    Point at a folder, a drive or an upload. Files stay where they are if you want them to.

  2. 02

    Understand

    Speech, faces, scenes, text, quality and fingerprints — extracted by workers you scale yourself.

  3. 03

    Search

    Full text, structured filters and vector similarity over everything that came out.

Speech

Search what was said.

Spoken audio becomes timecoded text. Jump straight to the sentence instead of scrubbing for it.

  • Transcripts
  • Timecoded segments
  • Runs locally
Speech-to-text node

People

Know who is on screen.

Faces are detected, described as vectors and matched across your whole library — so "every clip with her in it" becomes a query.

  • Face detection
  • Descriptions
  • Similarity search
Face detection node

Picture

Every scene, described.

Long takes are cut into scenes, and a vision model writes down what each one shows. Contact sheets come free.

  • Scene detection
  • Captions
  • Thumbnails
Scene detection node

Reach

Ship it in another language.

Extract, translate, and — when you want it — speak it back in a synthetic voice. One pipeline, end to end.

  • Translation
  • Subtitles
  • Dubbing
Translation playbook

Order

Never ingest the same shot twice.

Content hashes catch the exact duplicates. Perceptual fingerprints catch the re-encodes, the crops and the rescales.

  • Content hashes
  • Perceptual fingerprints
  • Dedup
Fingerprint node

Ask

Talk to your archive.

A built-in agent answers questions about your material in plain language, with the assets it used attached to the answer.

  • Chat agent
  • Skills
  • Memory
Chat & AI agent
19
built-in processing nodes
3
ways in: web, desktop, terminal
1
command to start the demo
0
frames sent to a cloud you don't own

Your material, your machines

Nothing has to leave the building.

MetaLoom runs on your own hardware — a single box under a desk or a Kubernetes cluster in your rack. The platform is Apache 2.0. The models are the ones you choose to install.

On-premise by default

No upload step, no per-minute processing bill, no third party holding your rushes.

Open source

Apache 2.0. Read it, fork it, run it after we stop answering emails.

Legal & licensing

Models you pick

Speech, vision and language models are configuration. Swap them for whatever your policy allows.

Model licenses

Who it is for

Built for people with more footage than time.

Post production

Find the take, the frame and the face across every project on the shared storage — without an assistant logging it by hand.

Broadcast & archive

Turn decades of tape transfers into something searchable, deduplicated and described.

Creators & agencies

Reuse what you already shot. Cut faster because the archive answers questions instead of listing folders.

Try it

Point it at a folder tonight.

The demo image runs the whole platform with seeded assets and pipelines. One command, no account, nothing to sign.

docker run -p 8092:8092 metaloom/loom-demo:latest

Then open the UI on port 8092 and log in as admin.

Getting started Read the docs Join the Discord