`.vtag`: an open layer for AI metadata

Position .vtag as a portable, open AI metadata format that travels with the file—not buried in a single vendor’s silo. VisionaryAI Suite is one way to create and update these sidecars; the file on disk is the contract with the rest of your stack.

What is `.vtag`?

A .vtag file is a sidecar document stored next to a media file (same basename, different extension). It holds structured results of AI and parsing steps—typically JSON—so any tool that can read a text file can participate: backup, Git, find/grep, custom scripts, and catalog software with indexing rules.

Why sidecar metadata matters

Portability: copy a folder, and the intelligence travels with the media.
Provenance: you can record which model produced which field, for audit and reproducibility as models change.
Interoperability: XMP, EXIF, and database-only solutions each have a role; a sidecar can remain the neutral interchange for rich AI output that outgrows a few embedded fields.

Example (illustrative JSON)

The real schema is defined by the product and may evolve. Treat this as a readability example, not a normative spec:

{
  "vtag_version": "1.0",
  "media": "interview_0445.mp4",
  "summary": "Outdoor interview, traffic noise, two speakers",
  "tags": ["interview", "outdoor", "car_passes"],
  "objects": [
    { "label": "person", "t": 12.4, "confidence": 0.89 }
  ],
  "caption": { "text": "Reporter with microphone, building facade behind", "confidence": 0.78 },
  "transcript": { "text": "…we filed the request under the public record law…" },
  "speakers": [ { "id": "spk1", "label": "Reporter" } ],
  "timeline": [ { "t0": 0, "t1": 3.1, "event": "siren" } ],
  "confidence": { "overall": 0.81 },
  "engine": { "name": "VisionaryAI Suite", "build": "0.0.0" }
}

What can be stored in `.vtag`

Summary

High-level overviews of the asset for quick triage and cards in search UIs.

Objects

Detections and regions that ground search in the visual or temporal structure of the file.

Captions

Natural language descriptions, with confidence where the pipeline provides it.

Transcripts

Text from speech for keyword and phrase access across long recordings.

Speakers

Roles or diarised identities when the workflow supports it.

Timeline events

Points or ranges for “what happens when” in video or audio.

Confidence scores

So humans and automations can down-rank low-trust inferences without deleting them.

Engine metadata

Model names, versions, and pipeline identifiers for your technical records.

Open metadata instead of a locked-in database

A database inside a single product can be excellent for that product and still be the wrong long-term archive contract if the vendor changes terms or the project ends. A sidecar you own is boring technology—in the best way. You can re-index it tomorrow in another system without a migration project that re-runs all inference on day one (unless you choose to, for a new model generation).

How `.vtag` can work with XMP and catalogs

Many workflows keep XMP in images and database rows in a DAM, while video and audio often carry less in embedded metadata. A .vtag can bridge that gap: one rich JSON for modern AI output, while you still selectively map into XMP or database columns where your tool chain expects them. The same sidecar can feed NeoFinder indexing in documented setups—without replacing how each tool stores its own cache.

Future of `.vtag`

As an open AI metadata pattern, the format can grow with the ecosystem: new fields, shared schemas between tools, and community tooling around validation and display. The direction is a portable, inspectable, versionable record of what machines inferred about your media—without pretending that a single file replaces legal records or human cataloguing policy.

Product overview Read the FAQ

.vtag: an open layer for AI metadata

What is .vtag?