Plug AI into the upload pipeline — block, tag, transcribe

CoreUpload exposes four opt-in AI hooks that run in the pre-upload pipeline, symmetric to virusScan. Your hook typically calls a server endpoint fronting AWS Rekognition, Google Vision, Azure AI, or a local model — the component stays cloud-agnostic.

What you should see: auto-upload is off, so files sit in the queue after the AI pass until you press “Scan & upload”. The analysis result appears against each file before any bytes are sent.

contentModeration — blocking NSFW/safety gate; rejected files never upload.
ocr — extract text; stored on task.metadata.meta.ocrText.
autoTag — label objects/scenes; stored on task.metadata.meta.tags.
generateAltText — accessibility captions; stored on task.metadata.meta.altText.

This demo uses mock hooks (client-side heuristics) so it runs with no cloud credentials. Files whose name contains "unsafe" are blocked by the mock moderator. Real deployments call your endpoint from inside the hook.

Drag & drop files here, or paste from clipboard

AI events appear here...

Wiring (server-backed in production)

const up = CoreUpload.create('#uploader', {
 metaFields: [{ id: 'tags' }, { id: 'altText' }],

 contentModeration: async (file) => {
 const r = await fetch('/ai/moderate', { method: 'POST', body: file });
 const { safe, reason } = await r.json();
 return safe ? true : { allow: false, reason };
 },
 ocr:             (file) => fetch('/ai/ocr', { method:'POST', body:file }).then(r => r.json()),
 autoTag:         (file) => fetch('/ai/tag', { method:'POST', body:file }).then(r => r.json()),
 generateAltText: (file) => fetch('/ai/caption', { method:'POST', body:file }).then(r => r.text())
});

Enricher results land on task.metadata.meta.*; because this demo declares matching metaFields (tags, altText), they populate the inline inputs and ship to the server automatically via the metadata field / X-Upload-Meta header.