sdk.tag()

Classify images with a custom prompt, optionally constraining to a list of options. Supports both cloud API and local in-browser processing via transformers.js.

Interactive Demo

Try it out ↓

Signature

sdk.tag(image: string, options: TagOptions): Promise<TagResponse>

Parameters

Parameter	Type	Required	Description
`image`	string	Yes	Image URL or base64 data URI
`options.prompt`	string	Yes	The classification question or context
`options.options`	string[]	No	Constrained list of valid tags to choose from
`options.multiple`	boolean	No	Allow multiple tags (default: `true`)
`options.mode`	string	No	`"api"` (default) or `"local"` (in-browser via transformers.js)

Returns

Promise<TagResponse>:

{
  tags: string[];  // selected tags
  raw: unknown;    // raw model response
}

Examples

import { CoreViz } from '@coreviz/sdk';

const sdk = new CoreViz({ apiKey: process.env.COREVIZ_API_KEY });

// Open-ended tagging
const { tags } = await sdk.tag('https://example.com/product.jpg', {
  prompt: 'What objects are visible in this image?',
});
console.log(tags); // ["red sneaker", "white background", "shoelace"]

// Single-choice classification
const { tags: [category] } = await sdk.tag('https://example.com/doc.jpg', {
  prompt: 'What type of document is this?',
  options: ['receipt', 'invoice', 'contract', 'other'],
  multiple: false,
});
console.log(category); // "receipt"

// Local processing (no API key needed, runs in browser)
const { tags: localTags } = await sdk.tag(imageBase64, {
  prompt: 'Describe the scene',
  mode: 'local',
});

Use Cases

Document classification — classify receipts, invoices, contracts
Content moderation — flag inappropriate content
Product categorization — auto-tag e-commerce products
Auto-organization — tag images on upload for searchability

Local mode

mode: 'local' runs entirely in the browser or Node.js using transformers.js — no API call is made. Results may differ from the cloud model. Not supported on React Native / Expo.

Raw API Endpoint

This method calls POST /api/ai/tag. See the Vision API Reference for raw HTTP details.

sdk.tag()

On this page