Reference retrieval for creative work — every result arrives license-tagged.

Search images, video, audio, and text as creative references across 18+ sources (The Met, Rijksmuseum, Openverse, Unsplash, Flickr, Internet Archive, Freesound, …) through one neutral API. Every result carries source + license + attribution + canonicalUrl, normalized per-result — so you can gate usage (commercial-product? ai-generation-input?) before an asset ever enters your pipeline.

import { createRefkit } from '@refkit/core'
import { openverse } from '@refkit/provider-openverse'
import { met } from '@refkit/provider-met'

const refkit = createRefkit({ providers: [openverse(), met()] })

// Fan out, RRF-merge, dedup — only return commercially-usable results:
const refs = await refkit.search({
  query: 'cyberpunk alley at night',
  modalities: ['image'],
  gateFor: 'commercial-product',
})

Multi-source search is a commodity. The hard part is rights: license normalization across wildly inconsistent source metadata, a strict-deny use-gate (allowed / allowed-with-attribution / denied / needs-review), and ready-to-paste attribution strings. refkit does that per-result, for visual and text references, in an embeddable BYOK SDK that agents can call directly (MCP server included).

Keyless providers work out of the box; API-key providers are BYOK.

pnpm add @refkit/core @refkit/provider-openverse @refkit/provider-met

📖 Full docs & examples: refkitjs/refkit · 📦 npm: @refkit/*

Apache-2.0. Contributions welcome — especially new providers: the provider testkit makes a new source a small, well-tested package.