recoengine

Batteries-included entry point for the recoengine recommendation system — an algorithmic, explainable, domain-agnostic recommendation engine for TypeScript. Runs on Node, Bun, Deno and in the browser.

This package re-exports @recoengine/core plus the full standard plugin set, so a single install gives you the whole thing.

npm i recoengine

What you get

| From | Exports | | --- | --- | | @recoengine/core | createEngine, the pipeline, ports, scoring maths, explainability | | @recoengine/strategies | nine scoring strategies (popularity, history, affinity, recency, similarity, …) | | @recoengine/modifiers | fatigue, novelty, boosts/penalties | | @recoengine/diversity | MMR, attribute quotas, blending, similarity | | @recoengine/features | reusable history-based feature extractors and transforms |

Everything imports from one place:

import {
  createEngine,
  popularityStrategy,
  historyStrategy,
  fatigueModifier,
  mmrDiversifier,
} from 'recoengine'

const engine = createEngine<Track>()
  .use(myProvider)           // your data (the only domain-specific part)
  .use(myFeatureExtractor)   // your payload → numbers
  .use(popularityStrategy())
  .use(historyStrategy())
  .use(fatigueModifier({ threshold: 2, halfLife: 3 }))
  .configure({ limits: { maxCandidates: 5_000, maxLimit: 100, timeoutMs: 200 } })
  .build()

const { recommendations, diagnostics } = await engine.recommend({ user, history, limit: 10 })

See @recoengine/core for the complete, tested end-to-end example.

Prefer a smaller tree?

Depend on @recoengine/core (zero dependencies) and add only the plugin packages you use. recoengine is the convenience meta-package; the core plus hand-picked plugins gives you the same engine with a smaller surface.

Links

• Repository & full docs: https://github.com/waleron07/recommendationEngine

MIT