@ontosdk/core

Shared HTML → Markdown cleaning + AIO (Agent Input Optimization) scoring engine.

Powers the Onto Read API (api.buildonto.dev), the @ontosdk/next SDK, and the public AIO scanner at buildonto.dev. One cleaning engine, one scoring model, one bot registry — used everywhere Onto reads or scores web pages.

Install

npm install @ontosdk/core

Use

import { extractContent, calculateAioScore, getHallucinationRisk } from '@ontosdk/core';

const html = await fetch('https://example.com').then(r => r.text());

const extraction = extractContent(html, 'https://example.com');
//   → { markdown, title, headings, ... }

const scoring = calculateAioScore({
  html,
  markdownPayload: extraction.markdown,
  targetUrl: 'https://example.com',
  robotsAllowed: true,
  isUsingOntoSdk: false,
});
//   → { score, grade, insights, penalties, benefits, recommendations, bot_preview }

const risk = getHallucinationRisk(scoring.score, scoring.insights);
//   → 'low' | 'medium' | 'high'

Entry points

| Import | Returns | |---|---| | @ontosdk/core | Everything from clean + score re-exported | | @ontosdk/core/clean | extractContent, ExtractionResult | | @ontosdk/core/score | calculateAioScore, getHallucinationRisk, scoreToGrade, AI_BOTS, matchBot, types |

What it does

Clean. Strips React/Tailwind/framework noise from HTML. Normalizes heading hierarchy. Extracts page title. Produces compact, semantic Markdown an AI agent can actually use — typically 10–100× smaller than the source HTML.

Score. Returns a subtractive-penalty AIO score (0–100) with structured insights — what's helping the page's AI readability, what's hurting it, and ranked recommendations. The same score the Onto Read API and public scanner return.

Match bots. Ships a registry of known AI crawlers (GPTBot, ClaudeBot, PerplexityBot, etc.) and a matchBot() helper for user-agent detection at the edge.

License

MIT — see LICENSE.