Skip to content

Latest commit

 

History

History

README.md

Built with Hyperbrowser

HyperGraph

Turn any technical topic into a traversable skill graph your agent can navigate. This is the difference between an agent that follows instructions and one that understands a domain.

HyperGraph scrapes real sources with Hyperbrowser, then uses Claude to generate a network of interconnected markdown nodes — each one a complete thought, linked with wikilinks the agent can follow.

Search → Hyperbrowser scrape → Claude graph generation → Interactive force graph

What it generates

  • MOC — Map of Content; the entry point your agent reads first
  • Concepts — foundational ideas, theories, and frameworks
  • Patterns — reusable techniques and approaches
  • Gotchas — failure modes and counterintuitive findings

Every node has a YAML description the agent can scan without reading the full file. Wikilinks are woven into prose so the agent knows why to follow them.

Get an API key

Get your Hyperbrowser API key at https://hyperbrowser.ai

Quick start

git clone https://github.com/yourusername/hypergraph
cd hypergraph
npm install
cp .env.example .env.local   # fill in your keys
npm run dev

Open http://localhost:3000 and enter any technical topic.

Environment variables

HYPERBROWSER_API_KEY=          # from hyperbrowser.ai
SERPER_API_KEY=                # from serper.dev — used to find source URLs
OPENAI_API_KEY=                # used for graph generation with GPT-4o
HYPERBROWSER_MAX_CONCURRENCY=1 # free plan: keep at 1. paid plans can increase this.

Free plan & concurrency

The free Hyperbrowser plan supports 1 concurrent browser session. By default the app scrapes URLs sequentially (one at a time), so it works out of the box on any plan.

If you hit a concurrency limit the app will show an amber warning banner with an Upgrade plan link rather than crashing silently. To unlock parallel scraping, upgrade at hyperbrowser.ai and set HYPERBROWSER_MAX_CONCURRENCY to match your plan's limit.

Stack

  • Next.js 16 — App Router, API routes
  • Hyperbrowser — scrapes source material with onlyMainContent for clean output
  • OpenAI GPT-4o — generates the skill graph JSON from scraped docs
  • Serper — Google search to find the best source URLs for any topic
  • react-force-graph-2d — force-directed graph canvas rendering
  • react-markdown — renders node content in the VS Code-style preview panel

Download

Every generated graph can be downloaded as a .zip containing ready-to-use markdown files. Drop the folder into .claude/skills/ or .cursor/skills/ and point your agent at moc.md as the entry point.


Follow @hyperbrowser for updates.