Margin

Minimal dark reading ritual for arXiv-powered papers.

What it does

• recent paper queue from arXiv metadata
• search across arXiv papers
• clean HTML-first reader when arXiv HTML exists
• abstract fallback when HTML is missing
• AI brief via OpenAI when OPENAI_API_KEY is set
• local fallback brief when no key is present
• saved text highlights in localStorage
• browser text-to-speech for summary or full paper

Stack

• Next.js 16 app router
• direct arXiv Atom API + HTML paper fetch
• fast-xml-parser for Atom
• cheerio for HTML cleanup/extraction

Run

bun install
bun dev

Open http://localhost:3000.

Env

Copy .env.example to .env.local.

OPENAI_API_KEY=
OPENAI_MODEL=gpt-4.1-mini

Without OPENAI_API_KEY, Margin still works and falls back to abstract-based summaries.

Notes

• arXiv HTML is not available for every paper yet. Margin falls back gracefully.
• Speech uses the browser speechSynthesis API, so voice quality depends on the device/browser.
• Product acknowledgment included per arXiv API guidance: Thank you to arXiv for use of its open access interoperability.