ApartureBringing the arXiv into focus
One daily briefing replaces raw arXiv skimming, with every citation verified against the source papers.
One daily briefing replaces raw arXiv skimming, with every citation verified against the source papers.
Aparture reads the daily arXiv listings so you can stop skimming them manually. You describe your research in plain English and select the arXiv categories you follow. Each run scores the new preprints in those categories against your description, reads the highest-scoring manuscripts in full, and writes a cohesive briefing that groups the papers into themes and explains how each one connects to your work. Every citation in the briefing is verified against the input papers before the page renders.
Aparture is designed for continuous daily use over weeks and months, and you can run it either as a local web application (npm run dev, using your own provider API keys in .env.local with interactive review gates) or as a headless Agent Skill (SKILL.md paired with node bin/aparture.mjs, running inside Claude Code, Antigravity, or OpenAI Codex on your existing agent subscription). As you read each briefing, you star papers that were valuable, dismiss papers that missed the mark, and leave comments where you have specific reactions. When you request profile suggestions, Aparture turns those accumulated signals into concrete proposed edits to your research profile that you can accept or reject individually.
You describe your research in natural prose the way you would explain it to a colleague. Every stage of the pipeline reads that description, from the initial abstract filter through the final synthesis.
Each run produces a single editorial document that synthesizes across the day's papers, opening with an executive summary and highlighting which papers to read first.
Star, dismiss, or comment on papers as you read. The Suggest Improvements workflow analyzes your feedback and proposes targeted edits to your profile, citing the specific events that motivated each change so you can accept or reject it.
Run the interactive web application in your browser (mixing Anthropic, OpenAI, and Google models per stage with keys in .env.local), or invoke the built-in SKILL.md workflow inside Claude Code, Antigravity, or Codex without separate API keys.
Install Node.js, clone the repository, and optionally set up Playwright in about 5 minutes.
STEP 02Add a provider key to .env.local for the browser app, or skip directly to the Agent Skill.
Run the free Dry Run and the Minimal API Test (~$0.20–$1 on paid tiers).
Follow a narrated 10-minute walkthrough of your first live briefing in the local web interface.
AGENT SKILL (SKILL.MD)Execute the pipeline headlessly via SKILL.md and node bin/aparture.mjs in Claude Code, Antigravity, or Codex.
If you already have Aparture running, the Guide covers daily workflows from running and reading briefings to recording feedback and refining your profile, Running as an agent skill documents the SKILL.md workflow and headless CLI runner, Under the Hood explains the pipeline architecture and briefing checks, and Reference documents environment variables, prompt templates, and troubleshooting steps.