About atavola

atavola turns "what should we eat tonight?" into a 60-second swipe. Free, non-commercial, hackathon-born.

What it is

atavola is a small web app that helps groups of 2-20 people decide what to cook or order. A host picks a vibe — a phrase like "Cozy Movie Night" or "Taco Tuesday" — and an AI generates four meal candidates with images, descriptions, and metric ingredient lists scaled to the headcount. The host shares a public voting link; everyone swipes right to like or left to dislike. The results screen re-orders cards in real time as votes come in.

There is no terminal "winner" state. The link stays open, the ranking keeps updating, and the group converges naturally.

Who built it

atavola was built at the IBM Bobathon in Zürich, 2026, by Alessandro, Marco, and Philipp. It started as a hackathon project and was kept alive afterwards as a free, open-source study project. There are no plans to monetise.

How it's built

atavola runs entirely on free tiers. The total infrastructure cost is $0/month at hackathon scale. The stack:

  • Frontend: React 19, TypeScript, Vite, Tailwind v4, Framer Motion, dnd-kit.
  • Backend: Vercel serverless functions (TypeScript).
  • Auth + DB + Realtime + Storage: Supabase.
  • LLM: OpenRouter, free models with automatic rotation.
  • Image generation: rotation across Pollinations.ai, Hugging Face FLUX.1-schnell, and Cloudflare Workers AI.
  • Email: Resend (magic links via Supabase SMTP, weekly digest via the Resend HTTP API).
  • Hosting: a single Vercel project at the same origin.

Why it exists

Most "what should we eat?" decisions are bottlenecked by either a single person deciding for everyone or an endless thread that converges on nothing. atavola replaces both with a 60-second swipe game. The app's job is to remove friction from a tiny daily decision, not to be a recipe app, a calorie tracker, or a delivery service.

Open source

atavola is MIT-licensed. The code, the schema, the deployment guide, and the threat model are all in the public repository. Pull requests welcome; security reports go to the address below.

Contact

[email protected] — the no-reply prefix is intentional, but write anyway if you have something to say.