👻 GhostFeed — RSS, back from the dead

Kiroween 2026 · Resurrection category · Built with Kiro

🌐 Try Live Demo | 📖 Deployment Guide

Algorithmic feeds bury what readers actually want. GhostFeed brings back reader-owned subscription: subscribe to any RSS/Atom feed, see at a glance which sources are alive or dead, and get a short daily digest. No sign-up required — judges can try it in under a minute.

What it does

• Subscribe to any feed — paste an RSS/Atom URL; it is fetched and parsed server-side (no CORS issues).
• Read newest-first — items across your feeds, with source and date; click through to the original.
• Graveyard — a feed is marked Dead if it hasn't published in 90+ days or its last 3 fetches failed. Dead feeds show their last-alive date and last error, and recover automatically if they publish again.
• Replacements — dead sources get a curated replacement suggestion (never an invented URL), addable in one click.
• Daily digest — top items of the last 24h. Rule-based by default; set LLM_BASE_URL + LLM_API_KEY (OpenAI-compatible, e.g. new-api) for an AI summary, with automatic fallback to rule-based.

Run it locally

npm install
npm run dev    # http://localhost:3000
npm run build  # production check

Deploy it

Multiple deployment options available — see DEPLOYMENT.md for complete guides.

Since Vercel is not available, we provide 5 deployment options:

1. Docker + VPS (Recommended) — Full control, easy to manage
2. Netlify — Easiest, free tier, auto-deploy
3. Railway — Simple, $5 free credit, auto-deploy
4. Cloudflare Pages — Free, global CDN, unlimited bandwidth
5. PM2 + Nginx + VPS — Traditional setup, full control

Quick Start (Docker + VPS):

# On your VPS
git clone <your-repo>
cd ghostfeed
docker compose up -d

Quick Start (Netlify):

# Push to GitHub, then connect at netlify.com
git push origin main
# Follow Netlify dashboard to import project

The app runs without environment variables (rule-based digest). For AI-powered digest, set LLM_BASE_URL, LLM_API_KEY, and LLM_MODEL in your deployment platform.

AI Digest (Optional)

GhostFeed uses a provider interface for digest generation with two implementations:

1. Rule-based (default) — simple numbered list, no LLM needed
2. AI-powered — connects to any OpenAI-compatible endpoint (OpenAI, new-api, local models)

To enable AI summaries, create a .env.local file:

LLM_BASE_URL=https://your-new-api.example.com/v1
LLM_API_KEY=sk-...
LLM_MODEL=gpt-4o-mini  # optional, defaults to gpt-4o-mini

The system automatically falls back to rule-based digest if the LLM fails — the page never crashes.

How Kiro was used

• Spec-driven development: .kiro/specs/ghostfeed/ holds requirements.md, design.md, and tasks.md — the spec was written before the code.
• Steering docs: .kiro/steering/ (product / tech / structure) keeps every change aligned: no sign-up, server-side RSS only, never invent a feed URL.
• Agent hooks: .kiro/hooks/on-save-typecheck.json runs tsc --noEmit on save.
• Property-based testing (Week 2): parser invariants — any valid RSS/Atom sample parses without throwing.

Tech

Next.js 14 (App Router) · TypeScript · Tailwind CSS · rss-parser · localStorage (no database, no auth in the MVP)

License

MIT — see LICENSE.