Atlas

Your personal intelligence feed — everything the web knows, ranked for you

Atlas is a Google Discover replacement for power users — a mobile-first PWA that aggregates every corner of the web (RSS, Reddit, Hacker News, GitHub, YouTube, AI news, finance, cybersecurity, privacy) and surfaces the most relevant content using a recommendation engine that learns from every interaction.

Key Highlights

• Works without login — full feed accessible as guest
• 13+ AI providers — Puter.js (zero-config), Gemini, Groq, OpenRouter, Cerebras, NVIDIA, Mistral, Cohere, HuggingFace, GitHub Models, Cloudflare AI
• 45+ RSS feeds pre-seeded across 10 categories
• Web discovery — DuckDuckGo search integration (no API key needed)
• Dark Observatory design — Playfair Display + Newsreader typography, deep indigo-black backgrounds
• Offline-first PWA — Serwist service worker, IndexedDB cache
• Knowledge management — bookmarks, highlights, notes, collections, spaced repetition (FSRS)
• Gamification — reading streaks, curiosity scores, knowledge graph visualization
• Full cloud sync — Supabase Postgres + pgvector for semantic search

Tech Stack

Quick Start

Prerequisites

• Node.js 22+
• pnpm 10+ (npm install -g pnpm)

Installation

git clone https://github.com/chirag127/atlas.oriz.in.git
cd atlas.oriz.in
pnpm install

Environment Variables

cp .env.example .env.local

Atlas works with zero configuration. The following features work without any API keys:

• Puter.js AI (client-side, uses user's Puter account)
• DuckDuckGo discovery (no key needed)
• Demo articles (when no database configured)

For full functionality, configure:

Development

pnpm dev

Open http://localhost:3000

Database Setup (Optional)

# Push schema to Supabase
pnpm db:push

# Seed default feeds
pnpm db:seed

# Open Drizzle Studio
pnpm db:studio

Features

Feed & Discovery

• 45+ pre-seeded RSS feeds across 10 categories
• Web discovery via DuckDuckGo (free, no key)
• Optional paid discovery: Serper, Tavily, Brave Search, Exa
• Personalized recommendation engine with interest learning
• Anti-clickbait scoring
• India content boost

AI Features

• Puter.js — zero-config, client-side, 400+ models
• 11 server-side providers with automatic fallback chain
• Per-task model selection (summarize, chat, tags, digest, embed)
• TL;DR, full summary, key insights, ELI5, article chat
• Live model discovery with "Refresh models" button
• AES-256-GCM encryption for stored API keys

Knowledge Management

• Bookmarks with FSRS spaced repetition
• Highlights with 4 color options
• Standalone and article-attached notes
• Collections (manual + AI-powered smart folders)
• Tags (user-defined + AI-generated)
• Knowledge graph visualization
• Export: Markdown, Obsidian vault, JSON backup

PWA & Offline

• Serwist service worker
• Offline article reading
• Background sync for pending actions
• Push notifications (VAPID)
• Add to Home Screen support

Gamification

• Reading streak tracking
• Curiosity score (topic breadth)
• Learning score (weighted activity)
• Knowledge graph connections
• Weekly recap email

Deploy to Vercel

1. Push to GitHub
2. Import in vercel.com
3. Add environment variables
4. Deploy

Vercel auto-detects Next.js and deploys with zero config.

Deploy to Cloudflare Workers

pnpm add -D @opennextjs/cloudflare
# Configure open-next.config.ts
pnpm build
npx wrangler deploy

Architecture

atlas.oriz.in/
├── src/
│   ├── app/           # Next.js App Router pages + API routes
│   ├── components/    # React components (ui, feed, reader, ai, etc.)
│   ├── lib/           # Core libraries (db, ai, feeds, discovery, etc.)
│   ├── hooks/         # React hooks
│   ├── stores/        # Zustand stores
│   └── middleware.ts  # Auth guard
├── public/            # Static assets, fonts, service worker
└── tests/             # Unit, integration, e2e tests

License

MIT — Chirag Singhal