Multi-Domain Research Intelligence with Autonomous Browser Automation
Where knowledge converges through parallel autonomous intelligence
Purple Prism is the first AI research engine with autonomous parallel browsing. Powered by Lightpanda headless browser + Claude Sonnet, it deploys multiple lightweight browsers simultaneously to research your query from every angle.
Key Innovation: While Perplexity searches and summarizes, Purple Prism autonomously browses 5+ websites in parallel, extracts information, cross-references claims, and synthesizes comprehensive reports.
- Parallel browsing: 5 Lightpanda instances scraping simultaneously
- 11x faster than traditional browser automation
- 9x less memory (24MB per browser vs 207MB for Chrome)
- Server-side execution: No user browser required
- Real-time synthesis: Claude combines findings into coherent reports
- Animated purple prism logo with light refraction
- Dynamic video background (120s prism loop)
- Liquid glass dark mode overlay
- A/B video toggle (V1: dynamic, V2: subtle)
- Standard Mode: Fast Claude synthesis from knowledge base
- Deep Research Mode: Autonomous multi-source web research
- Bring 1-5 API keys: Claude, GPT-4, Gemini, or all of them
- Automatic load balancing: Distribute synthesis across providers
- Redundancy: Auto-fallback if one API fails
- Advanced: Parallel multi-model analysis (coming soon)
- See MULTI-PROVIDER-AI.md for details
- Quick summary
- Notion-ready document
- Presentation slides
- Deep research report
- Podcast-ready audio script
- 🏃 Quick (5-min scan)
- 🚶 Standard (15-min dive)
- 🏊 Deep (30+ min immersion)
User Query
↓
Purple Prism API
↓
Lightpanda (5 parallel instances)
↓
[Site 1] [Site 2] [Site 3] [Site 4] [Site 5]
↓
Claude Sonnet 4.5 (synthesis)
↓
Comprehensive Report (streamed)
Frontend:
- Next.js 14 (App Router)
- React 18
- Tailwind CSS
- TypeScript
AI/Automation:
- Claude Sonnet 4.5 (Anthropic API)
- Lightpanda Browser (11x faster, 9x less memory than Chrome)
- Custom TypeScript wrapper for browser automation
Optional Integrations:
- Notion API (auto-save research)
- Vercel Analytics
- Node.js 18+
- Anthropic API key (get one here)
- (Optional) OpenAI API key for GPT-4 support
- (Optional) Google API key for Gemini support
- (Optional) Notion API key for auto-save
# 1. Clone the repo
git clone https://github.com/your-username/purple-prism.git
cd purple-prism
# 2. Install dependencies
npm install
# 3. (Optional) Install additional AI providers
npm install openai @google/generative-ai
# 4. Download Lightpanda binary
# For Mac (M1/M2/M3):
curl -L -o bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-aarch64-macos
chmod +x bin/lightpanda
# For Linux:
curl -L -o bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-x86_64-linux
chmod +x bin/lightpanda
# 5. Set up environment variables
cp .env.example .env
# Edit .env and add your API key(s)
# 6. Run development server
npm run devVisit http://localhost:3000
- Enter your query
- Select output format (Quick, Notion, Presentation, etc.)
- Choose research depth (Quick 🏃, Standard 🚶, Deep 🏊)
- Click "Refract Knowledge"
- Toggle "🌐 Deep Research ON" in header
- Enter complex query (e.g., "Latest quantum computing breakthroughs 2026")
- Watch real-time scraping:
🌐 Browsing: techcrunch.com 📊 Found: IBM announces 1000-qubit processor 🌐 Browsing: arxiv.org 📊 Found: New error correction breakthrough ... - Receive comprehensive synthesized report
# Required (at least one)
ANTHROPIC_API_KEY=sk-ant-xxx
# Optional (add for multi-provider support)
OPENAI_API_KEY=sk-xxx
GOOGLE_API_KEY=xxx
# Optional (for Notion integration)
NOTION_API_KEY=secret_xxx
NOTION_DATABASE_ID=xxxMulti-Provider Benefits:
- 1 provider: Basic Deep Research works
- 2-3 providers: Auto-fallback + load balancing
- 4-5 providers: Parallel multi-model synthesis
See MULTI-PROVIDER-AI.md for complete guide.
Edit lib/lightpanda.ts to customize:
// Timeout per scrape
{ timeout: 10000 } // 10 seconds
# Max concurrent scrapes
const limit = pLimit(5); // 5 parallel browsersTest: Research query requiring 5 source sites
| Mode | Time | Memory | Sources |
|---|---|---|---|
| Standard (Claude only) | ~3s | 50MB | 0 (knowledge base) |
| Deep Research (Lightpanda) | ~2-3s | 120MB | 5 (parallel web scraping) |
| Traditional (Puppeteer) | ~15s | 1GB+ | 5 (sequential) |
Result: Deep Research Mode is 5x faster than traditional browser automation with 8x less memory.
Toggle between two prism videos:
- V1 (default): 120-second dynamic prism loop
- V2: 61-second subtle geometric shifts
Button in UI header switches between them.
Purple Prism uses a purple/indigo gradient palette:
- Primary:
#8B5CF6(purple-500) - Dark:
#5B21B6(purple-900) - Light:
#C4B5FD(purple-300) - Accent:
#6366F1(indigo-500)
Edit app/globals.css and tailwind.config.js to modify.
# Build for production
npm run build
# Deploy
netlify deploy --prodImportant: Ensure bin/lightpanda is included in deployment.
Update netlify.toml:
[functions]
node_bundler = "esbuild"
included_files = ["bin/lightpanda"]vercel --proddocker build -t purple-prism .
docker run -p 3000:3000 -e ANTHROPIC_API_KEY=xxx purple-prism# Run benchmark
npx tsx test/lightpanda-benchmark.tsExpected output:
🚀 Benchmarking Lightpanda parallel scraping...
📊 Results:
Total time: 2847ms
Average per site: 569ms
Successful: 5/5
✅ Benchmark complete!
npm run dev
# Visit http://localhost:3000
# Toggle Deep Research ON
# Try query: "Latest RISC-V AI developments"- LIGHTPANDA-DEEP-DIVE.md: Comprehensive Lightpanda analysis
- LIGHTPANDA-INTEGRATION.md: Integration guide
- DEEP-RESEARCH-MODE.md: Deep Research feature docs
- VIDEO-COMPARISON.md: Background video details
This is currently a personal project, but suggestions are welcome!
- Fork the repo
- Create feature branch (
git checkout -b feature/amazing-feature) - Commit changes (
git commit -m 'Add amazing feature') - Push to branch (
git push origin feature/amazing-feature) - Open Pull Request
MIT License - feel free to fork and adapt.
Built by: Derayc
AI Partner: Claude Sonnet 4.5
Browser Engine: Lightpanda (22.2k ⭐)
Inspired by: Pink Floyd's Dark Side of the Moon (prism concept)
- ✅ Deep Research Mode with Lightpanda
- ✅ Parallel autonomous browsing
- ✅ Animated UI with video backgrounds
- ✅ Multiple output formats
- Interactive research steering ("Focus more on technical specs")
- Better source discovery (integrate web search API)
- Citation management
- Export to PDF/DOCX
- Multi-agent research (3 Claude agents researching different angles)
- Knowledge graph (entity relationships across queries)
- Voice input for queries
- Mobile app (React Native)
Q: Why Lightpanda instead of Puppeteer/Playwright?
A: Lightpanda is 11x faster and uses 9x less memory. For parallel research (5+ browsers), this means Purple Prism can run on cheap servers instead of requiring expensive compute.
Q: Does Deep Research Mode work without Lightpanda?
A: No, it requires the Lightpanda binary. But you can still use Standard Mode (Claude knowledge base) without it.
Q: Can I add my own research sources?
A: Yes! Edit lib/lightpanda.ts → findResearchSources() function.
Q: How much does it cost to run?
A: Only Anthropic API usage (~$0.01-0.10 per query depending on depth). Lightpanda is free and open-source.
Q: Is my data secure?
A: Yes. All research happens server-side. Your queries and results are not stored (unless you enable Notion integration).
- Lightpanda compatibility: Some sites with complex JavaScript may not scrape perfectly (fallback to standard mode)
- Rate limiting: Some sites may block rapid requests (add delays if needed)
- Memory limits: On free Netlify, limit concurrent browsers to 3 (vs 5)
See Issues for full list.
Purple Prism - Where knowledge converges 🔮✨