<<<<<<< HEAD A collaborative AI system where GPT-oss, Gemma, and Flux work together to provide the best possible answers.
A collaborative AI system where GPT-OSS-120B and Gemma 3-4B-IT work together with real-time streaming to provide the best possible answers.
- ⚡ Real-time Streaming - See responses as they're generated (like ChatGPT)
- 🤖 Individual Agent Views - Expand to see each agent's contribution
- 🛡️ Smart Error Handling - Retry logic + graceful degradation
- 🔄 Response Regeneration - Regenerate any response instantly
- 🎨 Enhanced UI/UX - Toast notifications, loading skeletons, keyboard shortcuts
- ⌨️ Keyboard Shortcuts - Cmd/Ctrl+K for new chat
- 📋 Code Copy Button - One-click copy for code blocks
- 🎭 Agent Transparency - See GPT-OSS → Gemma → Final synthesis
e65d0a6 (feat: Enhanced Light Mode UI, multi-agent persistence, and terminal code blocks)
- 🤖 Real AI Collaboration - Two premium AI models discuss and refine answers
- 💬 Sequential Discussion - Each AI builds on previous responses <<<<<<< HEAD
- 🎨 Image Generation - Fluxintegration for creating images
- 💾 Conversation History - Remember all your chats =======
- 🎨 Image Generation - Flux.2 integration for creating images
- 💾 Conversation History - Remember all your chats (coming soon)
e65d0a6 (feat: Enhanced Light Mode UI, multi-agent persistence, and terminal code blocks)
- ⚡ Best Quality - Synthesized answers from multiple perspectives
<<<<<<< HEAD
- GPT-oss provides initial comprehensive analysis
- Gemma reviews, finds gaps, and adds improvements and synthesizes both into the best final answer
- You get one polished response from true AI collaboration =======
- GPT-OSS-120B provides initial comprehensive analysis
- Gemma 3-4B-IT reviews, finds gaps, and adds improvements
- You get one polished response from true AI collaboration
- Streaming - Watch the magic happen in real-time!
e65d0a6 (feat: Enhanced Light Mode UI, multi-agent persistence, and terminal code blocks)
npm installnpm run setupOr manually:
npm install
npx prisma generate
npx prisma migrate dev --name initCreate/update .env.local:
# NVIDIA NIM API Keys (get from https://build.nvidia.com)
GPT_OSS_API_KEY=your_key_here
GEMMA_API_KEY=your_key_here
FLUX_API_KEY=your_key_here
# Database (optional)
DATABASE_URL="file:./dev.db"npm run dev<<<<<<< HEAD
- User submits a query
- Agent 1 (GPT-oss) provides initial analysis
- Agent 2 (Gemma) reviews and adds perspective and synthesizes the best final answer =======
Cmd/Ctrl + K- Start new conversationShift + Enter- New line in messageEnter- Send message
- Click "Show Agent Responses" to see individual agent contributions
- Click "Regenerate" to get a new response
- Hover over code blocks to see the Copy button
- Watch real-time streaming as agents think and respond
e65d0a6 (feat: Enhanced Light Mode UI, multi-agent persistence, and terminal code blocks)
User Query
↓
<<<<<<< HEAD
Agent 1: GPT-oss (Analysis)
↓
Agent 2: Gemma (Critical Review and Synthesis)
↓
Final Answer
=======
┌─────────────────────────┐
│ Streaming Pipeline │
├─────────────────────────┤
│ 1. GPT-OSS (Analysis) │ ← Real-time streaming
│ 2. Gemma (Synthesis) │ ← Real-time streaming
│ 3. Final Response │
└─────────────────────────┘
↓
Enhanced UI with Toast + Loading States
>>>>>>> e65d0a6 (feat: Enhanced Light Mode UI, multi-agent persistence, and terminal code blocks)
- Next.js 14 (App Router) with TypeScript
- NVIDIA NIM API (AI model access)
- Prisma + SQLite (Database - optional)
- React with hooks and streaming
- Tailwind CSS with custom design system
- Server-Sent Events (SSE) for streaming
Streaming endpoint with real-time updates
- Uses Server-Sent Events (SSE)
- Returns agent responses as they're generated
- Better UX with instant feedback
Non-streaming endpoint (still works)
- Returns complete response only
- Simpler but slower perceived performance
- SETUP_GUIDE.md - Detailed setup and feature guide
- ✅ Streaming responses with SSE
- ✅ Retry logic with exponential backoff (3 attempts)
- ✅ Graceful degradation (fallback to Agent 1 if Agent 2 fails)
- ✅ Better error messages and handling
- ✅ Individual agent response tracking
- ✅ Real-time streaming UI
- ✅ Toast notifications (success/error/info)
- ✅ Loading skeletons and smooth animations
- ✅ Expandable agent response cards
- ✅ Regenerate button on each response
- ✅ Code block copy buttons
- ✅ Keyboard shortcuts (Cmd/Ctrl+K)
- ✅ Professional design system
- 💾 Conversation persistence (database ready!)
- 🔍 Search conversations
- 📤 Export conversations (Markdown/PDF)
- 🔗 Share conversations (public links)
- 👤 User authentication
- 📊 Usage analytics
- Free with NVIDIA NIM (check their limits)
- No monthly subscription required
- Pay-per-use model
# Run development server
npm run dev
# Build for production
npm run build
# Start production server
npm start
# Open Prisma Studio (database GUI)
npm run prisma:studio
# Generate Prisma client
npm run prisma:generate
# Run database migrations
npm run prisma:migrate- Check browser console for errors
- Verify API keys in
.env.local - Try the legacy endpoint first:
/api/chat
- Run
npm run prisma:generate - Run
npm run prisma:migrate - Check
DATABASE_URLin.env.local
- Verify API keys are correct
- Check NVIDIA NIM API status
- Review rate limits
- Streaming: Perceived 3x faster response time
- Error Handling: 3 automatic retries with backoff
- Fallback: Always returns best available response
- Optimized: Production build with tree-shaking
Contributions welcome! Feel free to open an issue or submit a pull request.
MIT