A Claude-powered research assistant built with Next.js 16, TypeScript, and the Claude Agent SDK for discovering, analyzing, and synthesizing AI research papers.

This application provides an intelligent research interface powered by Claude Agent SDK, featuring:

• Exa Neural Search: Semantic search for discovering academic papers and research
• Real-time Progress Tracking: Live updates on research progress with tool call visualization
• Comprehensive Reports: AI-generated markdown reports with citations
• Modern Dashboard: Built with shadcn/ui and Tailwind CSS
• Agent Architecture: Orchestrator agent with custom tool integration

• Neural search powered by Exa for semantic understanding
• Content retrieval from academic papers and articles
• Similarity search to find related work
• Multi-source synthesis and analysis

• Live streaming of research progress via Server-Sent Events (SSE)
• Visual indicators for current research stage
• Detailed tool call timeline with collapsible details
• Statistics dashboard (searches, fetches, total steps)

• Comprehensive markdown reports with syntax highlighting
• Metadata tracking (duration, cost, API usage)
• Copy to clipboard functionality
• Download as Markdown or export as JSON

• Node.js 20+ (Download)
• Anthropic API Key (Get one here)
• Exa API Key (Get one here)

1. Clone the repository:

git clone https://github.com/dair-ai/deep-ai-research.git
cd deep-ai-research

2. Install dependencies:

npm install

3. Configure environment variables:

Copy the example environment file and add your API keys:

cp .env.example .env.local

Then edit .env.local and add your API keys:

# Anthropic API Key (required)
# Get your key at: https://console.anthropic.com/
ANTHROPIC_API_KEY=sk-ant-api03-your-actual-key-here

# Exa API Key (required for search functionality)
# Get your key at: https://exa.ai/
EXA_API_KEY=your-actual-exa-key-here

# Optional: Default model
DEFAULT_MODEL=claude-haiku-4-5-20251001

# Optional: Budget limits (in USD)
MAX_BUDGET_USD=2.0

# Optional: Max turns per conversation
MAX_TURNS=20

4. Run the development server:

npm run dev

5. Open the app: Navigate to http://localhost:3001

Note: The app will use port 3001 if port 3000 is already in use.

Try these research queries to see the agent in action:

• "Find latest papers on multimodal LLMs"
• "What are emerging trends in reinforcement learning?"
• "Explain mixture of experts in modern LLMs"
• "Recent advances in attention mechanisms"
• "The latest papers on multi-agent systems"

Progress Tab:

• Shows real-time research progress
• Displays current stage (searching, fetching, analyzing, etc.)
• Statistics panel with search counts
• Timeline of all tool calls (expandable for details)

Final Report Tab:

• Activated when research is complete
• Markdown-formatted comprehensive report
• Citations and references
• Metadata (cost, duration, API usage)
• Export options (copy, download MD, export JSON)

┌─────────────────────────────────┐
│   Claude Orchestrator Agent     │
│   (claude-haiku-4-5-20251001)  │
└──────────────┬──────────────────┘
               │
       ┌───────┴────────┐
       │                │
       ▼                ▼
┌─────────────┐  ┌─────────────┐
│ Exa Search  │  │ File Tools  │
│ - search    │  │ - Read      │
│ - contents  │  │ - Write     │
│ - similar   │  │ - Edit      │
└─────────────┘  └─────────────┘

Current Phase: Phase 1 - Exa Neural Search Integration

Agent Memory: Uses CLAUDE.md for comprehensive research instructions and workflows

deep-ai-research/
├── app/                           # Next.js App Router
│   ├── api/
│   │   └── agent/
│   │       └── query/
│   │           └── route.ts       # SSE streaming endpoint
│   ├── page.tsx                   # Main dashboard UI
│   ├── layout.tsx                 # Root layout
│   └── globals.css                # Global styles
│
├── components/                    # React components
│   ├── ui/                        # shadcn/ui components
│   ├── ProgressTracker.tsx        # Research progress display
│   ├── ToolCallSummary.tsx        # Tool call visualization
│   └── FinalReport.tsx            # Report renderer
│
├── lib/                           # Core libraries
│   ├── agent/
│   │   ├── config.ts              # Agent configuration
│   │   ├── tools.ts               # Exa MCP tools
│   │   └── agents/                # Subagent definitions (Phase 2)
│   ├── types/
│   │   └── agent.ts               # TypeScript type definitions
│   └── utils.ts                   # Utility functions
│
├── .claude/                       # Claude Code config
│   ├── agents/                    # Future subagent definitions
│   └── settings.json              # Agent settings
│
├── CLAUDE.md                      # Agent memory & instructions
├── .env.local                     # Environment variables (create this)
└── package.json                   # Dependencies

• Framework: Next.js 16.0.1 (App Router)
• Language: TypeScript 5
• Agent SDK: @anthropic-ai/claude-agent-sdk v0.1.37
• Search: Exa API
• UI Library: shadcn/ui
• Styling: Tailwind CSS
• Icons: Lucide React
• Markdown: react-markdown with syntax highlighting

# Start development server
npm run dev

# Build for production
npm run build

# Start production server
npm start

# Run linting
npm run lint

1. app/api/agent/query/route.ts: API endpoint that handles research queries and streams responses via SSE
2. lib/agent/config.ts: Agent configuration including tools, prompts, and settings
3. lib/agent/tools.ts: Exa search tool definitions using MCP (Model Context Protocol)
4. CLAUDE.md: Agent memory file with research workflows and best practices
5. components/ProgressTracker.tsx: Extracts and displays research progress from message stream

To add new tools to the agent:

1. Define the tool in lib/agent/tools.ts using createSdkMcpServer and tool
2. Add the MCP server to mcpServers in lib/agent/config.ts
3. Add tool names to allowedTools array (format: mcp__<server-name>__<tool-name>)
4. Update UI components to recognize the new tools (icons, colors, stage mapping)

Example:

const myToolsServer = createSdkMcpServer({
  name: "my-tools",
  version: "1.0.0",
  tools: [
    tool(
      "my_tool",
      "Description of what the tool does",
      { /* zod schema */ },
      async (args) => { /* implementation */ }
    )
  ]
});

Port already in use:

• The app will automatically try port 3001 if 3000 is busy
• Or manually specify: PORT=3002 npm run dev

API Key errors:

• Ensure .env.local is created in the root directory
• Verify API keys are valid and not expired
• Check keys don't have leading/trailing spaces

Build errors:

• Clear Next.js cache: rm -rf .next
• Delete node_modules and reinstall: rm -rf node_modules && npm install
• Check Node.js version: node --version (should be 20+)

Agent not responding:

• Check browser console for errors
• Verify API endpoint is working: curl http://localhost:3001/api/agent/query
• Review server logs in terminal

Enable verbose logging by checking the terminal output where npm run dev is running. All agent messages and tool calls are logged.

• Next.js + TypeScript setup
• Claude Agent SDK integration
• Exa neural search tools
• Dashboard UI with real-time progress
• SSE streaming
• Report generation and export
• Tool call visualization

• Specialized subagents (web-researcher, analyzer, report-generator)
• Multi-agent orchestration
• Research database for session persistence
• Advanced filtering and search options
• PDF export capabilities
• Citation management
• Collaborative research sessions

Contributions are welcome! To contribute:

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-feature
3. Make your changes and test thoroughly
4. Commit with clear messages: git commit -m "Add: feature description"
5. Push to your fork: git push origin feature/my-feature
6. Open a Pull Request

• Additional search integrations (arXiv API, Semantic Scholar, etc.)
• Enhanced UI components and visualizations
• Export formats (PDF, LaTeX, BibTeX)
• Testing infrastructure
• Documentation improvements
• Performance optimizations

MIT License - see LICENSE file for details

• Built with Claude Agent SDK
• Search powered by Exa
• UI components from shadcn/ui

Built with Claude Code 🤖