Autonomous AI researcher that hits APIs, runs experiments, and ships research reports in minutes.

AutoResearch Lab is a multi-agent system that conducts autonomous research using Claude AI agents. It orchestrates five specialized agents that work together to plan, execute, analyze, critique, and report on research questions.

• 🔄 Reflexion Loop - System learns from past research and improves over time
• 📦 Mock Data Fallbacks - Works without API keys using realistic mock data
• 🎨 Colorful Console Output - Beautiful, readable output with chalk
• 💾 Persistent Learning - Stores insights in Redis for future research
• 🔌 Flexible API Integration - Seamless fallback to mock data when APIs unavailable
• ⚡ Command-Line Interface - Easy to use: node src/index.js "Your question"

• Agents: Anthropic Claude (Planner, Executor, Analyst, Critic, Reporter)
• Memory: Redis (learnings + cache)
• APIs: TRM Labs, Finster AI, Senso via Postman
• Security: Skyflow
• Knowledge Base: Sanity CMS
• Compute: AWS / Parallel
• UI: Chalk for colorful console output

The system uses five specialized AI agents:

1. Planner Agent - Creates structured research plans with clear objectives
2. Executor Agent - Executes API calls and collects data (with mock data fallback)
3. Analyst Agent - Analyzes results and identifies insights
4. Critic Agent - Provides critical evaluation and extracts learnings
5. Reporter Agent - Synthesizes findings into comprehensive reports

All agents use:

• Redis for caching and storing learnings
• Reflexion loops to improve through self-critique
• Mock data for development without API credentials
• Sanity CMS to build a knowledge base over time

# Install dependencies
npm install

# Copy environment template
cp .env.example .env

# Add your Anthropic API key to .env
# ANTHROPIC_API_KEY=your_key_here

# Run the system with a question
node src/index.js "What are the highest risks in DeFi bridges right now?"

# Or run the default demo
npm run dev

# Run tests
npm test

The system requires an Anthropic API key as a minimum. Other services are optional:

• ANTHROPIC_API_KEY - Your Anthropic Claude API key

• Redis - For caching and learning storage (falls back to mock if unavailable)
• TRM Labs API - For blockchain/crypto research
• Finster AI API - For AI-powered analysis
• Senso API - For additional data sources
• Skyflow - For secure data handling
• Sanity CMS - For building a persistent knowledge base
• AWS - For parallel compute and storage

See .env.example for all available configuration options.

Command Line (Easiest):

# Run with a custom question
node src/index.js "What are the highest risks in DeFi bridges right now?"

# Run default demo question
npm run dev

Programmatically:

import { ResearchOrchestrator } from './src/orchestrator.js';

const orchestrator = new ResearchOrchestrator();

const result = await orchestrator.conductResearch(
  'Your research question here',
  {
    useCache: true,        // Use cached results when available
    enableReflexion: true, // Enable self-improvement loops
  }
);

console.log(result.report);

1. Memory - Retrieves past learnings for context
2. Planning - Breaks down the question into actionable steps
3. Execution - Queries APIs and collects data (with mock fallback)
4. Analysis - Identifies patterns and insights
5. Reporting - Generates comprehensive report
6. Critique - Evaluates findings and saves learnings (Reflexion Loop)

The system continuously improves through the Reflexion Loop:

• 🔄 Critic agent evaluates each research and extracts key learnings
• 💾 Learnings are stored in Redis, categorized by topic
• 🧠 Future research uses past learnings for better context
• 📈 Quality improves over time through accumulated insights

Example learnings:

// Get learnings for a specific topic
const defiLearnings = await redisService.getLearnings('defi');

// Save a new learning
await redisService.saveLearning('security', 
  'Always check multi-sig implementations in bridge protocols');

The system includes realistic mock data for all APIs, enabling:

• ✅ Development without API credentials
• ✅ Testing complete workflows offline
• ✅ Consistent, reproducible results
• ✅ Cost-free experimentation

Mock data is automatically used when:

• API keys are not configured
• API calls fail or timeout
• You're in development/testing mode

See examples/mock-data-demo.js for a full demonstration.

autoresearch-lab/
├── src/
│   ├── agents/          # AI agent implementations
│   │   ├── BaseAgent.js
│   │   ├── PlannerAgent.js
│   │   ├── ExecutorAgent.js
│   │   ├── AnalystAgent.js
│   │   ├── CriticAgent.js
│   │   ├── ReporterAgent.js
│   │   ├── prompts.js   # System prompts for all agents
│   │   └── index.js     # Agent wrapper functions
│   ├── services/        # External service integrations
│   │   ├── anthropic.js # Claude API with callClaude helper
│   │   ├── redis.js     # Redis with learning storage
│   │   ├── api.js       # External API clients
│   │   ├── mockData.js  # Mock data for development
│   │   ├── sanity.js
│   │   └── skyflow.js
│   ├── orchestrator.js  # Main research coordinator
│   └── index.js         # Entry point with CLI support
├── config/
│   └── index.js         # Configuration management
├── test/
│   └── test-basic.js    # Test suite
├── examples/            # Example scripts
│   ├── reflexion-loop.js    # Demonstrates learning system
│   ├── mock-data-demo.js    # Shows mock data usage
│   ├── custom-research.js
│   └── multiple-research.js
├── .env.example         # Environment template
├── package.json
└── README.md

Blockchain intelligence and crypto analytics.

AI-powered financial and business analysis.

Market data and sentiment analysis via Postman.

Each API can be queried through the Executor Agent:

await apiService.query('trmlabs', '/endpoint', { params });

Note: Mock data is automatically provided for all APIs when credentials are not configured.

# Run test suite
npm test

# Run mock data demo (no API keys needed)
node examples/mock-data-demo.js

# Run Reflexion Loop example (requires Anthropic API key)
node examples/reflexion-loop.js

The test suite validates:

• Mock data structure and availability
• Redis learning storage and retrieval
• Cache functionality
• Basic system integration

Sensitive data is protected using Skyflow:

• Tokenization of API keys and credentials
• Secure vault storage
• Data privacy compliance

# Install dependencies
npm install

# Run in development mode (with default question)
npm run dev

# Run with custom question
node src/index.js "Your research question"

# Run tests
npm test

# Try examples
node examples/mock-data-demo.js
node examples/reflexion-loop.js

MIT License

Contributions are welcome! Please feel free to submit a Pull Request.