Multi Agent Researcher — AI skill for Claude Code

Multi-Agent Research System

A sophisticated multi-agent research system that uses [Instructor](https://github.com/jxnl/instructor) for structured LLM outputs and [Exa.ai](https://exa.ai) for neural web search.

Features

• 🤖 Multi-Agent Architecture: Lead researcher decomposes queries and coordinates parallel subagents
• 🔍 Neural Search: Integrates Exa.ai for semantic web search
• 📊 Structured Outputs: Uses Instructor to get typed, validated responses from LLMs
• 💾 Persistent Memory: Stores research plans and intermediate results
• 🔄 Iterative Refinement: Subagents evaluate and refine their searches
• ⚡ Parallel Execution: Multiple agents work simultaneously for faster results

Quick Start

Installation

# Clone the repository
git clone 
cd researcher

# Install with uv (recommended)
pip install uv
uv sync

# Or with pip
pip install -e .

Environment Setup

# Required for LLM decomposition
export ANTHROPIC_API_KEY="your-anthropic-key"

# Required for web search (optional - will use mock data without it)
export EXA_API_KEY="your-exa-key" 

Run Examples

# Simple research query
uv run python examples/simple_research.py

# Compare AI frameworks
uv run python examples/comparative_research.py

# Find recent AI developments
uv run python examples/time_bounded_research.py

# Academic paper research
uv run python examples/academic_research.py

# Demo with mock data (no API keys needed)
uv run python examples/demo_with_mocks.py

Architecture

The system implements a hierarchical multi-agent architecture:

User Query → Lead Researcher → Query Decomposition (via Instructor)
                ↓
        Parallel Subagents → Iterative Search → Result Synthesis
                ↓
        Memory Storage → Citation Addition → Final Report

Key Components

1. LeadResearcherV2: Orchestrates the research process using Instructor for structured task decomposition
2. **R