Deep Research Agent

An AI-powered research agent that autonomously searches the web, analyzes content, and generates comprehensive research reports on any topic.

Features

• Automated Web Research: Generate search queries, scrape content, and synthesize information
• Multi-cycle Research: Progressively build deeper understanding with multiple research cycles
• Reflective Analysis: Identify knowledge gaps and areas for further exploration
• Structured Reports: Generate well-organized markdown reports with clear sections
• Multiple Interfaces: Use via command line, Python API, or Streamlit web UI

Installation

pip install deep-research-agent

Usage

Command Line

deep-research --topic "Your research topic" --cycles 2 --output report.md

Streamlit UI

deep-research-ui

Then open your browser to the URL shown (typically http://localhost:8501).

Python API

from deep_research_agent import ResearchConfig, ResearchController

# Configure research parameters
config = ResearchConfig(
    topic="Your research topic",
    max_research_cycles=2,
    max_search_results_per_query=5,
    max_urls_to_scrape_per_cycle=3
)

# Initialize and run the research controller
controller = ResearchController(config)
results = controller.run_full_research()

# Access the final report
print(results["final_report"])

Requirements

• Python 3.9+
• Local LLM (Ollama with QwQ model recommended)

Environment Configuration

Create a .env file with:

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=qwq

How It Works

1. Query Generation: Creates effective search queries based on the topic and current knowledge
2. Web Search: Retrieves search results from DuckDuckGo
3. Content Scraping: Extracts and cleans content from web pages
4. Summary Generation: Integrates new information with existing knowledge
5. Reflection: Identifies gaps and contradictions to guide further research
6. Report Generation: Creates a comprehensive final report

License

MIT License - see LICENSE file for details.