Deep Research ✨ NEW ✨

With the latest "Deep Research" trend in the AI community, we're excited to implement our own Open source deep research capability! Introducing GPT Researcher's Deep Research - an advanced recursive research system that explores topics with unprecedented depth and breadth.

Each deep research takes around 5 minutes to complete and costs around $0.4 (using o3-mini on "high" reasoning effort)

How It Works

Deep Research employs a fascinating tree-like exploration pattern:

1. Breadth: At each level, it generates multiple search queries to explore different aspects of your topic
2. Depth: For each branch, it recursively dives deeper, following leads and uncovering connections
3. Concurrent Processing: Utilizes async/await patterns to run multiple research paths simultaneously
4. Smart Context Management: Automatically aggregates and synthesizes findings across all branches
5. Progress Tracking: Real-time updates on research progress across both breadth and depth dimensions

Think of it as deploying a team of AI researchers, each following their own research path while collaborating to build a comprehensive understanding of your topic.

Process Flow

Quick Start

from gpt_researcher import GPTResearcher
from gpt_researcher.utils.enum import ReportType, Tone
import asyncio

async def main():
    # Initialize researcher with deep research type
    researcher = GPTResearcher(
        query="What are the latest developments in quantum computing?",
        report_type="deep",  # This triggers deep research modd
    )
    
    # Run research
    research_data = await researcher.conduct_research()
    
    # Generate report
    report = await researcher.write_report()
    print(report)

if __name__ == "__main__":
    asyncio.run(main())

Configuration

Deep Research behavior can be customized through several parameters:

• deep_research_breadth: Number of parallel research paths at each level (default: 4)
• deep_research_depth: How many levels deep to explore (default: 2)
• deep_research_concurrency: Maximum number of concurrent research operations (default: 4)
• total_words: Total words in the generated report (recommended: 2000)

You can configure these parameters in multiple ways:

1. Environment Variables:

export DEEP_RESEARCH_BREADTH=4
export DEEP_RESEARCH_DEPTH=2
export DEEP_RESEARCH_CONCURRENCY=4
export TOTAL_WORDS=2500

2. Config File:

deep_research_breadth: 4
deep_research_depth: 2
deep_research_concurrency: 4
total_words: 2500

researcher = GPTResearcher(
    query="your query",
    report_type="deep",
    config_path="path/to/config.yaml"  # Configure deep research parameters here
)

Progress Tracking

The on_progress callback provides real-time insights into the research process:

class ResearchProgress:
    current_depth: int       # Current depth level
    total_depth: int         # Maximum depth to explore
    current_breadth: int     # Current number of parallel paths
    total_breadth: int       # Maximum breadth at each level
    current_query: str       # Currently processing query
    completed_queries: int   # Number of completed queries
    total_queries: int       # Total queries to process

Error Handling

The deep research workflow is designed to be resilient:

• Failed queries are automatically skipped
• Research continues even if some branches fail
• Progress tracking helps identify any issues

Best Practices

1. Start Broad: Begin with a general query and let the system explore specifics
2. Monitor Progress: Use the progress callback to understand the research flow
3. Adjust Parameters: Tune breadth and depth based on your needs:
   • More breadth = wider coverage
   • More depth = deeper insights
4. Resource Management: Consider concurrency limits based on your system capabilities

Limitations

• Usage of reasoning LLM models such as o3-mini
• Deep research may take longer than standard research
• Higher API usage and costs due to multiple concurrent queries
• May require more system resources for parallel processing

Happy researching! 🎉