This project provides an agentic LLM workflow for comprehensive, multi-hop reasoning research on complex topics, mimicking human research processes. It's designed for researchers, students, and anyone needing in-depth, well-cited content, enhancing traditional web search with structured information gathering and source verification.

How It Works

The system employs a multi-stage, agentic LLM process that plans, searches, evaluates, and iterates to produce detailed research reports. It leverages multiple self-reflection stages to ensure quality information gathering and includes source verification with citations for all information. The architecture is designed for extensibility, allowing community contributions.

Quick Start & Requirements

• Install: Use uv (a faster alternative to pip) for installation.

  # Create and activate virtual environment
  uv venv --python=3.12
  source .venv/bin/activate
  # Install project dependencies
  uv pip install -r pyproject.toml
  

• Prerequisites: Python 3.12+, Pandoc, pdfLaTeX (via BasicTeX on macOS or texlive-xetex on Ubuntu).
• API Keys: Requires TOGETHER_API_KEY, TAVILY_API_KEY, and HUGGINGFACE_TOKEN.
• Usage: Run via CLI (python src/together_open_deep_research.py --config configs/open_deep_researcher_config.yaml) or Gradio webapp (python src/webapp.py).
• Docs: Overview, Features, Installation, Usage.

Highlighted Details

• Generates long-form, well-cited research reports.
• Supports output formats including PDF, HTML, and Podcast.
• Includes source verification and citations for all information.
• Extensible architecture for community contributions.

Maintenance & Community

The project is from Together Computer. Further community or roadmap details are not explicitly provided in the README.

Licensing & Compatibility

The README does not specify a license. Users should verify licensing terms before use, especially for commercial applications.

Limitations & Caveats

As an LLM-based system, it may generate hallucinations, exhibit biases from training data, misinterpret queries, or present outdated information. Users are advised to always verify critical information with primary sources.