OmniThink is a machine writing framework designed to enhance long-form article generation by simulating human cognitive processes of iterative expansion and reflection. It targets researchers and developers looking to create more insightful and knowledge-dense content, offering a structured approach to outline generation and article composition.

How It Works

OmniThink employs a unique mechanism that mimics human thinking to deepen understanding of complex topics. This iterative expansion and reflection process aims to enhance knowledge density and expand the boundaries of machine writing, resulting in coherent and contextually robust articles.

Quick Start & Requirements

• Installation: Clone the repository and install dependencies via pip install -r requirements.txt within a Python 3.11 conda environment. Docker is also available (docker pull zjunlp/omnithink:latest).
• Prerequisites: Requires LM API key and SEARCH API key set as environment variables (LM_KEY, SEARCHKEY). The LM output must be a LIST.
• Running: Execute sh run.sh. Articles, outlines, and mindmaps are saved in ./results/.
• Demos: Online demos are available on Modelscope and HuggingFace Spaces.

Highlighted Details

• Simulates human cognitive behaviors for iterative expansion and reflection.
• Focuses on expanding knowledge boundaries for information-rich articles.
• Supports outline and comprehensive article generation.
• Includes evaluation scripts for Rubric_Grading, Knowledge_Density, and Information_Diversity using FActScore.

Maintenance & Community

The project is implemented by DsPY and STORM, with contributions from Zhangjiabao-nudt and techshoww. Contact is available via email or GitHub issues.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README.

Limitations & Caveats

The README does not specify compatibility for commercial use or closed-source linking. The project appears to be recently open-sourced (January 2025) and may still be under active development.