Overview
STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking) is an open-source knowledge curation and writing engine from Stanford OVAL that performs internet-based research to generate outlines and produce citation-backed article drafts, useful for researchers and editors during pre-writing stages.
Key Features
- Two-stage writing pipeline: retrieval and outline generation followed by citation-aware article generation.
- Multi-perspective question asking: discovers diverse perspectives and simulates conversations to generate deeper research questions.
- Co-STORM collaborative mode: supports human-AI collaborative discourse for better alignment and curation.
- Rich retriever support: multiple retrievers (Bing, You, DuckDuckGo, Vector, etc.) and vector grounding options.
Use Cases
- Pre-writing and research assistance for academics and editors.
- Automated report, review, or Wikipedia-style article drafting.
- Educational tools and dataset generation for knowledge-base construction.
Technical Highlights
- Implemented in Python with a modular
knowledge_stormpackage, easy to extend for custom retrievers and model backends. - Integrates with litellm and other model adapters to flexibly switch language and embedding models.
- Provides example scripts, datasets (FreshWiki, WildSeek), and reproduction branches for research validation.