STORM is an LLM system that writes Wikipedia-like articles from scratch based on Internet search. Co-STORM extends it to enable human–AI collaborative knowledge curation through a collaborative discourse protocol and a dynamically updated mind map.
Conducting Internet-based research to collect references, generating an outline, and producing full-length articles with citations on a given topic; Co-STORM supports human-AI collaborative knowledge curation and information seeking. It has been found helpful by experienced Wikipedia editors in their pre-writing stage.
STORM is a research-and-analysis LLM system that writes Wikipedia-like articles from scratch. Its design output is a generated article/outline — an advisory artifact for the user's pre-writing stage. The README explicitly notes the system cannot produce publication-ready articles and is meant to help editors in their pre-writing stage, signaling advisory intent. No autonomy signals were detected in the entrypoint, no confirmation gates exist, and the system does not take consequential multi-step actions on external systems. Code surface shows outbound network calls are for LLM inference and search/retrieval (reading data), and filesystem writes are for producing the article output. Although the surface analysis flags 'executes arbitrary code' (likely litellm/dynamic provider loading or eval in the pipeline), the agent is advisory by design, so it is capped at Tier 2.