STORM
Open-source research system from Stanford University that turns web sources into long, structured reports with citations.
What it does
STORM researches a topic on the web and writes a long, structured report in the style of an encyclopedia article, with citations. To do this, the system asks questions from different perspectives, simulates conversations with experts and drafts an outline first. In the Co-STORM variant you follow a discussion between AI agents and steer it with your own questions; a mind map organises the information collected. You can use the web version free of charge after signing in with a Google account and accepting the terms of service. The code is available as a Python package (knowledge-storm) under the MIT licence. You connect language models via LiteLLM, for example from OpenAI or Azure; sources can be search services such as Bing, Google, Brave, Tavily, DuckDuckGo, SearXNG or You.com, or your own documents in a vector database.
Who it suits
Anyone who needs a structured overview of a new topic quickly, for example for market research, background texts or preparing specialist articles. Development teams can build the library into their own research applications.
Data protection for businesses
The web version is a research preview run by Stanford University in the USA. The vendor does not offer EU data hosting and does not mention a data processing agreement. The web version uses your Google account for sign-in. The team has already used input from the web version to build a research dataset (WildSeek); it does not describe how to opt out. There are no business plans. For confidential topics, it is better to host STORM yourself – data protection then depends on the model and search providers you connect.
Limits
According to the team, the reports are not ready for publication and need substantial editing. Self-hosting requires Python skills and API keys. There is no support and no availability commitment. The project is barely being developed any more: the latest package update (1.1.1) and the latest commit to the code repository date from September 2025 (as of November 2026).