storm
LLM system that researches a topic and generates full-length reports with citations.
Build comprehensive sustainability reports and knowledge bases using multi-perspective AI research agents.
STORM is an LLM-powered knowledge curation system designed to research topics and generate full-length reports complete with citations. It suits energy researchers and sustainability analysts who need to synthesize large volumes of technical literature.
What it is
The system breaks long-form report generation into a pre-writing internet research stage and a writing stage, using perspective-guided question asking and simulated expert conversations. It also includes Co-STORM for human-AI collaborative knowledge curation with a dynamic hierarchical mind map.
Who it's for
It suits professionals who need to draft comprehensive background reports on complex energy and sustainability topics. It does not suit those looking for publication-ready outputs, as the README explicitly notes the system cannot produce final-draft-quality text without significant edits.
Key features
Perspective-Guided Question Asking
Surveys existing articles on similar topics to discover diverse angles and control the research query process.
Simulated Expert Conversations
Simulates a dialogue between a writer and a topic expert grounded in internet sources to update topic understanding.
Collaborative Co-STORM Mode
Implements a discourse protocol with AI experts, moderators, and human input alongside a dynamic concept mind map.
Flexible Search Integration
Supports multiple retrievers including YouRM, BingSearch, VectorRM for local documents, SerperRM, and TavilySearchRM.
Up and running in 3 steps
Our honest verdict
What's great
- Generates structured, full-length reports with proper citations from scratch.
- Supports human-in-the-loop collaboration via Co-STORM and dynamic mind maps.
- Integrates with litellm to let you choose your preferred language and embedding models.
- VectorRM allows grounding reports on your own proprietary user-provided documents.
Where it falls short
- Outputs require significant manual editing because they are not publication-ready out of the box.
- Requires active setup of external language model APIs and search engine credentials which incur ongoing costs.
- Running deep research loops with frontier models like GPT-4o can consume a high number of tokens.
Pricing & license
Free under its MIT license, plus the cost of running external language models and search APIs such as OpenAI and Bing.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Can I use my own documents instead of internet search?
Yes, STORM includes VectorRM to support grounding reports on user-provided documents alongside standard search engines.
What models are supported for generating the reports?
The package integrates with litellm, allowing you to use any language model and embedding model supported by that ecosystem.
Are the generated reports ready to publish directly?
No, the developers note that the system cannot produce publication-ready articles without a significant amount of human editing.
