We’re thrilled to announce the release of Co-STORM, a major update to the STORM project that brings collaborative human-AI knowledge curation into the spotlight! This new version empowers users to engage with language models in a more interactive and aligned way, transforming how we explore and curate knowledge together.

The knowledge-storm package is now at v1.0.0—make sure to upgrade by running:

pip install knowledge-storm --upgrade

News 🔥

• [2024/09] Co-STORM is now live and fully integrated into the knowledge-storm Python package. Try it out by upgrading the package today!
• [2024/09] Our paper on Co-STORM has been accepted to EMNLP 2024! You can read it here.

New Features 🎉

🚀 Co-STORM Engine

• Co-STORM is now integrated into the knowledge-storm package. Check out the API documentation for more details.
• We introduce the Agent interface in Co-STORM, providing a unified framework for defining different LM agent policies for information seeking and knowledge curation. You can explore this interface here.
  • Co-STORM LLM Experts: These agents ground their responses in external knowledge sources and ask follow-up questions based on the discourse.
  • Moderator Agent: Guides the conversation by generating insightful questions, drawing attention to discovered but underexplored areas.
  • Human Engagement: Users can inject their own utterances to steer the conversation, enabling an interactive and collaborative experience.

🧠 Dynamic Mind Map

Co-STORM introduces the KnowledgeBase data class that structures retrieved information into a concept-oriented hierarchy, forming a shared conceptual space between the user and the system. This is presented as a mind map in the graphical UI, allowing users to easily navigate and explore deep knowledge curation processes.

💻 Graphical UI Update

An interactive graphical UI of Co-STORM will be available soon on our live research preview website. Stay tuned for updates!

🔧 Modularity and Flexibility

• Just like STORM, Co-STORM is built using the dspy library, ensuring full modularity. You can easily customize both language models (LMs) and retrieval modules (RMs) for advanced use cases. Check out the customization guide.
• Co-STORM supports the same LMs and RMs as STORM. See the full list of supported models here.

Contributors 🙌

This release wouldn’t be possible without the hard work of:

• @Yucheng-Jiang
• @shaoyijia
• @dekunma