Storm research

storm-research

Stanford's STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective question asking) — an LLM-powered research pipeline that produces Wikipedia-quality articles with citations.

Upstream: https://github.com/stanford-oval/storm

Install

pip install knowledge-storm

Usage with Claude

STORM has native Claude support via its ClaudeModel class:

from knowledge_storm import STORMWikiRunnerArguments, STORMWikiRunner
from knowledge_storm.lm import ClaudeModel

claude_model = ClaudeModel(
    model="<your-model>",  # e.g. claude-sonnet-4-6
    max_tokens=4096
)

runner_args = STORMWikiRunnerArguments(
    output_dir="./output"
)

runner = STORMWikiRunner(runner_args, claude_model, claude_model)
runner.run(
    topic="Your Research Topic",
    do_research=True,
    do_generate_outline=True,
    do_generate_article=True,
    do_polish_article=True
)

API Keys Required

1. ANTHROPIC_API_KEY — for Claude models
2. Search engine key — one of: Bing, You.com, Serper, Brave, or Tavily

Co-STORM Mode

Co-STORM enables collaborative research where you and the LLM investigate together through multi-perspective questioning. Start a Co-STORM session to iteratively refine research direction.

How It Complements Loam

• /researcher — quick web research, single-pass
• STORM — systematic multi-perspective literature review with citations

Use /researcher for fast lookups. Use STORM when you need a structured article with proper citations.

Requirements

• Python 3.x
• dspy, litellm, streamlit
• Two API keys minimum (LLM + search engine)

Notes

• MIT license, Stanford-backed
• Development pace has slowed — may be less actively maintained
• No MCP server or Claude Code skill integration — requires Python invocation
• Best-in-class research quality for systematic reviews