Stanford Storm

Stanford STORM Skill

This skill allows you to use Stanford STORM, an LLM-powered system for generating detailed, Wikipedia-style articles. It uses litellm for flexible LLM configuration.

Setup

1. Dependencies: Requires knowledge-storm and litellm.

   pip install knowledge-storm dspy-ai litellm python-dotenv
   

2. Configuration: STORM needs API keys for the LLM (e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY) and a Search Provider (BING_SEARCH_API_KEY or YDC_API_KEY). LiteLLM reads these standard environment variable names.

   # Ensure keys are set. Example for OpenAI and Bing:
   if [ -z "$OPENAI_API_KEY" ] || [ -z "$BING_SEARCH_API_KEY" ]; then
       echo "STORM requires API keys."
       echo "Ensure your LLM key (e.g., OPENAI_API_KEY) and Search key (BING_SEARCH_API_KEY or YDC_API_KEY) are set in .env."
       # Add interactive setup here if desired, ensuring the correct variable names are used.
   fi
   

Usage

Use the scripts/run_storm.py script to generate an article.

Command

python3 scripts/run_storm.py --topic "<topic>" [--rm-name <bing|you>] [--fast-model <model>] [--strong-model <model>]

Parameters

• --topic (Required): The subject to research.
• --rm-name (Optional): Retriever module (default bing). Ensure the corresponding API key is set.
• --fast-model (Optional): LLM for simulation/questions (e.g., gpt-3.5-turbo).
• --strong-model (Optional): LLM for outline/writing (e.g., gpt-4o, claude-3-5-sonnet-20240620).

Example

python3 scripts/run_storm.py --topic "The History of Quantum Computing" --strong-model gpt-4o --rm-name bing

Output

The script outputs the final article in Markdown format to stdout. Intermediate files (outline, raw research) are saved in the storm_output/ directory (logged to stderr). The process can take several minutes.