Storm

Storm is an open-source platform from Stanford designed for building, executing, and analyzing large-scale AI agent workflows. It provides a structured environment for orchestrating multi-step, multi-agent systems that combine language models, tools, and external data sources. Users can define complex workflows as graphs, where each node represents an agent, tool call, or transformation, and Storm manages the execution, data flow, and state tracking across the entire process.

A core capability of Storm is its support for reproducible experiments. It logs intermediate states, prompts, model outputs, and tool interactions, enabling detailed inspection, debugging, and comparison of different workflow configurations. This makes it well-suited for research on agent behavior, tool use, and long-horizon reasoning.