Discover and explore top open-source AI tools and projects—updated daily.
jayminwestClaude Code agent swarm orchestrator
New!
Top 64.6% on SourcePulse
This project provides a project-agnostic swarm system designed to orchestrate multiple Claude Code agents, transforming a single session into a coordinated multi-agent team. It addresses the complexity of advanced AI-driven development workflows by enabling agents to collaborate on tasks, decompose objectives, and merge their work, offering a powerful tool for users seeking to scale their AI development capabilities.
How It Works
Overstory employs a two-layer architecture featuring a persistent coordinator agent that manages task decomposition and dispatch. Worker agents are spawned in isolated git worktrees using tmux, communicating via a custom SQLite mail system with a typed protocol and broadcast messaging. A mechanical watchdog daemon monitors agent health. This approach leverages git worktrees for isolation, a structured SQLite mail system for reliable communication, and tiered health monitoring to manage complex agent swarms robustly.
Quick Start & Requirements
bun install, and link the CLI globally using bun link.overstory init and install hooks via overstory hooks install.overstory coordinator start or spawn individual worker agents using overstory sling --capability <type>.STEELMAN.md for risk analysis and the Agentic Engineering Book (web version) for deeper insights.Highlighted Details
overstory dashboard) provides real-time monitoring of the agent fleet.Maintenance & Community
No specific details regarding notable contributors, sponsorships, or community channels (e.g., Discord, Slack) are provided in the README.
Licensing & Compatibility
The project is released under the MIT license, which is generally permissive for commercial use and integration into closed-source projects.
Limitations & Caveats
Agent swarms are explicitly warned against as a universal solution. Users must understand the inherent risks of multi-agent orchestration, including compounding error rates, cost amplification, debugging complexity, and merge conflicts, which are presented as the "normal case." A full risk analysis is available in STEELMAN.md.
22 hours ago
Inactive
ag2ai