This project provides a framework for orchestrating multiple Claude Code instances as a collaborative AI development team, enabling specialized agents with distinct roles, tools, and directory contexts. It's designed for developers and researchers tackling complex projects who need to delegate tasks across a hierarchy of AI agents, improving efficiency and specialization.

How It Works

Claude Swarm utilizes a YAML configuration to define swarm topologies, specifying instance roles, models, directories, and inter-instance connections. Communication between instances is managed via the Model Context Protocol (MCP), allowing agents to delegate tasks and share information. The system launches a main instance that coordinates other connected agents, each operating within its defined context and toolset.

Quick Start & Requirements

• Install: gem install claude_swarm or add to Gemfile and run bundle install.
• Prerequisites: Ruby 3.2.0+, Claude CLI installed and configured.
• Usage: Initialize with claude-swarm init or claude-swarm generate, then launch with claude-swarm.

Highlighted Details

• Supports multi-level swarm hierarchies for complex team structures.
• Allows mixing Claude and OpenAI providers within a single swarm.
• Features experimental session management and restoration capabilities.
• Includes Git worktree integration for isolated development environments.

Maintenance & Community

• The project is hosted on GitHub: https://github.com/parruda/claude-swarm.
• Contributions and bug reports are welcome via pull requests.

Licensing & Compatibility

• Licensed under the MIT License.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The session restoration feature is experimental and may not fully restore the main instance's conversation context.