<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> ClawTeam enables "Agent Swarm Intelligence," allowing AI agents to self-organize into collaborative teams for complex task execution. It addresses the isolation of individual agents by facilitating autonomous division of labor, real-time insight sharing, and goal convergence. This system benefits researchers, developers, and power users by automating complex workflows across AI research, software engineering, and financial analysis, leading to faster delivery and breakthrough solutions.

How It Works

A leader agent orchestrates specialized sub-agents, all communicating and delegating tasks via simple CLI commands. Each agent operates in an isolated environment (git worktree, tmux window) to prevent conflicts. Task dependencies are managed, and progress is visualized via terminal or web dashboards. ClawTeam's novelty lies in agents self-organizing via CLI interactions, unlike frameworks requiring manual orchestration code. Its architecture uses a filesystem (~/.clawteam/) for state management, offering simplicity and crash safety without databases or servers.

Quick Start & Requirements

Install via pip install clawteam. Requires Python 3.10+, tmux, and a compatible CLI coding agent (e.g., claude, codex) installed and on PATH. Optional P2P transport: pip install -e ".[p2p]".

Highlighted Details

• Autonomous ML Research: 8 agents on 8 H100 GPUs conducted 2430+ experiments, improving LLM training (val_bpb: 1.044→0.977).
• Agentic Software Engineering: Automates full-stack development via specialized agents for API design, backend, frontend, and testing, with dependency resolution.
• AI Hedge Fund: TOML templates launch multi-agent investment teams with diverse strategies, coordinated by risk/portfolio managers.
• Workspace Isolation: Each agent uses a git worktree/branch for isolated experimentation.
• Inter-Agent Messaging: Supports point-to-point and broadcast messaging via file or ZeroMQ P2P transport.
• Monitoring: Terminal kanban, live dashboards, tiled tmux views, and a Web UI visualize swarm activity.

Maintenance & Community

A roadmap details future enhancements (Redis Transport, Shared State, Agent Marketplace) towards production-grade v1.0. Contributions are welcomed for agent integrations, team templates, and transport backends.

Licensing & Compatibility

Released under the permissive MIT License, allowing free use, modification, and distribution for commercial and closed-source projects.

Limitations & Caveats

Integration with Cursor is "Experimental." Relies on external CLI agents functioning independently. Advanced cross-machine capabilities are planned for v0.4+, indicating current versions are primarily single-machine focused. Core production features (auth, permissions) are slated for v1.0.