Overview

ClawTeam: Agent Swarm Intelligence (One Command → Full Automation) is an open-source Python framework that enables AI agents to form dynamic teams, coordinate in real time, and achieve complex goals with zero ongoing human intervention. Built by HKUDS, it transforms isolated agents into self-organizing swarms using simple CLI orchestration.

Provide a goal once — the leader agent spawns workers, assigns tasks with dependencies, handles messaging, and delivers results. Fully compatible with Claude Code, Codex, OpenClaw, nanobot, Cursor, and any CLI-based agent.

Key Features

One-Command Swarm Launch: Start complete teams with predefined templates (e.g., clawteam launch hedge-fund --team fund1).
Agent Self-Organization: Leader automatically spawns and manages specialized workers with auto-injected coordination prompts.
Workspace Isolation: Every agent runs in its own git worktree (separate branch) to prevent merge conflicts.
Task & Dependency Management: Shared kanban, dependency chains, and auto-unblocking.
Inter-Agent Communication: Inbox messaging, broadcasts, and file/ZeroMQ transport.
Real-Time Monitoring: Terminal kanban, auto-refresh dashboards, tiled tmux views, and optional web UI.
Team Templates: TOML-defined archetypes for instant swarm creation.
Plan Approval & Lifecycle: Agents submit plans for review; graceful shutdown and idle notifications.
Cross-Machine & Multi-User Support: Works across machines or users via shared filesystem or P2P.
Structured JSON Output: Perfect for scripting and integration.

Use Cases

Autonomous ML Research: Spawn 8+ agents across GPUs to run thousands of experiments. Example: Improved LLM validation bpb from 1.044 to 0.977 (6.4% gain) with zero human input.
Agentic Full-Stack Development: Leader spawns architect, backend, frontend, and tester agents that code, test, and merge a complete app autonomously.
AI Hedge Fund: One-command launch of 7-agent team (Portfolio Manager + analysts) for real-time market analysis, risk assessment, and trading decisions.
Custom Teams: Scientific research squads, investment committees, business ops teams, or content studios.

Getting Started

pip install clawteam

Or install from source:

git clone https://github.com/HKUDS/ClawTeam.git && cd ClawTeam && pip install -e .

Requires Python 3.10+, tmux, and a compatible CLI agent. Optional P2P support: pip install -e ".[p2p]".

Launch examples:

clawteam spawn tmux claude --team myteam
clawteam board attach myteam to monitor live.

Supported Agents

Full support for Claude Code, Codex, OpenClaw, nanobot. Experimental for Cursor. Works with any shell-executable CLI agent.

Why ClawTeam?

Move beyond single-agent limitations. ClawTeam unlocks true collective intelligence for multi-day, multi-step workflows — making AI agents dramatically more powerful and practical. MIT licensed, actively maintained (v0.1.2 released March 2026).