ai-team-os

Turn Claude Code into a multi-agent team operating system with persistent coordination, task management, and autonomous loop execution.

What is AI Team OS?

AI Team OS is a Claude Code plugin that adds full team orchestration capabilities to your AI workflow. It provides 155 MCP tools, 25 agent templates, and 14 hook scripts across 12 lifecycle events to coordinate multiple AI agents on complex projects — with persistent state, meeting systems, task walls, and a company-style loop engine.

Installation

Option 1: Local Development (recommended for contributors)

git clone https://github.com/CronusL-1141/AI-company.git
cd AI-company
claude --plugin-dir ./plugin

Option 2: GitHub Plugin Marketplace

In any Claude Code session:

/plugin install CronusL-1141/AI-company

Option 3: From source (script install)

git clone https://github.com/CronusL-1141/AI-company.git
cd AI-company
python3 install.py

python3 install.py (repo root) is the primary installer; the older python scripts/install.py still works but the root version is preferred. The install script will:

• Check Python 3.11+ and Node.js availability
• Install Python dependencies editable against the working tree (pip install -e .; greenlet is a core dependency so Apple Silicon installs directly)
• Build the Dashboard (if Node.js is available)
• Create data directory at ~/.claude/data/ai-team-os/
• Generate .mcp.json for MCP tool discovery

Features

Category	Details
MCP Tools	155 tools across team, task, workflow, loop, meeting, memory, channel, git, guardrail, and debate domains
Agent Templates	25 pre-built agent roles (tech-lead, researcher, reviewer, debate roles, etc.)
Hook Events	14 hook scripts wired across 12 lifecycle events: SessionStart, SubagentStart, SubagentStop, PreToolUse, PostToolUse, TaskCreated, TaskCompleted, SessionEnd, Stop, UserPromptSubmit, PermissionDenied, PreCompact
Team Management	Create teams, register agents, assign roles, track status
Task Wall	Decompose, assign, and monitor tasks across agents
Loop Engine	Autonomous company loop: plan → execute → review → iterate
Watchdog	Health checks, issue reporting, system self-healing
Meeting System	Structured meetings with conclusions and action items
Memory Store	Persistent cross-conversation memory search
Auto Port Discovery	API server finds available port automatically, avoids multi-project conflicts
MCP HTTP Streamable	/mcp/ endpoint mounted on FastAPI for HTTP-based MCP access

System Requirements

• Python 3.11+
• SQLite (included with Python)
• Claude Code (latest)
• Node.js 18+ (optional, for Dashboard UI)

Quick Start

1. Install using one of the methods above
2. Open your project directory in Claude Code
3. Create a team: /os-up
4. Start working: /os-task

Commands

Command	Description
/os-up	Start the OS and create a team
/os-status	View team and system status
/os-task	Manage tasks
/os-meeting	Start or join a meeting
/os-doctor	Run health diagnostics
/os-hooks	Manage hook configuration
/os-help	Show help information
/os-init	Initialize project setup

Troubleshooting

Plugin not loading

• Ensure Python 3.11+ is installed: python3 --version
• Run the install script: python3 install.py
• Check MCP server is running: python3 -m aiteam.mcp.server --check

MCP tools not showing

• Verify .mcp.json exists in project root
• Restart Claude Code after installation
• Run /os-doctor for automated diagnostics

Hooks not firing

• Check hooks.json is present in plugin/hooks/
• Use --plugin-dir ./plugin flag when starting Claude Code locally

License

MIT — see LICENSE