AI Maestro 代理協調系統

Hacker News·4 個月前

AI Maestro 是一個開源的 AI 代理協調系統，旨在為 Claude 等 AI 編碼代理提供記憶搜尋、程式碼圖譜查詢和代理間通訊等進階功能。它提供一個統一的儀表板，用於跨多台機器管理多個 AI 代理。

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

To see all available qualifiers, see our documentation.

AI Agent Orchestrator with Skills System - Give Claude Code agents superpowers: memory search, code graph queries, agent-to-agent messaging. Manage Claude, Aider, Cursor from one dashboard. Multi-machine support.

License

Uh oh!

There was an error while loading. Please reload this page.

23blocks-OS/ai-maestro

Folders and files

Latest commit

History

Repository files navigation

AI Maestro

Stop juggling terminal windows. Orchestrate your AI coding agents from one dashboard.

Quick Start • Features • Updating • Documentation • Contributing

The Problem

Your AI agents are scattered across terminals, computers, and cloud servers. They forget everything between sessions. They can't talk to each other. You're the bottleneck - copy-pasting context, relaying messages, and manually coordinating their work.

The Solution

AI Maestro is an AI Agent Orchestrator that gives your agents superpowers:

Your agents become a coordinated team, not isolated tools.

One Dashboard, Unlimited Machines

Benefits:

✨ Features

🌐 Manager/Worker Architecture (New in v0.8.0!)

Distribute your AI agents across unlimited machines - all managed from one browser dashboard.

⚠️ macOS 15+ Users: If setting up distributed workers, you MUST run ./scripts/fix-pm2-daemon.sh first to fix Local Network Privacy blocking. See Known Issues for details.

Example Setup:

→ See Setup Tutorial | Use Cases | Concepts Guide

Universal Agent Support

Works with any terminal-based AI:

Smart Organization

Agent Management

Agent Communication System

Agent Intelligence System (New in v0.11!)

Your AI agents become smarter over time with persistent memory and deep code understanding.

See Agent Intelligence Guide for setup and configuration

Portable Agents (New in v0.15!)

Move your AI agents anywhere. Export, import, transfer, and clone agents across machines.

See Multi-Computer Guide for detailed transfer workflows

Built for Speed

Access from Anywhere

🚀 Quick Start

Windows Users: See Windows Installation Guide for WSL2 setup (5-10 minutes)

Zero to Hero (Easiest - for new users)

macOS/Linux - One command installs everything:

With options:

Windows - Install via WSL2:

Full Windows guide: docs/WINDOWS-INSTALLATION.md

This installs:

Time: 5-10 minutes (depending on what's already installed)

Manual Install (for developers)

Configure tmux for optimal scrolling (highly recommended):

This enables:

Configure SSH for tmux sessions (CRITICAL for git operations):

This ensures:

Start the dashboard:

Dashboard opens at http://localhost:23000

Network Access: By default, AI Maestro is accessible on your local network at port 23000. See Security below for important information.

⚠️ After System Restart: tmux and the dashboard won't auto-start by default. To avoid socket errors after restart, see Auto-start Setup Guide for one-time configuration using macOS LaunchAgents and pm2.

Optional: Configure settings

2. Create Your First Agent

Option A: From the UI (Recommended)

Option B: From Terminal (For tmux users)

💡 Hierarchy Tip: Agent names with hyphens create automatic organization:

3. Start Coding

Your agent is now live in the dashboard. Click to switch between agents. Add notes. Organize your work. That's it.

🔄 Updating AI Maestro

AI Maestro automatically checks for updates and shows you when a new version is available.

Update Notification

When a new version is released, you'll see a green badge in the footer next to the version number:

Click the badge to see:

Note: The update check requires internet access. If you're running in an air-gapped environment, it will silently skip the check - AI Maestro works perfectly offline.

One-Command Update

To update AI Maestro to the latest version:

What this updates:

Example output:

Manual Update (Alternative)

If you prefer to update manually:

After Updating

Important: After any update, restart your Claude Code agents to reload updated skills:

This ensures your agents have the latest skills and capabilities.

📱 Access from Mobile Devices

AI Maestro is fully mobile-optimized, letting you monitor and control your AI agents from your phone or tablet - perfect for checking progress while away from your desk.

Secure Remote Access with Tailscale

The best way to access AI Maestro from anywhere is using Tailscale - a zero-config VPN that creates a secure network between your devices.

Note: AI Maestro is not endorsed by or affiliated with Tailscale in any way. We simply use it and recommend it based on our positive experience.

Why Tailscale?

Setup Guide

Install Tailscale on your development machine:
Install Tailscale on your mobile device:
Connect both devices:
Start AI Maestro:
Access from your mobile device:

Find your machine name in Tailscale settings (e.g., macbook-pro, desktop-work)

Mobile Features

Use Cases

📬 Inter-Agent Communication

The next evolution in AI pair programming: Your agents can now talk to each other.

When you're running a backend-architect agent and a frontend-developer agent, they need to coordinate. The backend agent finishes an API endpoint and needs to notify the frontend agent. The frontend agent hits an error and needs help from the backend team. Previously, you were the middleman - copying messages, switching contexts, losing flow.

Not anymore.

How It Works

AI Maestro provides a dual-channel messaging system designed specifically for agent-to-agent communication:

Perfect for detailed requests, specifications, and async collaboration:

Features:

For when agents need immediate attention:

Three delivery methods:

Real-World Use Case

Claude Code Integration

Every agent can use the messaging system automatically via a Claude Code skill (📁 View Skill):

No manual scripting needed - agents understand natural language messaging commands.

Installation:

Built-In UI

Each agent has a Messages tab with:

Get Started in 2 Minutes

Easy Install (Recommended):

Update Existing Installation:

Manual Install: See Installation Guide

For Claude Code: Install the skill to use natural language - Skills Guide

Documentation

Why This Matters

Before: You're the bottleneck. Every agent interaction goes through you.

After: Agents coordinate directly. You orchestrate, they collaborate.

Result: Faster development, better context retention, true multi-agent workflows.

📸 Screenshots

Agents organized automatically using hyphens, with color coding and icons:

Example agent names:

Displays as:

Each top-level category gets a unique color automatically - no configuration needed.

Take notes for each agent. They're saved automatically to your browser:

🎯 Why AI Maestro?

🧠 Agents That Remember

Every agent has persistent memory powered by CozoDB. They learn your codebase and remember past conversations. No more re-explaining context every session.

💬 Agents That Communicate

Direct agent-to-agent messaging. Your frontend agent can request APIs from backend agent without you playing messenger. Built-in inbox/outbox with priority levels.

🌐 Agents Everywhere

Run agents on your laptop, remote servers, Docker containers, or cloud VMs. Manage them all from one dashboard with the Manager/Worker architecture.

🗺️ Agents That Understand

Code Graph visualization shows your entire codebase structure. Agents know what files relate to what before they even start. Delta indexing keeps everything current (~100ms updates).

Why not just use tmux directly?
You can! AI Maestro is built on tmux. But instead of memorizing keybindings and switching between panes, you get visual organization, point-and-click switching, persistent memory, agent communication, and Code Graph visualization.

Is it just a tmux GUI?
It started that way, but now it's an AI Agent Orchestrator. Think of it as tmux + memory + communication + code understanding + visual hierarchy. You still have full access to your tmux sessions from the terminal.

📋 Requirements

macOS

Windows

Linux

🛠️ Tech Stack

Built with modern, battle-tested tools:

📚 Documentation

Getting Started

Manager/Worker (Distributed Setup)

Agent Communication

Agent Intelligence

General

🗺️ Roadmap

Phase 1 ✅ Complete

Phase 2 ✅ Complete

Phase 3 ✅ Complete (v0.8.0)

Phase 4 ✅ Complete (v0.11.0)

Phase 5 (2026)

Phase 6 (Future)

🤝 Contributing

We love contributions! AI Maestro is built for developers, by developers.

Ways to contribute:

See CONTRIBUTING.md for guidelines.

⚠️ Important Notes

Security

⚠️ Network Access Enabled by Default

AI Maestro runs on 0.0.0.0:23000 which means:

Safe for:

NOT safe for:

For the most secure setup, restrict AI Maestro to only accept connections from your local machine:

Option 1: One-time run

Option 2: Persistent configuration (recommended)

Create a .env.local file in the project root:

Then run normally:

Production build:

Testing your configuration:

Agent Logging (Disabled by Default)

AI Maestro can optionally log terminal agent content to ./logs/{agentName}.txt files. This is useful for:

What gets logged:

Controls:

Privacy considerations:

Disk usage:

Built-in protections:

Recommended practices:

⚠️ Not for production use without adding:

⚠️ Known Issues

If you're setting up Manager/Worker architecture on macOS 15+ (Sequoia) or macOS 26+ (Tahoe), you MUST apply this fix.

Symptoms:

Root Cause: macOS Local Network Privacy restricts PM2 (user-level process) from accessing local network IPs. This is a macOS security feature introduced in macOS 15.

Quick Fix (5 minutes):

What this does:

Alternative: Use Tailscale to connect workers (may bypass restriction).

Full documentation: See GitHub Issue #24 for complete technical details and troubleshooting.

Known Limitations

When Claude Code updates status indicators (like "Thinking..."), you may see duplicate lines in the scrollback buffer. This is a known limitation of xterm.js (the terminal library used by VS Code, JupyterLab, and most web terminals).

Why this happens:

Workarounds included:

Note: This is not specific to AI Maestro - it affects all web terminals using xterm.js with tools that update status indicators in place.