English | 简体中文 | 繁體中文 | 한국어 | 日本語

🚀 OpenManus-Max

An Advanced Autonomous AI Agent Framework with Multi-Level Permissions & Trust Attenuation

👋 Introduction

OpenManus is great, but OpenManus-Max takes it to the next level!

Inspired by the sandboxed security of Manus and the local execution flexibility of IronClaw, OpenManus-Max is a fully refactored, enterprise-grade AI Agent framework. It features a DAG Task Scheduler, Hierarchical Memory, 20+ Built-in Tools, and a revolutionary Multi-Level Permission Engine with Skill Trust Attenuation.

Whether you want an agent that safely runs in a Docker sandbox, or a fully unleashed local assistant that controls your desktop, OpenManus-Max has you covered.

✨ Key Features

• 🛡️ Multi-Level Permission Engine: Choose between YOLO (full access), STANDARD (intercepts high-risk), STRICT (requires approval), or SANDBOX (Docker isolation).
• 🧩 Skill System & Trust Attenuation: Load custom SKILL.md files. Third-party skills automatically trigger "Trust Attenuation", stripping the agent of write/execute permissions to prevent prompt injection attacks.
• 🧠 Hierarchical Memory: Working Memory $\rightarrow$ LLM-driven Episodic Summary $\rightarrow$ Global Blackboard.
• ⚡ DAG Task Scheduler: Automatically breaks down complex goals into Directed Acyclic Graphs (DAG) for parallel execution.
• 🛠️ 20+ Built-in Tools: Python/Shell execution, Web Search (Multi-engine), Deep Web Crawling, Vision Analysis, Data Visualization, Desktop Automation (RPA), and more.
• 🔌 MCP & A2A Protocol: Built-in Model Context Protocol (MCP) client and Agent-to-Agent (A2A) HTTP server.
• ⏰ Routine Daemon: Run scheduled tasks (Cron/Interval) in the background with SQLite persistence.

🚀 Installation

We recommend using uv for a faster installation experience.

# 1. Clone the repository
git clone https://github.com/your-repo/OpenManus-Max.git
cd OpenManus-Max

# 2. Create virtual environment
uv venv --python 3.11
source .venv/bin/activate

# 3. Install dependencies
uv pip install -e ".[all]"

⚙️ Configuration

Create a config.toml file in the root directory (you can copy from config.example.toml):

cp config.example.toml config.toml

Edit config.toml to set your API keys and preferred permission mode:

[llm]
model = "gpt-4o"
base_url = "https://api.openai.com/v1"
api_key = "sk-..."

[permission]
# Choose from: yolo, standard, strict, sandbox
mode = "standard"

🎯 Quick Start

Run OpenManus-Max in interactive mode:

openmanus-max

Other Execution Modes

# Single task mode
openmanus-max -t "Clean up my Downloads folder"

# YOLO mode (Full local access, no confirmations)
openmanus-max --mode yolo -t "Analyze system logs and fix errors"

# Sandbox mode (Safe execution in Docker)
openmanus-max --mode sandbox -t "Run this untrusted script"

# DAG Planning mode for complex tasks
openmanus-max --dag "Research AI trends and generate a PPT presentation"

# Start Routine Daemon
openmanus-max --routine

📚 Skill System

You can extend the agent's capabilities by adding skills to ~/.openmanus-max/skills/. A skill is simply a directory containing a SKILL.md file with YAML frontmatter:

---
name: github-pr-reviewer
version: "1.0.0"
description: Review GitHub Pull Requests
activation:
  keywords: ["github", "pr", "review"]
---
# GitHub PR Reviewer
When asked to review a PR, follow these steps...

Note: Skills placed in ~/.openmanus-max/installed/ are treated as untrusted and will automatically attenuate the agent's dangerous tools.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgements

Special thanks to the original OpenManus team and the IronClaw project for their inspiring architectures.