OpenManus Framework

To run OpenManus within a Daytona sandbox environment, you must first install the required dependencies in your existing OpenManus Python environment and then configure your Daytona API key. This setup allows the agent to use sandbox-specific tools like sb_browser_use, sb_shell, and sb_files to perform tasks in an isolated environment.

# 1. Activate your environment
conda activate 'Your OpenManus python env'

# 2. Install dependencies
pip install daytona==0.21.8 structlog==25.4.0

# 3. Prepare configuration
cd OpenManus
cp config/config.example-daytona.toml config/config.toml

# 4. Run the sandbox main script
python sandbox_main.py

Depending on the mode you want to use, run one of the following commands:

• Standard Mode: Run the general agent.

  python main.py

• MCP Tool Version: Run using Model Context Protocol tools.

  python run_mcp.py

• Multi-Agent Flow: Run the experimental multi-agent version.

  python run_flow.py

# Standard
python main.py

# MCP version
python run_mcp.py

# Multi-agent flow
python run_flow.py

If you require browser automation capabilities, install Playwright dependencies:

playwright install

OpenManus requires LLM API configuration.

1. Create your configuration file by copying the example:

cp config/config.example.toml config/config.toml

2. Edit config/config.toml to provide your API keys and model preferences.

Supported configuration sections include [llm] for global settings and [llm.vision] for vision-specific model settings.

The recommended installation method uses uv for faster dependency management.

1. Install uv:

   curl -LsSf https://astral.sh/uv/install.sh | sh

2. Clone the repository and enter the directory:

   git clone https://github.com/FoundationAgents/OpenManus.git
   cd OpenManus

3. Create and activate a Python 3.12 virtual environment:
   • Unix/macOS:

     uv venv --python 3.12
     source .venv/bin/activate

   • Windows:

     uv venv --python 3.12
     .venv\Scripts\activate

4. Install dependencies:

     uv pip install -r requirements.txt

5. (Optional) Install browser automation tools:

   playwright install

This integration combines the OpenManus agent with the A2A protocol (Agent-to-Agent). Note that current support is limited to non-streaming interactions.

Prerequisites

1. Activate your OpenManus Python environment (e.g., via conda).
2. Install the required SDK:

   pip install a2a-sdk==0.2.5

Running the A2A Server

Navigate to the OpenManus directory and run the server module:

cd OpenManus
python -m protocol.a2a.app.main

Running an A2A Client

You can use the A2A official samples to interact with the agent via CLI or a web frontend. To run the Python CLI sample:

git clone https://github.com/google-a2a/a2a-samples.git
cd a2a-samples
echo "GOOGLE_API_KEY=your_api_key_here" > .env
cd samples/python/hosts/cli
uv run .

To install the tool on Mac or Linux, you must first install Node.js (version 18 or higher) using nvm, then install the tool's dependencies.

1. Install and activate nvm and Node.js:

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
source ~/.bashrc
nvm install node
nvm use 22 # Replace 22 with your desired version

2. Install dependencies:

cd app/tool/chart_visualization
npm install

Alternatively, you can use conda to set up your environment:

1. Create and activate a new conda environment with Python 3.12:

   conda create -n open_manus python=3.12
   conda activate open_manus

2. Clone the repository and enter the directory:

   git clone https://github.com/FoundationAgents/OpenManus.git
   cd OpenManus

3. Install dependencies:

   pip install -r requirements.txt

When the agent performs tasks in the Daytona sandbox, you can monitor its progress and results using different interfaces depending on the tool being used:

• sb_browser_use: View live browser operations via the VNC link printed in the terminal (e.g., https://6080-sandbox-123456.h7890.daytona.work).
• sb_shell: View shell execution results via the Daytona Dashboard Sandboxes page.
• sb_files: The agent uses this tool to operate on files within the sandbox.

Once configured, you can start using OpenManus with the following commands:

• Standard Agent: Run the main entry point and type your ideas in the terminal:

  python main.py

• MCP Tool Version: To use the Model Context Protocol (MCP) tools version:

  python run_mcp.py

• Multi-Agent Version (Unstable): To run the experimental multi-agent flow:

  python run_flow.py