A production-ready, all-in-one Docker image designed for AI agents and autonomous systems that need to execute code across multiple programming languages.

Why AgentBox?

AI agents often need to:

Execute code in multiple languages (Python, JavaScript, Go, etc.)
Process documents, images, and multimedia
Interact with web browsers for automation
Access databases and external services

Instead of managing separate containers or installing dependencies on-the-fly, AgentBox provides everything pre-configured in a single, secure image.

What's Inside

Multi-language runtime: Python, Node.js, Go, Ruby, Rust
Document processing: PDF, Excel, PowerPoint, Word, HTML
Browser automation: Playwright with Chromium & Firefox
Data tools: pandas, numpy, database clients (MySQL, PostgreSQL, SQLite)
Media processing: ffmpeg, image manipulation
Development tools: git, build tools, debuggers

Quick Start

docker run -it --rm -v $(pwd):/workspace dify/agentbox:latest

Use Cases

AI Agent Execution: Run LLM-generated code safely in isolated environments
Code Interpreters: Execute user code across multiple languages
Document Processing: Convert, extract, and transform documents at scale
Browser Automation: Scrape websites, test web applications, automate workflows
CI/CD Pipelines: Universal build environment with all tools included

Managing This Repository

Project Structure

dify-agentbox/
├── versions/versions.yaml    # Version configuration for all packages
├── Dockerfile.j2              # Jinja2 template for generating Dockerfile
├── build.py                   # Build script to render and build images
└── .github/workflows/         # CI/CD automation

Making Changes

1. Update Package Versions

Edit versions/versions.yaml:

languages:
  python:
    version: "3.12"
  nodejs:
    version: "20"

python_packages:
  - name: "pandas[excel,html,xml]"
    version: "~=2.2.3"

2. Render Dockerfile locally

# Install dependencies
uv sync

# Render Dockerfile
uv run python build.py

# Build the image with Docker
docker build -t dify/agentbox:dev .

3. Test Changes

# Run the image
docker run -it --rm dify/agentbox:dev

# Verify installations
python --version
node --version
go version

CI/CD Workflow

Images are automatically built and pushed when:

Push to main: Builds and tags as latest
Create tag (e.g., v1.0.0): Builds and tags with semantic version
Pull request: Validates build without pushing

Multi-architecture builds (amd64, arm64) are enabled by default.

Adding New Packages

System packages (apt):

system_packages:
  utilities:
    - your-package-name

Python packages:

python_packages:
  - name: "your-package"
    version: "~=1.0.0"

Node.js packages:

nodejs_packages:
  - your-global-package

After making changes, rebuild and test locally before pushing.

License

Apache 2.0