TFGrid AI Agent

AI-powered coding agent with loop technique for safe AI coding environments.

Overview

TFGrid AI Agent is a standalone application that provides an isolated, safe environment for AI-assisted coding. It uses the "loop technique" to enable AI agents to iteratively work on projects without risking your local development environment.

Features

• 🤖 AI-Powered Coding - Uses qwen-cli for AI assistance
• 🔄 Loop Technique - Iterative AI coding with safety
• 🏝️ Isolated Environment - Runs on dedicated VM
• 📦 Project Management - Create, edit, and manage projects
• 🔒 Safe - No access to your local files
• ⚡ Concurrent Projects - Run multiple AI agents simultaneously

Quick Start

Complete AI Development Environment (Recommended)

For the full experience with automatic Git hosting, use tfgrid-ai-stack:

# Deploy complete AI + Git environment
tfgrid-compose up tfgrid-ai-stack

# Authenticate with Qwen
tfgrid-compose login

# Create project (AI generates code + creates Git repo automatically)
tfgrid-compose create "modern web application"

# Monitor progress
tfgrid-compose monitor web-application

# View results in web interface
tfgrid-compose projects

Standalone AI Agent

For standalone usage without Git integration:

# Deploy just the AI agent
tfgrid-compose up tfgrid-ai-agent

# Connect to the agent
tfgrid-compose ssh tfgrid-ai-agent

# Use agent scripts directly
/opt/ai-agent/scripts/create-project.sh my-project
/opt/ai-agent/scripts/agent-loop.sh

Structure

tfgrid-ai-agent/
├── tfgrid-compose.yaml      # Deployment manifest
├── src/                     # Application source
│   ├── scripts/             # Agent scripts
│   └── templates/           # Project templates
├── deployment/              # Deployment hooks
│   ├── setup.sh             # Install dependencies
│   ├── configure.sh         # Configure service
│   └── healthcheck.sh       # Verify deployment
├── patterns/                # Pattern-specific configs
└── docs/                    # Documentation

Requirements

Minimum:

• 2 CPU cores
• 4 GB RAM
• 50 GB disk

Recommended:

• 4 CPU cores
• 8 GB RAM
• 100 GB disk

Dependencies

The deployment automatically installs:

• Node.js (>=18.0.0)
• npm (>=9.0.0)
• qwen-cli (@qwen-code/qwen-code)
• git, curl, wget

Configuration

Environment Variables

QWEN_API_KEY (optional)

• Qwen API key for AI access
• Can be set after deployment

PROJECT_WORKSPACE (default: /opt/ai-agent)

• Agent workspace directory
• All projects stored here

Patterns

Supported Patterns

single-vm (Recommended)

• Single VM deployment
• Private access via Wireguard/Mycelium
• Best for development

k3s (Future)

• Kubernetes deployment
• For team environments

Usage

Creating Projects

# Create from template
/opt/ai-agent/scripts/create-project.sh my-web-app

# Edit existing project
/opt/ai-agent/scripts/edit-project.sh my-web-app

# List all projects
/opt/ai-agent/scripts/list-projects.sh

# Delete project
/opt/ai-agent/scripts/delete-project.sh my-web-app

Running the Agent Loop

# Start the agent loop (interactive)
/opt/ai-agent/scripts/agent-loop.sh

# Advanced loop with more options
/opt/ai-agent/scripts/agent-loop-advanced.sh

Project Workspace

All projects are stored in /opt/ai-agent/projects/:

/opt/ai-agent/
├── projects/
│   ├── my-web-app/
│   ├── my-api/
│   └── my-tool/
├── logs/
└── scripts/

Logging

Logs are available via:

# Systemd logs
journalctl -u tfgrid-ai-agent -f

# Application logs
tail -f /var/log/ai-agent/output.log
tail -f /var/log/ai-agent/error.log

Service Management

The agent runs as a systemd service:

# Check status
systemctl status tfgrid-ai-agent

# Start/stop
systemctl start tfgrid-ai-agent
systemctl stop tfgrid-ai-agent

# Restart
systemctl restart tfgrid-ai-agent

# View logs
journalctl -u tfgrid-ai-agent -f

Troubleshooting

Service not running

# Check status
systemctl status tfgrid-ai-agent

# View logs
journalctl -u tfgrid-ai-agent -n 50

# Restart
systemctl restart tfgrid-ai-agent

qwen-cli issues

# Check installation
qwen --version

# Reinstall
npm install -g @qwen-code/qwen-code

Permission issues

# Fix workspace permissions
chown -R root:root /opt/ai-agent
chmod -R 755 /opt/ai-agent/scripts

Development

Local Development

You can run the agent locally without TFGrid:

# Clone the repo
git clone https://github.com/tfgrid-studio/tfgrid-ai-agent
cd tfgrid-ai-agent

# Run setup manually
sudo ./deployment/setup.sh
sudo ./deployment/configure.sh
./deployment/healthcheck.sh

Contributing

Contributions welcome! Please:

1. Fork the repo
2. Create a feature branch
3. Make your changes
4. Submit a pull request

Security

• Agent runs in isolated VM
• No access to local files
• All code execution in safe environment
• Review all AI-generated code before use

License

Apache 2.0

Support

• 📚 Documentation: docs.tfgrid.studio
• 🐛 Issues: GitHub Issues
• 💬 Discussions: GitHub Discussions
• 📧 Contact: tfgrid.studio/contact

Links

• TFGrid Studio: github.com/tfgrid-studio

---

Part of: TFGrid Studio
Status: ✅ Production Ready
Version: 0.3.0 Concurrent Projects: ✅ Multiple projects supported Compatible with: tfgrid-compose v0.11.0+ (all patterns)