Forkybot AI: Optimized and Secured Bytebot

This is an optimized and secured fork of Bytebot, an open-source AI agent that controls a virtual desktop to automate complex tasks.

The original project was archived and left unmaintained. This fork rescues the idea, modernizes it, and prepares it for production, applying software engineering, security, and documentation best practices.

🎯 What does this project do?

Imagine a virtual employee that has its own computer. You can give it instructions in natural language such as "Download this month's invoices from my three supplier portals and organize them in a folder" and the agent will open the browser, navigate the portals, download the files, and organize them. All autonomously, but with the option for you to take control at any time (Human-in-the-loop).

✨ Key Improvements (Engineering Phases)

This fork is not a simple copy. Three engineering phases were carried out to take it to the next level:

Phase A: Security (CVE-2026-78250 Patch)

• Infinite loop mitigation: Configurable safety limits (MAX_TASK_ITERATIONS and TASK_TIMEOUT_MS) were implemented to prevent denial of service (DoS) and uncontrolled API spending.
• Human-in-the-loop: If a task exceeds the limits, it does not fail. It transitions to NEEDS_HELP so that a human can intervene and resume it.

Phase B: Optimization (85% size reduction)

• Migration to Debian Bookworm-slim: Maintains full compatibility with glibc, VS Code, and 1Password, but removes unnecessary software and build tools.
• Multi-Stage Build: The final image reduces its size from ~7 GB to ~1.09 GB, improving efficiency and reducing the attack surface.
• Deep cleanup: Thunderbird, xfce4-goodies, and redundant editors were removed.

Phase C: Integration Validation (Smoke Test)

• Concurrent startup of the desktop, the agent, the UI, and the database was verified.
• Internal communication (VNC on port 9990) and external communication (UI on port 9992) were checked.

📚 Living Documentation (Obsidian)

One of the major contributions of this fork is its Project Brain, an Obsidian vault (vault/) that serves as context for the AI (and humans) to navigate the architecture without getting lost. It includes:

• ADRs (Architectural Decision Records) that explain the why behind each change.
• Detailed File Maps for the Agent, the UI, and the Virtual Desktop.
• Work Protocols (PROTOCOLO_TRABAJO.md and PROTOCOLO_OBSIDIAN.md) that define how an AI agent should work on this project.

🛠️ Technologies Used

• Agent: NestJS (Node.js 20 Alpine)
• Virtual Desktop: Debian Bookworm-slim, XFCE, Xvfb, TigerVNC
• Frontend: Next.js 15, Tailwind CSS, noVNC
• Database: PostgreSQL 16
• Infrastructure: Docker, Docker Compose, Multi-Stage Builds

🚀 Getting Started

# Clone the repository
git clone https://github.com/YOUR_USERNAME/forkybot_AI.git
cd forkybot_AI

# Configure the environment variables (e.g. Gemini API Key)
echo "GEMINI_API_KEY=your_key_here" > docker/.env

# Bring up the stack
docker-compose -f docker/docker-compose.yml up -d

# Access the interface
# Open http://localhost:9992 in your browser