Apppublic

nimesh24/computer-use-agent

sourceHugging Faceupdated 8mo agoView on Hugging Face

0likes

App README

Computer Use Agent 🤖

An AI autonomous agent powered by Google Gemini 2.5 and Playwright. It browses the web, extracts information, and saves summaries to local files, all while operating within a secure, sandboxed environment.

Now features a Streamlit Web Interface for easy interaction and Docker support for cloud deployment.

🌟 Features

—🧠 AI-Powered Browsing: Uses Gemini 2.5's computer use capabilities to navigate websites and interact with elements.
—🛡️ Secure by Design: Custom "Security Kernel" (security.py) prevents prompt injection, restricts browsing to allowed domains (Wikipedia), and sandboxes file access.
—💻 Dual Interface:
—Web UI: User-friendly dashboard built with Streamlit.
—CLI: Direct terminal access for power users.

—🐳 Docker Ready: Containerized environment for safe and easy deployment to Hugging Face Spaces or any cloud provider.
—📄 Automatic Summarization: Reads articles and creates structured text summaries in a designated output folder.

📂 Project Structure

text

computer-use-agent/
├── app.py           # 🖥️ Streamlit Web Interface (Entry Point)
├── agent.py         # 🧠 Main Agent Logic (Gemini + Playwright)
├── security.py      # 🔒 Security Guardrails (Policy Enforcement)
├── Dockerfile       # 🐳 Container configuration
├── requirements.txt # 📦 Python dependencies
├── .env             # 🔑 API Keys (Optional)
└── runs/            # 📂 Output folder for generated files

🚀 Quick Start

Prerequisites

—Google GenAI API Key: Get one from Google AI Studio.

Option 1: Run with Docker (Recommended)

This method guarantees the browser installs correctly without messing up your local machine.

1.Build the Container:

bash

docker build -t browser-agent .

1.Run the App:

bash

docker run -p 8501:8501 browser-agent

1.Open Browser: Go to http://localhost:8501

Option 2: Run Locally

1.Clone & Install:

bash

git clone https://github.com/yourusername/computer-use-agent.git
cd computer-use-agent

pip install -r requirements.txt
playwright install chromium

1.Start the Web UI:

bash

streamlit run app.py

1.Start via CLI (Alternative):

bash

# Set key first: export GOOGLE_API_KEY=AIzaSy...
python agent.py "Find information about Apollo 11"

🔒 Security Architecture

This project implements a Policy Enforcement Point (PEP) pattern in security.py to ensure the AI behaves safely.

Guardrail	Function	Description
Input Sanitization	`validate_goal()`	Blocks malicious keywords (e.g., `rm -rf`, `/etc/passwd`) and limits prompt length.
Domain Allowlist	`validate_url()`	SSRF Protection. The agent is strictly limited to browsing `*.wikipedia.org`. It cannot access local networks or malicious sites.
File Sandbox	`validate_file_path()`	Path Traversal Protection. Files can ONLY be written to the `./runs` directory.

🛠️ Configuration

You can configure the agent via the Web UI sidebar or by setting environment variables in a .env file.

Variable	Description	Default
`GOOGLE_API_KEY`	Your Gemini API Key	Required
`HEADLESS`	Run browser without UI (faster)	`true`
`SCREEN_WIDTH`	Browser viewport width	`1440`
`SCREEN_HEIGHT`	Browser viewport height	`900`
`MAX_TURNS`	Maximum actions per run	`8`

☁️ Deployment

Deploy to Hugging Face Spaces

1.Create a new Space on Hugging Face.
2.Select Docker as the SDK.
3.Upload all project files (Dockerfile, app.py, etc.).
4.Hugging Face will automatically build and launch your agent!

📝 License

This project is open-source.

Author: Sachintha Dissanayaka