Agent OS is an open-source framework designed to provide a secure, isolated environment for running autonomous agents. By leveraging containerization technology, Agent OS ensures that each agent operates within its own sandbox, minimizing security risks and preserving system integrity. This tool is ideal for developers and researchers building scalable, safe, and efficient agent-based systems. Agent OS currently utilizes the Browserless image to power its browser instances.

• Secure Execution
  Agents operate in isolated containers, preventing interference and bolstering security.

• Autonomous Task Management
  Supports a wide range of automated tasks, including web browsing, file operations, and system interactions.

• Seamless Playwright Integration
  Easily integrates with Playwright for advanced automation and testing capabilities.

• Scalable Infrastructure
  Efficiently spawns, manages, and monitors multiple agents concurrently.

• Command Execution
  Execute shell commands directly within a container for enhanced control and debugging.

Note: Agent OS is currently in development. The features below reflect its initial capabilities, with more to come soon!

Currently, Agent OS supports creating a browser instance for AI agents using the Browserless image. To get started, simply run:

agentos create --name my-container

This command sets up a containerized browser environment ready for agent operations, accessible at IP-ADDRESS:3000.

Puppeteer can connect to the Agent OS browser instance via the browserWSEndpoint option. Here's how to adapt your code:

Before:

const puppeteer = require('puppeteer');
const browser = await puppeteer.launch();

After:

const puppeteer = require('puppeteer');
const browser = await puppeteer.connect({
  browserWSEndpoint: 'ws://IP-ADDRESS:3000',
});

Agent OS supports Playwright's remote connection protocols. Here's an example using Firefox:

Before:

const pw = require('playwright');
const browser = await pw.firefox.launch();

After:

const pw = require('playwright-core');
const browser = await pw.firefox.connect({
  wsEndpoint: 'ws://IP-ADDRESS:3000/firefox/playwright',
});

Run shell commands directly inside a running container using the exec command. This feature allows you to interact with the container environment seamlessly.

agentos exec --containerId <container-id> --cmd "<shell-command>"

To execute a simple command like echo Hello, World in a container:

agentos exec --containerId mycontainerId --cmd "echo Hello, World"

• --containerId (required): The ID of the container where the command will be executed.
• --cmd (required): The shell command to run inside the container.
• The container must be in a "running" state for the command to execute successfully.
• Output from the command is logged for easy inspection.

This feature is useful for debugging, testing, or performing one-off tasks within the containerized environment.