browser-use/browsercode

What it solves

BrowserCode is a coding agent designed to automate browser interactions. It solves the problem of rigid browser automation tools by allowing an AI agent to write and execute JavaScript directly via the Chrome DevTools Protocol (CDP), enabling it to adapt to any website at runtime and create reusable scripts for future tasks.

How it works

The project turns browser interaction into a coding problem. Instead of using high-level abstractions, the agent uses a core primitive called browser_execute(code), which runs JavaScript in-process and communicates with Chrome through CDP. This allows the agent to maintain a session across calls and receive logs, values, and screenshots to inform its actions.

Who it’s for

It is intended for developers and users who need an autonomous agent capable of performing complex web tasks, such as QA testing local sites, extracting data from websites, or finding deals on e-commerce platforms.

Highlights

• Direct Browser Control: Uses unconstrained CDP to drive real browsers.

• Flexible Connectivity: Can connect to an existing browser tab, create a new local browser profile, or use a remote cloud browser.

• LLM Agnostic: Supports any model reachable via API key, including providers supported by OpenCode.

• Reusable Scripts: Automatically writes scripts to a local workspace for later reuse.

• Cloud Integration: Integrates with Browser Use Cloud for stealth, captcha solving, and proxies.