Repolode
ntegrals

ntegrals/openbrowser

A TypeScript framework that lets AI agents control a Playwright browser to autonomously complete web tasks like clicking, typing and extracting data.

Stars
9,581
Forks
866
Open issues
20
Last commit
7 months ago

What is it?

Open Browser gives an AI agent a browser it can control. You describe a task in natural language, and the agent sends the current page state to an LLM, which chooses commands such as click, type, scroll, navigate and extract. The commands run against a Playwright browser, and the agent loops until the task completes, with stall detection to catch repeated failures.

The project is a monorepo with a core library, a CLI and a sandbox package. The CLI offers an agent `run` command, direct browser commands like open, click, screenshot and eval, and an interactive REPL for live browser sessions. The core library can be imported directly in TypeScript to create a viewport and agent.

Model support covers OpenAI, Anthropic and Google through the Vercel AI SDK, selectable by a provider flag. The sandbox package adds timeouts, memory limits, domain allowlists and step limits for running agents in constrained environments.

Key features

  • 01Runs natural-language tasks by letting an LLM choose browser actions
  • 02Supports OpenAI, Anthropic and Google models via a provider flag
  • 03Provides direct browser commands and an interactive REPL
  • 04Includes a sandbox with timeouts, memory limits and domain restrictions
  • 05Offers URL allow and block lists, proxy and cookie file configuration

Use cases

  • Automating multi-step web workflows such as filling forms or checking prices
  • Extracting structured content from websites for research
  • Debugging and prototyping browser interactions interactively
  • Running untrusted agent browsing in resource-limited sandboxes

Quick start

  1. 1

    Install dependencies

    bun install
  2. 2

    Set up your API keys by copying the example env file, then edit it with your keys

    cp .env.example .env
  3. 3

    Run an agent with a natural language task

    bun run open-browser run "Find the top story on Hacker News and summarize it"

Things to know

  • Requires an API key from OpenAI, Anthropic, or Google to run agents.
  • Uses Bun as the package runner, which must be installed separately.

In collections

Topics

FAQ

What is openbrowser?
A TypeScript framework that lets AI agents control a Playwright browser to autonomously complete web tasks like clicking, typing and extracting data. Open Browser gives an AI agent a browser it can control. You describe a task in natural language, and the agent sends the current page state to an LLM, which chooses commands such as click, type, scroll, navigate and extract. The commands run against a Playwright browser, and the agent loops until the task completes, with stall detection to catch repeated failures.
Is openbrowser free to use? What license does it have?
openbrowser is open source under the MIT license. Check the license terms before using it in commercial products.
Is openbrowser still maintained?
No commits for 6+ months. The last commit to openbrowser was on Apr 2, 2026.
What can I use openbrowser for?
Automating multi-step web workflows such as filling forms or checking prices Extracting structured content from websites for research Debugging and prototyping browser interactions interactively Running untrusted agent browsing in resource-limited sandboxes
How do I install openbrowser?
Install dependencies: bun install Set up your API keys by copying the example env file, then edit it with your keys: cp .env.example .env Run an agent with a natural language task: bun run open-browser run "Find the top story on Hacker News and summarize it"
Where can I get openbrowser?
The source code and installation instructions are on GitHub: https://github.com/ntegrals/openbrowser

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects