browsergym-core

What it is and what it does

BrowserGym is a gymnasium-based framework for building and evaluating web automation agents. It wraps Chromium via Playwright and provides a standardized interface for agents to interact with websites—observing page state, taking actions, and receiving rewards. The core package registers open-ended web tasks as gym environments, allowing you to implement custom agents that navigate, click, type, and reason about web content using HTML parsing (beautifulsoup4, lxml) and visual features (pillow, numpy).

The package is designed for research into web agent capabilities rather than production automation. It integrates with gymnasium's standard environment loop, making it straightforward to plug in your own agent logic—whether rule-based, learning-based, or LLM-driven. The framework is extensible: you can define new tasks by inheriting from AbstractBrowserTask. While the core package provides the foundation, additional benchmark packages (miniwob, webarena, workarena, etc.) add pre-built task suites; the core alone supports open-ended tasks and custom benchmarks.

Use it for:

• Research and development of web automation agents using reinforcement learning or LLM-based reasoning.
• Benchmarking agent performance on standardized web tasks like form filling, navigation, and information retrieval.
• Building custom web task evaluation suites by extending the AbstractBrowserTask class.
• Testing agent behavior on interactive, open-ended web scenarios with real Chromium rendering.
• Prototyping multi-step web workflows that require visual understanding and DOM interaction.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

BrowserGym provides a gymnasium environment for automating web tasks in Chromium, letting you build and test agents that interact with websites through a standardized interface.

Yes, if you are building or researching web automation agents. The low install friction, active maintenance, permissive license, and gymnasium integration make it a solid foundation. Not recommended for production web scraping or automation—it is explicitly a research tool. No known security vulnerabilities.

Install

browsergym-core on PyPI

pip install browsergym-core

uv add browsergym-core

poetry add browsergym-core

Installing browsergym-core

Before you install

Low friction installation with a pure Python wheel. The package is actively maintained with recent commits and has 1315 repository stars, indicating active development and community use.

License in practice

Licensed under Apache-2.0 (permissive), allowing commercial and private use with minimal restrictions—just retain attribution and license notices.

Quickstart

pip install browsergym-core

import gymnasium as gym
import browsergym.core

env = gym.make("browsergym/openended", task_kwargs={"start_url": "https://www.google.com/"})
obs, info = env.reset()
action = ...  # your agent logic
obs, reward, terminated, truncated, info = env.step(action)
env.close()

Requires Python >3.9 and Playwright browser installation (run `playwright install chromium` after pip install).

Verify before relying

• Whether browsergym-core alone is sufficient for typical use cases or if additional benchmark packages are typically required.
• Performance characteristics and resource requirements when running agents on complex web tasks.
• Compatibility and integration details with specific LLM backends beyond the demo agent examples.

Package facts

Evidence: browsergym_core-0.14.3-py3-none-any.whl

Tags

More Artificial Intelligence packages

Further reading