Open-source Browserbase alternatives

Skyvern automates browser-based workflows using large language models and computer vision. It offers a Playwright-compatible SDK that adds AI capabilities on top of Playwright, and a no-code workflow builder so that both technical and non-technical users can automate manual tasks on any website, replacing brittle automation scripts.

Traditional browser automation depends on custom scripts for each site, often relying on DOM parsing and XPath selectors that break whenever the layout changes. Skyvern instead relies on vision-capable LLMs to understand and interact with pages. Its design was inspired by task-driven autonomous agents such as BabyAGI and AutoGPT, with the added ability to act on sites using browser automation libraries. A swarm of agents comprehends a website, plans actions and executes them.

Skyvern is written in Python and licensed under AGPL-3.0, and topics note RPA and workflow use. You can self-host it or use the hosted service at skyvern.com. It targets teams that automate tasks across many different websites, such as form filling or data retrieval, where writing and maintaining per-site scripts is costly.

BrowserAct gives AI agents and teams a way to automate real browsers for web data extraction and account-based workflows. The repository contains skills for agents such as Claude Code, Cursor, and Codex, along with a CLI. The project describes features for handling anti-bot protections, handing control to a human when an agent gets stuck, and running tasks in parallel without cross-contamination. It is released under the MIT license.

There are two ways to use it. In the cloud mode, you describe the website and the fields you need and BrowserAct builds and tests a reusable scraping bot in a cloud browser, then runs it for you. In the local mode, agents load the BrowserAct skill and drive a local browser, including reuse of local Chrome login state, from their own environment. Independent sessions and isolated multi-account browsing keep identities separate.

It suits developers building AI agents that need dependable browser access, and teams that want repeatable data collection. The cloud service is run by the vendor, so check browseract.com for plans and usage limits.

Magnitude is a browser agent that uses vision AI so you can control a browser with natural language. It navigates by looking at the interface and planning actions, interacts with precise mouse and keyboard actions, extracts structured data, and verifies results with a built-in test runner that supports visual assertions.

Rather than drawing numbered boxes around page elements, which the authors say generalizes poorly on complex modern sites, it has a visually grounded model give pixel coordinates, making it independent of the page's DOM structure and extendable to desktop apps or virtual machines. It can be used to automate web tasks, integrate apps that have no APIs, extract data, test web apps or build other browser agents. The README reports a 94% score on the WebVoyager benchmark.

Magnitude needs a large visually grounded model; Claude Sonnet 4 is recommended and Qwen-2.5VL 72B is also compatible. A setup command creates a project with an example script, and the test runner adds a tests directory with a configuration file and example test. It is written in TypeScript and licensed under Apache-2.0.