Skyvern

Skyvern is an open-source web app and server that automates browser-based workflows using Large Language Models and computer vision. It replaces traditional RPA scripts that rely on brittle DOM parsing and XPath selectors with AI agents capable of reasoning through website layouts. The software is deployed as a server with a web-based UI or used via a Playwright-compatible SDK.

Users can automate manual tasks on any website without writing custom code for every single page. The system is designed to handle complex interactions, including form filling and data extraction, while remaining resistant to layout changes. It can be installed locally via pip or Docker Compose, or used through a managed cloud version that includes anti-bot detection and proxy networks.

Key features

• No-code workflow builder for non-technical users to automate browser tasks
• Playwright-compatible SDK that adds AI functionality to standard browser automation
• Natural language interaction via AI-powered page commands like act and validate
• Structured data extraction from web pages using optional JSON schemas
• AI-augmented element location that falls back to LLMs when CSS selectors fail
• Support for multi-step task execution through an autonomous agent swarm
• Integration with credential managers like Bitwarden and 1Password for authentication
• Ability to operate on previously unseen websites without customized code

Skyvern is built for developers and automation engineers who need to scale browser interactions across multiple different websites. It integrates with various LLM providers to map visual elements to actions, allowing a single workflow to be applied to various sites with different structures. The architecture supports both SQLite for simple local setups and Postgres for more scalable deployments. The SDK provides specific methods for downloading files, validating page states, and executing pre-built workflows through a high-level agent interface.

It is a flexible tool for AI-driven browser automation and robotic process automation, moving away from static selectors toward vision-based navigation.