Agent TARS leverages a combination of advanced computer vision and natural language processing techniques to understand and manipulate graphical user interfaces. By capturing visual representations of web pages, TARS can identify buttons, forms, tables, and other page elements. Users interact with TARS through natural language prompts, instructing it to click, scroll, extract text, or fill forms across multiple pages. It supports customizable workflows that chain tasks—such as logging into accounts, scraping data, and exporting results to CSV or JSON. With support for headless and headful browser modes, TARS enables both interactive exploration and unattended automation, making it ideal for testing, data acquisition, and routine browser-based operations.

AutoBrowse is a developer library enabling LLM-driven web automation. By leveraging large language models, it plans and executes browser actions—searching, navigating, interacting, and extracting information from web pages. Using a planner-executor pattern, it breaks down high-level tasks into step-by-step actions, handling JavaScript rendering, form inputs, link traversal, and content parsing. It outputs structured data or summaries, making it ideal for research, data collection, automated testing, and competitive intelligence workflows.