Pardus Browser is a headless browser built specifically for AI agents that eliminates the need for Chromium. Unlike traditional browsers, it doesn't render pixels or take screenshots — instead, it fetches URLs, parses HTML, and outputs structured semantic state in milliseconds. The tool provides AI agents with exactly what they need: knowledge of page content, interactive elements, and navigation paths.

Built entirely in Rust with zero dependencies on Chrome, Pardus Browser offers three output formats (Markdown, tree, JSON) and supports semantic tree output with ARIA roles, landmarks, headings, and interactive elements. The browser can strip static content to show only actionable elements, annotate interactive elements with their actions (navigate, click, fill, toggle, select), and typically completes parsing in under 200ms. It includes navigation graph capabilities for internal routes, external links, and form descriptors.

The architecture consists of a core library handling HTML parsing and semantic tree building, a planned CDP WebSocket server for Playwright/Puppeteer integration, and a CLI tool. Future roadmap items include JavaScript execution for SPAs, page interaction capabilities, session persistence, and optional screenshot support. Available as open source under the MIT License, Pardus Browser represents a fundamental rethinking of browser architecture for AI agent workflows.