Run lightweight headless browsing for AI agents and data extraction
When you need to process thousands of web pages simultaneously on a server, this tool performs the task while significantly saving memory and CPU resources. It is not a fork of a Chrome-based browser but a new browser engine written from scratch in the Zig language. It provides only the core features required for AI agents or automation scripts to read and interact with web pages.
16x less memory and 9x faster speed compared to Chrome
A benchmark was conducted on an AWS EC2 m5.large instance requesting 933 actual web pages over the network. When processing 100 pages, peak memory usage was 123MB, which is approximately 16 times less than the 2GB used by headless Chrome. The execution time was also 5 seconds, approximately 9 times faster than headless Chrome's 46 seconds. This performance difference directly impacts cost and resource efficiency when launching large numbers of instances on a server.
Agent mode and MCP server support
You can control the browser with natural language commands through the built-in agent mode. If you instruct it to navigate pages, click, fill forms, or extract structured data, the browser executes these actions directly. The work logs generated during this process are saved as a JavaScript file called Pandascript, allowing for deterministic re-execution later without an LLM. Additionally, the MCP server allows multiple agents to use the browser simultaneously in independent sessions, ensuring that each agent does not contaminate the cookies or page state of others.
Installation methods and platform support
It can be installed via Homebrew, Arch Linux repositories, or nightly build binaries. It supports Linux x86_64 and aarch64, as well as macOS x86_64 and aarch64, and provides an official Docker image. Since there is no native binary for Windows, you must follow the Linux installation steps in a WSL2 environment. The Linux binary is linked to glibc, so it will not run on musl-based distributions like Alpine Linux; you must use a glibc-based image or build from source.
Implemented web standards and limitations
It implements features such as CORS, HTTP loader, HTML parser, DOM tree, JavaScript engine (v8), DOM API, Ajax, Fetch API, CDP WebSocket server, click, form input, cookies, proxy, network interception, robots.txt compliance, WebDriver Bidi, and ad blocker. Because it lacks a graphics rendering engine, it focuses on data extraction and interaction rather than visual screen output. Web Platform Tests (WPT) results are published daily to verify standard compliance, though running the full test suite takes a considerable amount of time.
Things to check before use
This browser is suitable for automation centered on data collection and API calls rather than tasks requiring visual rendering. It may not be suitable for cases requiring complex CSS layouts or visual verification. Usage telemetry is collected by default, so if you do not want this, you must set the environment variable LIGHTPANDA_DISABLE_TELEMETRY=true. Additionally, it requires Zig version 0.15.2 and dependencies on v8, Libcurl, and html5ever, so you must prepare this environment in advance if building from source.
