Skip to main content
Back to timeline
lightpanda-io/browserSource publication:

No Chromium fork: a Zig-built headless browser runs 16x lighter and 9x faster than Chromium across 933 real pages

Synopsis

Lightpanda released a headless browser written from scratch in Zig for AI agents and automation, rather than a Chromium fork or a WebKit patch. In a benchmark requesting 933 real web pages over the network on an AWS EC2 m5.large instance, it is reported as 16x lighter and 9x faster than Chromium. It exposes a CDP server that existing Puppeteer and Playwright clients connect to, and a fetch command that dumps HTML, Markdown, PNG, PDF or structured data directly.

Source-provided article image: lightpanda-io / browser
GitHub Trending

Interpretation

A headless browser now exists as a new implementation written from scratch in Zig, aimed at AI agents and automation rather than built on Chromium or WebKit. The prior route was to fork Chromium or patch WebKit; this implementation takes neither path. In a benchmark requesting 933 real web pages over the network on an AWS EC2 m5.large instance, it is reported as 16x lighter and 9x faster than Chromium.

It is offered as a CDP server, so existing automation scripts only change the connection address. A Puppeteer script that connected to a local browser process instead connects to Lightpanda's CDP server through browserWSEndpoint, with the rest of the script unchanged. The published Puppeteer example connects to ws://127.0.0.1:9222, then creates a browser context, opens a page and extracts links as usual.

A command-line fetch mode retrieves a page and emits the format the caller needs. The fetched result no longer has to pass through a downstream converter, because --dump selects html, markdown, png, pdf or structured_data directly. structured_data prints the page's JSON-LD, OpenGraph, Twitter card, meta and link tags as JSON, matching the MCP structuredData tool.

Perspective

The public material shows a benchmark requesting 933 real web pages over the network on an AWS EC2 m5.large instance, plus the fetch and serve modes; there is no native Windows or Android build, and the Linux binaries depend on glibc, so they do not run directly on musl distributions or Termux.

The 16x lighter and 9x faster multipliers are reported for 933 real web pages on an AWS EC2 m5.large instance, and the material does not say whether the ratio holds on other page sets or hardware.