Vercel’s Agent-Browser Goes Fully Native Rust, Cuts Memory Use by 18x & Install Size by 99x

Vercel has moved agent-browser to a fully native Rust implementation, a change the company says improves performance and efficiency.

According to Chris Tate, the update delivers 1.6× faster cold starts, 18× lower memory consumption, and a 99× smaller installation footprint compared to the previous architecture.

“Agent-browser is now fully native Rust. Less abstraction means faster shipping, more control, and capabilities that weren’t possible before,” Tate wrote while announcing the release.

The tool now includes more than 140 commands covering navigation, interaction, state management, network control, debugging, and multi-engine browser support.

Reacting to Tate’s announcement on X, one user questioned the impact of the rewrite, arguing that the tool still runs Chrome underneath and would therefore only reduce memory marginally.

Tate responded that the motivation was primarily architectural rather than just performance gains.

“Faster shipping, more control, and capabilities that weren't possible before… This is the main reason. The CLI was already Rust—we just rewrote the daemon as Rust too,” he said.

Earlier versions of agent-browser used a hybrid architecture, with a Rust command-line interface communicating with a Node.js daemon that managed Playwright browser instances. The latest release rewrites that daemon in Rust, making the entire toolchain native and removing the Node.js runtime layer.

Vercel introduced agent-browser in early 2026 through its Vercel Labs initiative as an open-source command-line tool designed specifically for AI agents. 

The tool enables coding assistants such as Claude Code, Cursor, or Copilot to control a real web browser using simple commands, allowing them to open pages, click elements, fill forms, take screenshots, and run automated workflows.

Unlike traditional browser automation frameworks that return large amounts of DOM or accessibility data, agent-browser is optimised for AI agents by returning compact snapshots of actionable elements, reducing the amount of context an AI model needs to process during automation tasks.

The project is open source under the Apache License 2.0 and is positioned as a lightweight alternative to tools like Playwright for AI-driven development workflows, where agents need to verify deployments, test UI flows, or complete multi-step web tasks autonomously.