Browser Use Rewrites Its Core in Rust Racing Toward 100k Stars

Browser Use teases a total rewrite as its Python AI web automation library closes in on 100k GitHub stars

·
·
·
Browser Use Rewrites Its Core in Rust Racing Toward 100k Stars
  • 100k milestone incoming: Browser Use is approaching 100,000 GitHub stars and teasing a "totally new version" to mark it.
  • Rust rewrite underway: An experimental Rust agent (from browser_use.rust import Agent) is already in the repo, signaling the direction of the rewrite.
  • CLI 2.0 shipped: The Browser Use CLI moved from Playwright to direct CDP, achieving 2x speed and 50% fewer tokens for coding agents.
  • Full platform now: The project has expanded to Browser Harness, stealth cloud browsers, proxies, and a 24/7 Claude Code box beyond the core Python library.
  • Current pain points: Memory usage, anti-bot detection, and parallel agent management remain the main open-source limitations driving users to the cloud tier.
  • Industry context: A broad shift toward Rust in AI agent infrastructure is underway in 2026, driven by memory safety, parallelism, and token efficiency needs.

Browser Use, the open-source Python library that lets AI agents control a real browser, just crossed a milestone that few open-source AI projects reach this fast: it's closing in on 100,000 GitHub stars. And the team isn't celebrating with a blog post or a changelog. They're promising something bigger: a totally new version of the library.

From zero to nearly 100k, faster than almost anyone

The project reached over 60,000 GitHub stars in just a few months, and the team secured over $17 million in seed funding. Since then the momentum hasn't stopped. The repo now sits at 78,000+ stars, is trusted by Fortune 500 companies, and bills itself as the #1 open-source browser automation platform. The GitHub releases page currently shows 86,400 stars, putting the 100k milestone well within sight.

For context, this is a library that didn't exist two years ago. Most production-grade developer tools take years to reach that kind of community size. Browser Use did it by solving a problem that suddenly became very real: how do you give an LLM a browser it can actually use?

What Browser Use actually does

Browser Use is an open-source Python project that makes websites accessible to AI agents. It identifies all interactive elements on a webpage, enabling agents to perform meaningful interactions with them. In short, the library allows AI to control and interact with your browser programmatically.

It combines advanced AI with robust browser automation to simplify web interactions for AI agents, integrating visual understanding and HTML structure extraction for more effective navigation and decision-making. Under the hood, the library combines visual input with AI control to automate browsers via Playwright.

The basic usage is minimal:

css
from browser_use import Agent
import asyncio
async def main():
    agent = Agent(
        task="Go to Amazon and find the cheapest USB-C hub with 4+ ports",
        llm=ChatBrowserUse(),
    )
    result = await agent.run()
    print(result.final_result())
asyncio.run(main())

You give it a task in plain English, point it at an LLM, and it figures out the rest: clicking, scrolling, form-filling, navigating across pages. The library has 79k+ GitHub stars and lets you connect any LLM and run locally or self-hosted.

The ecosystem has grown well beyond the core library

What started as a single Python package has expanded into a full platform. The current product lineup includes:

  • Browser Harness , a self-healing harness that enables LLMs to complete any task, now spun out as its own project at browser-harness.com
  • Browser Use Cloud , a hosted platform offering their SOTA model, stealth browsers, CAPTCHA solving, and managed infrastructure
  • Browser Use Box , a 24/7 Claude Code agent for Playwright-style browser automation with Browser Use Cloud, Telegram, and a real browser on any box you own
  • Browser Use CLI , a command-line interface for coding agents like Claude Code and Codex
  • Proxies , residential IPs across 195+ countries for stealth automation

The CLI deserves a closer look. Version 2.0 shipped with a notable architectural change: it's billed as the fastest browser automation for AI coding agents, running 2x faster and using 50% fewer tokens, working with any CLI agent including Claude Code, Codex, and more. It achieves this by bypassing Playwright entirely and talking directly to Chrome via CDP (Chrome DevTools Protocol), with a background daemon keeping the browser alive between commands for roughly 50ms latency per command.

Where it struggles

The library's main friction points are well-documented in the codebase and community. Chrome can consume a lot of memory, and running many agents in parallel can be tricky to manage. For production use cases, the Browser Use Cloud API handles the heavy lifting. Anti-bot detection is another real wall: Amazon's anti-bot system can detect and stop AI automation, showing a CAPTCHA or an error page, which effectively halts the agent. The cloud tier addresses this with stealth browsers and proxies, but the open-source version requires extra configuration to handle it.

A "totally new version" is coming

The tweet that kicked off this coverage is short: "100k special will be a totally new browser-use version." No architecture diagram, no changelog preview, no release date. But reading the recent commit history gives some clues about what direction the team is heading.

The most telling signal is from browser_use.rust import Agent, which already appears in the current README. An experimental Rust agent is available now via Browser Use Terminal, installable with a single curl command. This isn't a side project , it's a preview of where the core is going.

The broader industry context makes this move obvious. In Q1 2026, a pattern emerged across GitHub's fastest-growing AI agent repos: they're written in Rust. Not Python. Not TypeScript. Rust. The reason is straightforward: when an agent has root access and runs autonomously, memory safety isn't optional. Rust gives you process isolation, predictable memory layout, and compile-time guarantees that Python's GIL never will.

Token efficiency is the other forcing function. A typical web page is 15,000 to 30,000 tokens of raw HTML. The part your LLM actually cares about is maybe 2,000 to 3,000. Every unnecessary token costs money and slows the agent down. The CLI's move to direct CDP access already cut token usage by 50% , a full rewrite would likely push that further.

What the rewrite probably means for you

If you're using Browser Use today, the Python API isn't going away overnight. The team has been careful to maintain backward compatibility across the 0.11 and 0.12 release series. But the "totally new version" framing suggests this isn't just a performance patch , it's a rethinking of the architecture.

Here's what to watch for when it drops:

  • Rust core, Python bindings , the most likely shape, following the pattern of libraries like Pydantic v2 and Polars
  • Lower token consumption , the CLI already proved 50% reduction is achievable; a native rewrite could go further
  • Better parallelism , Python's GIL makes running dozens of browser agents simultaneously painful; Rust removes that constraint entirely
  • Tighter memory footprint , relevant for anyone running agents at scale in cloud environments

The cloud platform already advertises new browser infrastructure at 3x cheaper pricing ($0.02/hr), sub-second cold starts, and the most stealth browsers for AI agents. The open-source rewrite is likely the foundation that makes those numbers possible at the library level too.

The 100k milestone is a number. The rewrite is the actual story. Browser Use built its reputation on making web automation feel trivially easy for AI agents. The question the new version has to answer is whether it can make that same ease work reliably at production scale , without the memory leaks, token bloat, and anti-bot failures that currently push teams toward the paid cloud tier.

Trending
  • No trending articles

Comments

avatar

Next Reads