Agent Bulletproof Browser
100% stable headless browser kit for AI Agents(Linux, macOS)
Start new thread
Vaibhav Sharma

Agent Bulletproof Browser - 100% stable headless browser kit for AI Agents(Linux, macOS)

by•
Agent Bulletproof Browser is an enterprise-grade headless browser architecture for AI agents.Stop bolting agents to brittle Playwright scripts. V2 is now live! • Cross-platform: Native macOS & Linux. Zero zombie processes. • Dual-Profiles: Sync auth without hijacking active logins. • Orchestrators: JSON logging & concurrent task queuing. • Stealth: Dynamic bot cloaking to bypass WAFs. Give your agent a stable OS-level primitive and let it work.

Add a comment

Replies

Best
Vaibhav Sharma
Maker
📌
Hey Product Hunt! 👋 I’m Vaibhav, Founder at Kubar Labs. I recently asked X for a recommendation: I needed a CLI-based headless browser designed specifically to work with AI agents. I was tired of existing solutions treating agent integration as an afterthought. The market had no good answer. Every solution was either a heavy testing framework repurposed poorly (like Playwright), or a raw protocol wrapper that broke at the slightest breeze. When we tried to automate workflows in the background, our agents either lost access to their cookies, hallucinated CSS selectors on React sites, got blocked by Cloudflare, or left zombie Chromium processes hoarding our RAM. So, we built it ourselves using bare-metal Linux primitives. Introducing the Agent Bulletproof Browser Automation Kit. It is an enterprise-grade bash wrapper and AI Skill that completely insulates your agents from browser automation hell. ✨ What makes it bulletproof? 1. The "Chain of Death" (prctl): Biologically tethers the Chromium process to the agent's execution script. If the agent gets OOM-killed or hard aborted, the OS instantly vaporizes the browser. Zero orphaned memory leaks. 2. Safe Concurrency (flock): Queues multiple subagents so they don't crash the profile database or steal tab focus from each other. Bot Cloaking: Native --headless=new and User-Agent spoofing completely bypasses WAFs and navigator.webdriver footprinting. 3. Vision-First Policy: The included SKILL.md teaches your agent (Hermes, OpenClaw, Claude) to stop hallucinating DOM selectors and use coordinate-based clicking via screenshots instead. It’s completely open-source. If you use Hermes or OpenClaw, you don't even need to use the terminal, just paste the prompt from our README and the agent will install it autonomously. I’d love to hear your thoughts. What’s the most frustrating part of browser automation for your agents today? Drop your questions below, I’ll be hanging out here all day answering them! 👇
Vaibhav Sharma
Maker

V2 is here: Native macOS support, Dual-Profiles, and JSON logging


We heard the feedback from the community and completely overhauled the underlying architecture. Agent Bulletproof Browser now natively supports macOS.

If you are running multi-agent orchestrators, you no longer have to worry about background tasks hijacking your active login sessions. We split the architecture into dual profiles(foreground/background) and added a secure cookie-sync utility. We also added structured JSON logging so your code can natively subscribe to task lifecycles and errors instead of parsing fragile text outputs.

Check out the updated repo and let us know what you think!

Thami Benjelloun

This is a real pain for agent builders. A lot of “browser agents” still break because the browser layer is too fragile: auth sessions, concurrent tasks, zombie processes, random Playwright failures.

I like the idea of treating the browser as a stable OS-level primitive for agents.

What’s been the hardest part to make reliable: auth/session isolation, concurrency, or anti-bot detection?