Market Trends
Hi HN, I forked chromium and built agent-browser-protocol (ABP) after noticing that most browser-agent failures aren’t really about the model misunderstanding the page. Instead, the problem is that the model is reasoning from a stale state.ABP is designed to keep the acting agent synchronized with the browser at every step. After each action (click, type, etc), it freezes JavaScript execution and rendering, then captures the resulting state. It also compiles the notable events that occurred during that action loop, such as navigation, file pickers, permission prompts, alerts, and downloads, and sends that along with a screenshot of the frozen page state back to the agent.The result is that browser interaction starts to feel more like a multimodal chat loop. The agent takes an action, gets back a fresh visual state and a structured summary of what happened, then decides what to do next from there. That fits much better with how LLMs already work.A few common browser-use failures ABP helps eliminate:
* A modal appears after the last Playwright screenshot and blocks the input the agent was about to use
* Dynamic filters cause the page to reflow between steps
* An autocomplete dropdown opens and covers the element the agent intended to click
* alert() / confirm() interrupts the flow
* Downloads are triggered, but the agent has no reliable way to know when they’ve completedAs proof, ABP with opus 4.6 as the driver scores 90.5% on the Online Mind2Web benchmark. I think modern LLMs already understand websites, they just need a better tool to interact with them. Happy to answer questions about the architecture, forking chrome or anything else in the comments below.Try it out: `claude mcp add browser -- npx -y agent-browser-protocol --mcp` (Codex/OpenCode instructions in the docs)Demo video: https://www.loom.com/share/387f6349196f417d8b4b16a5452c3369
Show HN: Open-source browser for AI agents
A specialized browser protocol designed to eliminate 'stale state' failures in AI agent-browser interactions, making the process feel like a 'multimodal chat loop' and providing a 'better tool' for LLMs to interact with websites reliably.
View Origin LinkProduct Positioning & Context
AI Executive Synthesis
A specialized browser protocol designed to eliminate 'stale state' failures in AI agent-browser interactions, making the process feel like a 'multimodal chat loop' and providing a 'better tool' for LLMs to interact with websites reliably.
The agent-browser-protocol (ABP) directly tackles a fundamental reliability challenge in AI agent development: the problem of agents reasoning from stale browser states. By forking Chromium and implementing a mechanism to freeze JavaScript execution and rendering after every agent action, ABP ensures a real-time, synchronized feedback loop. This 'multimodal chat loop' provides agents with a fresh visual state and a structured summary of critical events (navigations, modals, downloads), aligning perfectly with how LLMs process information and make sequential decisions. This approach fundamentally improves upon traditional browser automation, which often fails to account for the dynamic, asynchronous nature of modern web applications in a way that's consumable for an LLM.
Developers building AI agents will find ABP invaluable. It eliminates common, frustrating failure points like unexpected modals, dynamic page reflows, or unhandled alerts/downloads, which typically require complex, brittle workarounds. This leads to significantly more robust, predictable, and easier-to-debug agents, accelerating development cycles and increasing the success rate of automated tasks. The impressive 90.5% score on the Online Mind2Web benchmark serves as strong validation of its practical effectiveness, offering a compelling reason for adoption over existing, less agent-aware solutions.
From a market perspective, ABP is a critical enabler for the burgeoning 'agentic AI' trend. As businesses increasingly deploy AI agents for complex B2B SaaS interactions, data extraction, and automated workflows, the demand for specialized, reliable browser infrastructure will intensify. ABP represents a shift towards purpose-built tools that bridge the gap between LLM capabilities and the complexities of the web, moving beyond general-purpose automation. This innovation could unlock new levels of autonomy and reliability for web-based AI agents, fostering broader adoption and transforming how enterprises leverage AI for digital operations.
Developers building AI agents will find ABP invaluable. It eliminates common, frustrating failure points like unexpected modals, dynamic page reflows, or unhandled alerts/downloads, which typically require complex, brittle workarounds. This leads to significantly more robust, predictable, and easier-to-debug agents, accelerating development cycles and increasing the success rate of automated tasks. The impressive 90.5% score on the Online Mind2Web benchmark serves as strong validation of its practical effectiveness, offering a compelling reason for adoption over existing, less agent-aware solutions.
From a market perspective, ABP is a critical enabler for the burgeoning 'agentic AI' trend. As businesses increasingly deploy AI agents for complex B2B SaaS interactions, data extraction, and automated workflows, the demand for specialized, reliable browser infrastructure will intensify. ABP represents a shift towards purpose-built tools that bridge the gap between LLM capabilities and the complexities of the web, moving beyond general-purpose automation. This innovation could unlock new levels of autonomy and reliability for web-based AI agents, fostering broader adoption and transforming how enterprises leverage AI for digital operations.
Community Voice & Feedback
These comments are curious. Any number are paraphrasing the same sort of LLMing compliment ("the problem is real" and freezing browser is the right framing), then there are commenters with 10 year old accounts and no comments until this topic cluster last couple days, aside from the dozen greens.Meanwhile OP theredsix comes off like the only other human here besides Retr0id...
Really impressive work! The deterministic “freeze then capture” approach highlights how much complexity happens when the system state isn’t guaranteed.In identity systems like Gnobu, we face a similar challenge: ensuring that authentication flows remain consistent across multiple services and sessions, especially in environments with multiple asynchronous actions.Curious if you’ve considered adding deterministic checkpoints or logging hooks that could integrate with external identity systems for agent-level session management?
Freezing JS execution between actions is the kind of obvious idea that nobody did properly untill now. Kudos for actually forking Chromium instead of hacking around Playwright like everybody else.But here's my thought: you're solving the "stale state" problem by making the browser deterministic. Real websites aren't deterministic. WebSOcket pushes, long-polling, background fetches, animations that don't finish — freezing execution doesn't pause the server. The moment you unfreeze, the world may have moved.90.5% on Mind2Web is great. But Mind2Web tasks are mostly "fill a form, click submit." The brutal failures happen on SPAs with optimistic UI updates, where the DOM says "saved" but the network request hasn't finished. Does ABP handle that case, or does the freeze just delay the confusion?Genuine question — not trying to tear this down. The architecture is smart. I just wonder if "make the browser simpler for the agent", eventually hit s a wall where you need to make the agent smarter about async instead.
The stale state problem is real and underappreciated. I've been running browser automation through OpenClaw and the failure modes you describe — modal appears after screenshot, dropdown covers the target element — are exactly what causes silent failures that are hard to debug. The agent "succeeds" from its perspective because it acted on the last known state.The freeze-then-capture approach is interesting. Curious how it handles pages with aggressive anti-bot detection that fingerprints headless Chromium forks — that's the other failure mode I keep hitting.
Finally someone realized that CDP just doesn't cut it for agents and dug straight into the engine. Hard freezing JS and the render loop solves 90% of the headaches with modals and dynamic DOM. Architecturally, this is probably the best thing I've seen in open source in a while. The only massive red flag is maintaining the fork - manually merging Chromium updates is an absolute meat grinder
> Pause JavaScript + virtual timeVery cool! Sometimes when I try to debug things with chrome dev tools MCP, Claude would click something and too many things happen then it kind of comes to the wrong conclusions about the state of things, so sounds like this should give it a more accurate slice of time / snapshot of things.
Freezing the browser at every step is a very good approach. I am also working on an agent browser. It uses wireframe snapshots instead of screenshots to reduce token cost.
https://github.com/agent-browser-io/browser
https://github.com/agent-browser-io/browser
From the commit history, it looks like you are using Claude for some of the development. Would love to hear how you are using Claude to go through such a massive code base.btw, impressive project.
> As proof, ABP with opus 4.6 as the driver scores 90.5% on the Online Mind2Web benchmarkAnd what does opus score with "regular" browser harnesses?
Interesting, I wonder if this would help with other projects too, one project that comes to mind is archivebox, I don't know if they still have the issue I'm thinking of, but archivebox eventually had the Chrome instances (as the meme goes) basically consume all available RAM. If by freezing execution this could stop that, it could be useful for more than just AI agents.
Related Early-Stage Discoveries
Discovery Source
Hacker News Aggregated via automated community intelligence tracking.
Tech Stack Dependencies
No direct open-source NPM package mentions detected in the product documentation.
Media Tractions & Mentions
No mainstream media stories specifically mentioning this product name have been intercepted yet.
Deep Research & Science
No direct peer-reviewed scientific literature matched with this product's architecture.