Meet Rover, your site's new hands. Rover lives inside your website, and takes actions for your users. It onboards users, runs workflows, fills forms, and converts visitors through conversation. Your user says "help me checkout", Rover fills the fields, clicks the buttons, and finishes the purchase. Think Stripe for AI agents: embed script tag, your site is agentic.
I'm Arjun, co-founder and CEO of rtrvr.ai. I pioneered Vertical Federated Learning at Google and left to build a vision for the agentic web. We've spent the last year building the leading DOM/text-only web agents, and today we're bringing that technology directly into your website with Rover.
Every website has a chat widget. They all do the same thing: answer questions and link you somewhere else. They are just glorified FAQ answerers and leave users frustrated when trying to navigate complex workflows.
Rover actually completes the task. It clicks buttons, fills forms, navigates pages, and finishes checkout flows, all through a conversational interface embedded on your site.
Our script tag is a complete agentic harness that can propagate types/clicks/selects and other interactions to complete tasks on your site for your users.
We already power 1.5 MM+ web automation workflows with SOTA performance of 81.4% task completion on WebBench. Rover brings that same engine to your website.
We built this because the agentic web shouldn't mean handing your users to the mercy of Google's Chrome agent that can for example redirect users to your competitors. Your website should have its own agent, engaging and serving users on your own turf.
The "glorified FAQ answerer" framing is exactly right — the shift from answering to completing is what matters for multi-step workflows. Drop-off compounds when a user has to take over from the bot mid-flow. Curious what the failure mode looks like at the 18.6% gap from WebBench: does Rover roll back cleanly, or does it hand off to the user with context on where it stopped?
This feels like the end of ‘book a demo to see it’ - instant value is the new default.
Super cool use case!! Curious, how does Rover handle authentication-gated workflows (e.g., when a user needs to log in before completing a purchase)? Does it store session context or does each action start fresh?
Congrats, strong vision. Сan site owners control which actions it’s allowed to perform?
I love that the agent can be embedded as a script tag, and that's literally all you need to onboard customers - little to no friction to start using Rover right away! Another use case is navigating complex (and often ill-designed) government websites, for example finding POCs for specific grants on highergov and adding them on LinkedIn with a personalized message. Very excited for the launch and excited to continue to use rover!
Congrats on the launch!, AI web agents are becoming essential, and Rover looks like a very smooth implementation.
Quick tip from a growth perspective: Most people market “AI Browsing” as a cool feature, but the real pain you're solving is Manual Search Exhaustion. Users don't want to “browse with AI”; they want to get the answer without the digging.
If you pivot your messaging from “AI-powered browsing” to “The End of Manual Search”, you move from being a browser extension to a cognitive shortcut. I would send a couple of specific copy tweaks to you if you want it, to help you nail that “Efficiency” angle. Good luck today!
"Think Stripe for AI agents" is a sharp analogy and it lands well. One script tag to make your site agentic is the kind of primitive that could quietly become infrastructure for a lot of products.
The checkout completion use case is compelling because it hits a real pain point. Curious how Rover handles sites with heavy dynamic rendering or multi-step auth flows. Those are usually where browser agents struggle most.
Building AI-powered workflows myself and the embed-first distribution model is smart. Developers adopt it where they already are, no new tool to learn. Congrats on #3 today, well deserved! 🚀
Finally, a chat widget that actually does stuff instead of just linking me around! 🚀
This is the direction everything is heading! The gap between AI that talks and AI that actually does is where the value is. How Does Rover handle edge cases mid-workflow, like when a form has conditional logic or a checkout has an unexpected step? Impressive execution on the embed simplicity though, the Stripe analogy is spot on
Grats on launching!! Which industry would benefit the most using Rover according to you? Maybe Ecom?
Congrats. Help me understand, is this designed to replace chatbots and it takes over the function of a chatbot as well as an agent for the website?
Hey PH!
I'm Bhavani, CTO and Co-founder of rtrvr.ai. Ex-Google, ex-Adobe. We spent 2 years building DOM-native web intelligence before shipping Rover.
The Problem
Every website has a chat widget. They all do the same thing... answer questions and link you somewhere else. When a user says "help me checkout," the chatbot says "here's the link to our cart page" and wishes them luck. The user still has to click, navigate, and fill everything themselves.
What Rover Does Differently
Rover lives inside your website and takes real actions for your users. It reads your live DOM, builds a semantic action tree, and executes with native browser precision. Sub-second. First-party. One script tag.
Why Existing Solutions Fall Short
RAG chatbots (Intercom, Chatbase, Drift): Answer questions, spit links, take zero actions. Users are left to figure it out themselves.
Screenshot agents (Operator, Computer Use): Take a picture of your page, guess where to click. 2-5 seconds per action. Run in a remote browser VM. Can't be embedded on your site.
WebMCP: You expose your internal APIs to Google so their agent can act on your behalf. You build it, you maintain it, Google owns the user.
Rover: Reads the live DOM, not screenshots. No APIs to expose, no knowledge base to maintain. Just one script tag and your site is agentic.
Key Features
✅ Checkout automation: User says "help me checkout," Rover fills the fields, clicks the buttons, completes the purchase.
✅ Guided product tours: Walks users through every feature, step by step, through conversation.
✅ Smart form filling: Forms filled from natural language. Zero manual input.
✅ Universal embed: One script tag. Works on any website. No backend changes needed.
Who It's For
SaaS companies that want to onboard users without hand-holding.
E-commerce sites that want higher checkout completion.
Any website that wants to convert visitors through action, not just conversation.
The Numbers
81.39% task success rate on WebBench, the highest in web automation, ahead of every vision-based agent we've tested against. 22,000+ users. 1.6M+ workflows completed. 2,000+ websites already live.
Hey Product Hunt! 👋
I'm Arjun, co-founder and CEO of rtrvr.ai. I pioneered Vertical Federated Learning at Google and left to build a vision for the agentic web. We've spent the last year building the leading DOM/text-only web agents, and today we're bringing that technology directly into your website with Rover.
Every website has a chat widget. They all do the same thing: answer questions and link you somewhere else. They are just glorified FAQ answerers and leave users frustrated when trying to navigate complex workflows.
Rover actually completes the task. It clicks buttons, fills forms, navigates pages, and finishes checkout flows, all through a conversational interface embedded on your site.
Our script tag is a complete agentic harness that can propagate types/clicks/selects and other interactions to complete tasks on your site for your users.
We already power 1.5 MM+ web automation workflows with SOTA performance of 81.4% task completion on WebBench. Rover brings that same engine to your website.
We built this because the agentic web shouldn't mean handing your users to the mercy of Google's Chrome agent that can for example redirect users to your competitors. Your website should have its own agent, engaging and serving users on your own turf.
Try it → rover.rtrvr.ai
🧡 rtrvr.ai Team