Rabbit R1 gets LAM playground for custom web agents

When the $199 Rabbit R1 launched earlier this year, it promised the pocket-sized AI dream: an assistant that didn't just chat, but actually operated the web for you. But early reviewers quickly realized the gadget was wildly underbaked. Battery life was disappointing, and its signature Large Action Model (LAM) only worked with Uber, Spotify, and DoorDash, failing more than half the time.

Now, the startup is finally giving early adopters the feature they actually paid for.

Teach your pocket AI new tricks

Rabbit just introduced LAM playground, a cross-platform agent environment hosted inside its cloud hub, Rabbithole. The new system lets users teach their R1 agents how to navigate and interact with standard websites across the open web.

Instead of being trapped inside a handful of pre-programmed services, the playground allows users to build workflows for destinations like Google, Walmart, and YouTube. Rabbit pitched the release as a shared testing ground where people can experiment, share feedback, and refine how their agents handle real-world sites.

Ahead of the rollout, Rabbit CEO Jesse Lyu demonstrated the tool to ZDNet with a simple prompt: find a six-pack of Diet Coke and add it to an Amazon cart. The multimodal agent scanned the page elements on Google, clicked through to Amazon, and placed the item in the cart. The entire process took about 45 seconds, which is noticeably sluggish, but it proved the agent can complete multi-step tasks end-to-end without manual intervention.

The CAPTCHA wall and desktop ambitions

Of course, setting an AI loose on the live internet immediately brings friction.

In a second test asking the agent to locate an iPhone 16 deal at Walmart, the AI ran straight into a CAPTCHA verification and failed. Lyu explained that the demo wasn't running on Rabbit's dedicated IP cluster, but it highlights a massive hurdle. Everyday web browsing is packed with bot-detection screens, and human intervention ruins the magic of autonomous agents.

Lyu also teased where Rabbit wants to take the tech next: operating beyond browser tabs at the desktop and application level. The roadmap includes running isolated Linux instances and manipulating native programs like Adobe Photoshop to edit images on the fly.

Why it matters

If you have an R1 gathering dust in a drawer, this update makes the device worth picking up again, if only to see how web agents work in real life. But don't expect it to replace your phone just yet. A 45-second checkout flow and inevitable bot-check roadblocks mean it remains an intriguing experiment rather than a daily utility.

Strategically, Rabbit made a smart pivot by branding this as a playground. By framing it as an experimental lab, they manage expectations while outsourcing the messy work of web navigation testing to enthusiastic users. It shifts R1 away from being just another quirky hardware gimmick toward a serious cross-platform agent platform.

However, moving toward desktop control and native app manipulation raises major security and privacy red flags. If Rabbit wants mainstream trust, letting cloud agents browse user accounts and edit files requires bulletproof guardrails. For now, Rabbit has taken a genuine step toward fulfilling its CES promise, but navigating the messy, bot-hostile open web is going to be an uphill battle.