The Robot App Store Is Real. The Customers Are Still Figuring Out What They Want.

Joel Cohen spent a few days building his first robot. He's 78. He'd never coded, never worked in robotics, never assembled anything more complicated than IKEA furniture. His creation sits on his desk in Raleigh-Durham, North Carolina, listens for the phrase "Hey Reachy," and responds in the voice of his self-described "VP of Future Thinking." It can greet 29 CEO peers by name, push back on surface-level answers during meetings, and summarize key themes before a session closes. Cohen built it by describing what he wanted in plain English. An AI wrote the code. No robotics background required.

Cohen is not alone. More than 150 creators have contributed to the Reachy Mini App Store, most of whom had never written a line of robotics code before, according to Hugging Face's own blog. Clement Delangue, the Hugging Face CEO, built an office receptionist app in under two hours — face recognition to detect arrivals, a message to alert him who walked in and why. "For me, it would have been impossible," he told VentureBeat. "If you weren't a robotics developer, it probably would have been impossible, or it would have taken a few months." Other community-built apps include a chess opponent that mocks your blunders, a language tutor that corrects your accent, and an F1 race commentator for your desk. None of these are enterprise software. All of them are real.

What Hugging Face built — and what Cohen represents — is the robot app store. The company that hosts open-source AI models launched a marketplace for Reachy Mini, its $299 open-source desktop robot, where users can browse, install, and build apps in the same way smartphone owners do. It's the same bet Apple made with the iPhone in 2008, the same bet Google made with Android: remove the gatekeeping, let anyone build, see what happens.

"For sixty years, robots were built by roboticists," wrote Delangue. "When the hardware is open-source, the software is open-source, and an AI agent can write the code, the gating that used to come from technical knowledge mostly disappears."

Hugging Face acquired Pollen Robotics, the French maker of the Reachy line, in April 2025. The App Store now holds over 200 community-built applications. Users build apps by describing what they want in plain English; an AI handles the code, tests, and shipping. The platform is designed to be model-agnostic, supporting a range of leading AI engines — meaning Hugging Face isn't locked into any single AI provider and can take a cut if and when app creators start charging. Real-time interaction draws on OpenAI Realtime and Gemini Live, according to Hugging Face's own blog.

The install base tells a story in numbers. Nearly 10,000 Reachy Mini units have shipped since launch, according to Hugging Face's own blog. Another 3,000 went out last week alone. More than 1,000 will go out in the next 30 days. By robotics industry standards — where a single industrial arm can cost $50,000 and a humanoid quadruped runs $70,000 or more — 10,000 units in under a year with a third of that coming in the last seven days is a real footprint with momentum.

The pricing helps. Reachy Mini Lite starts at $299, a tethered version that runs off an external computer. The wireless standalone version is $449, with a Raspberry Pi compute module and Wi-Fi built in. That's hobbyist territory, not enterprise procurement territory. Compare that to Boston Dynamics' Spot, which retails for around $70,000, or Chinese competitors starting above $1,900.

The app store lives on the Hugging Face Hub alongside millions of existing AI projects. Every app is a repo — searchable, forkable, one-click installable to a physical robot or a browser-based simulator. No hardware required to browse or build. The whole stack is open-source: the robot schematics, the software, the apps themselves. Even the AI agent traces that generate the code are public. "We could have built this as a closed app store with a 30% cut," Delangue wrote. "We didn't, and we won't."

The honest gap is this: whether the 200 apps represent the beginning of a platform ecosystem or the ceiling of a hobbyist curiosity market is genuinely unknown. Delangue himself built a receptionist for his own office. Cohen built a meeting facilitator for his peer groups. The source material does not yet contain a factory task, a call center replacement, or a medical use case. The apps are real. Their professional utility remains an open question. There is no monetization yet, and no creator has publicly disclosed a paid deployment at scale.

What to watch: whether any professional developer shows up once monetization arrives, whether any app crosses from desk curiosity into task replacement, and whether the model-agnostic platform can hold together as the robotics-specific AI layer matures. The robot app store is open for business. The customers are still figuring out what they want.