Hugging Face Unveils Agentic Toolkit for Reachy Mini

The compact Reachy Mini is designed for accessibility and engagement. | Source: Hugging Face

This week, Hugging Face introduced an agentic toolkit that enables anyone to create a functional application for the Reachy Mini, its open-source desktop robot. The toolkit can be set up in under an hour without requiring any coding. Instead, an AI agent handles all the code generation.

To utilize the toolkit, users simply describe the desired robot behavior in plain English. The AI agent then writes, tests, and deploys the code directly to the robot.

“For six decades, robots were exclusively built by roboticists. Starting today, anyone can build them,” stated Clément Delangue, co-founder and CEO of Hugging Face. “When the software is open-source and an AI agent can write the code, the traditional barrier of technical expertise simply vanishes.”

Throughout the history of robotics, three major hurdles have stood between an idea and a functional robot: specialized expertise, costly hardware, and lengthy integration processes. Hugging Face aims to eliminate all three. An AI agent replaces the need for expertise. The hardware is an affordable, open-source desktop robot available for purchase. Integration is streamlined into a one-click workflow on a platform already used by millions of developers.

Hugging Face is an open platform for AI developers, often referred to as “the GitHub of AI.” Millions of developers and tens of thousands of companies use it to share AI models, datasets, and applications. In 2024, Hugging Face acquired Pollen Robotics, the French creator of the open-source Reachy robot line.

Meet one of Hugging Face’s early uses

Joel Cohen is a 78-year-old retired marketing executive in the Raleigh-Durham area who facilitates CEO peer groups. He is not a developer and has no background in robotics. Cohen is also colorblind and wears hearing aids. According to Hugging Face, it took him over two weeks to assemble his Reachy Mini Lite (a process that typically takes about three hours).

Despite this, Cohen successfully built an application. He created a voice-controlled AI co-facilitator for the CEO peer groups he conducts on Zoom. The Reachy Mini sits on his desk. The robot stands 11 inches (27.9 cm) tall, is 6.3 inches (16 cm) wide, and weighs just 3.3 pounds (1.5 kg).

When Cohen says, “Hey, Reachy,” the robot wakes up, listens, and responds. It has a distinct personality, which Cohen describes as his “VP of future thinking.” The system features four facilitation modes, a library of over 60 questions, and greets each of his 29 group members by name.

During sessions, it can call on a member, challenge a superficial answer, generate a new question on the fly, or summarize key discussion points before concluding.

“I built this by simply describing what I needed in plain English. Claude wrote all the code,” said Cohen. “No SDK required. No robotics knowledge. No prior development experience.”

App store for Reachy Mini is now open

Apps for the Reachy Mini are hosted on the Hugging Face Hub, which is searchable, forkable, and allows one-click installation. If you find an app you like, you can duplicate it, ask the AI agent to modify it, and publish your own version. The company states that creating a new app takes just minutes.

Furthermore, every Hugging Face app can also run in a browser-based simulator, allowing anyone to explore the app catalog without needing the physical hardware.

The Reachy Mini App Store, featuring over 200 apps, is now live. Users can build their own apps by instructing a machine learning intern or their preferred AI agent. Some featured apps include: