Reachy Mini: the $300 open source robot you can actually hack — Andres Marafioti, Hugging Face

TL;DR

• 1

  The big bet is that voice, not keyboards, will define everyday robot interaction — Marafioti argues voice AI is already mature enough thanks to tools like GPT realtime, open models such as Voxtral and Kokoro, and Hugging Face’s own speech-to-speech stack, but almost nobody is building open robot experiences around it.

• 2

  Reachy Mini is deliberately cheap, weird, and hackable — the robot comes in $300 and $450 versions, ships unassembled so owners can repair everything themselves, and is designed to avoid the "human replacement" trap of humanoids by looking expressive but non-human.

• 3

  People are already personalizing the robot in unexpected ways — users have turned it into a Halloween pumpkin with 3D-printed parts and discovered you can pet it and make it react or even purr, which is exactly the kind of emergent behavior Hugging Face wants the community to explore.

• 4

  Hugging Face’s robot fleet is already large enough to expose real serving problems — with 7,500 units shipped, Reachy Mini has become the company’s most-used app, forcing them to build a production stack with load balancing, separate LLM endpoints, and robot-side echo cancellation, face tracking, and tool dispatch.

• 5

  Marafioti spent two weeks making Coqui 3 TTS actually fast enough for real-time agents — by adding streaming, replacing a dynamic KV cache with a static one, and using CUDA graph capture, he says they pushed performance from 0.8x real time to roughly 5.8x and cut first audio latency to under 200 ms.

• 6

  The endgame is communal robot software, not a closed platform — Marafioti repeatedly frames Reachy Mini as a way for high schools, universities, researchers, and even non-coders using vibe coding to shape how people interact with robots before that design space gets captured by a few companies.