Reachy Mini: the $300 open source robot you can actually hack
summary
Reachy Mini is a $300 open source robot shipped to 7,500 users, arrives unassembled, and runs a voice-first stack that actually works: Parakeet transcribes every 150 ms with partial results, Qwen 3.5 27B handles LLM inference, and Coqui 3 TTS is optimized from 0.8x to 5.8x real time by switching to a static KV cache and using CUDA graph captures, bringing time to first audio under 200 ms; infrastructure separates LLM endpoints from conversation nodes to match user load.
AI-written summary · about
chapters (11)
- 0:00Reachy Mini intro
- 1:39Robotics problems today
- 4:27Vision for Reachy Mini
- 5:57Design and features
- 7:38Voice AI future
- 8:22Live demo conversation
- 10:30Voice system architecture
- 12:39Optimizing Coqui TTS
- 15:25Faster Coqui TTS demo
- 17:10Call to action
- 18:35Q&A
related talks
16mServing Voice AI at $1/hr: Open-source, LoRAs, Latency, Load Balancing
47mOwn Your LLM: Running Generative and Agentic AI Locally
37mEmpowering Developers with local AI
34mVoice-Controlled Lights on a Raspberry Pi
33mDesigning Voice Agents for Real Conversations
43mLet’s build K.I.T.T. with JavaScript
36mPanel Discussion
48mThe next wave of AI: Open source, robotics & the future of India’s tech powerhouse