Hugging Face's Reachy Mini, a limbless desktop robot for human interaction experiments, now features a guide for implementing expressive, local conversational AI with head movements and antenna wiggles. The system aims for natural, low-latency responses with the ability to interrupt, running entirely on local hardware if desired, though it can also use remote services or work with a desktop machine. The software stack consists of VAD (voice activity detection) → STT (speech-to-text) → LLM (large language model) → TTS (text-to-speech), allowing users to tweak or swap components independently. For example, a frontier AI model via remote API can be used for the LLM while keeping everything else local. The example configuration uses effective, modest local models, including Qwen3-4B-Instruct for the LLM and even smaller models for the rest. Since Reachy Mini launched as a kit last year, Hugging Face has built an impressive software suite and infrastructure for sharing applications, and a simulator is available to demonstrate its capabilities.