Show HN: Multi-modal Muse assistant on an ESP32 (github.com)

A new AI assistant dubbed Multi-modal Muse has been announced, designed to run on an ESP32 microcontroller for about $50. This personal assistant can perform tasks like speech recognition, transcription, and chat using OpenAI-compatible servers, allowing users to choose various models for each function. Users can activate the assistant with a wake word or a button, and it streams voice replies sentence-by-sentence while still generating content, making interactions feel more dynamic and immediate. The assistant supports both English and Chinese, employing different text-to-speech (TTS) models depending on the language used. The significance of this innovation lies in its ability to break free from the vendor lock typical of cloud-based AI assistants. Unlike conventional systems that tether users to a single provider, Multi-modal Muse enables users to customize their AI stack, selecting the best or cheapest options for each task through a simple configuration file. The project, built on Meta's open-source muse-gadget-sdk, retains user autonomy over AI interactions, allowing for transparency and flexibility in how data is processed. With headless provisioning and low latency, this solution opens the door for DIY enthusiasts and developers to create personalized AI experiences with minimal cost and maximum customization.