Description
A Xiaozhi device on your desk can chat, but to make it actually do work, like turning this week of notes into a checklist, something has to sit in the middle. xiaozhi-hub-server is that piece: the device listens and speaks, the hub handles recognition and speech synthesis, and then hands the request to whichever AI you called by name. Say "Claude" and it goes to Claude Code on your machine, say "DeepSeek" and it goes to DeepSeek.
Long jobs do not leave you waiting in silence: it keeps you company, reports progress now and then and speaks the result when done; tap the screen or say the wake word to interrupt. Run it on your own computer to use your local Claude Code, or on a cloud server with model APIs, all managed from a web console.
Switch AI by name: Names are set in the console; each AI keeps its own thread and the device remembers the last one used
Claude Code at work: Uses the account you are already logged in with locally, no separate API setup
Many models: Claude, OpenAI, Gemini, DeepSeek, Qwen, Doubao, GLM, Kimi and any OpenAI-compatible endpoint
Grok Bot: Hand tasks to a cloud Grok Bot; the device reports back when a long job finishes
Waiting and barge-in: Progress updates on multi-minute tasks, auto-resume after a disconnect, and "continue last time" on request
Accurate recognition: Offline X-ASR by default handles mixed Chinese and English, with hotwords and a correction table
Web console: Configure models, names, recognition and synthesis in the browser, with a one-click installer
Long jobs do not leave you waiting in silence: it keeps you company, reports progress now and then and speaks the result when done; tap the screen or say the wake word to interrupt. Run it on your own computer to use your local Claude Code, or on a cloud server with model APIs, all managed from a web console.
Features
Switch AI by name: Names are set in the console; each AI keeps its own thread and the device remembers the last one used
Claude Code at work: Uses the account you are already logged in with locally, no separate API setup
Many models: Claude, OpenAI, Gemini, DeepSeek, Qwen, Doubao, GLM, Kimi and any OpenAI-compatible endpoint
Grok Bot: Hand tasks to a cloud Grok Bot; the device reports back when a long job finishes
Waiting and barge-in: Progress updates on multi-minute tasks, auto-resume after a disconnect, and "continue last time" on request
Accurate recognition: Offline X-ASR by default handles mixed Chinese and English, with hotwords and a correction table
Web console: Configure models, names, recognition and synthesis in the browser, with a one-click installer
