coderzc/open-xiaoai-bridge

What it solves

Open-XiaoAI Bridge breaks the closed ecosystem of Xiaoai smart speakers, allowing them to be bridged to external AI services. It enables users to replace the native AI with more flexible, powerful, or customized LLM-based agents and services.

How it works

The project acts as a server-side bridge between a modified Xiaoai speaker (which requires a custom Rust client installed via firmware flashing) and various AI backends. It captures audio from the speaker, processes it using VAD (Voice Activity Detection), KWS (Keyword Spotting), and optional local ASR (Automatic Speech Recognition), and then routes the request to a configured AI service. The bridge supports multiple interaction modes, including single-turn commands and continuous multi-turn conversations.

Who it’s for

Users who own Xiaoai smart speakers and want to integrate them with advanced AI agents (like OpenClaw or QwenPaw) or OpenAI-compatible APIs to create a personalized, voice-controlled AI assistant.

Highlights

• Multi-Backend Support: Integrates with OpenAI-compatible services (Ollama, LM Studio), OpenClaw, QwenPaw, and Xiaozhi AI.
• Custom Wake Words: Supports multiple custom wake words that can route to different AI services or specific agents.
• Multi-Agent Routing: Allows a single speaker to host multiple isolated AI personalities by dynamically switching agent sessions based on the wake word used.
• Continuous Conversation: Enables multi-turn dialogue without needing to repeatedly say the wake word.
• Remote Control API: Provides an HTTP API to remotely trigger text-to-speech (TTS), play audio URLs/files, or wake the device.
• Hybrid ASR: Supports both local offline ASR (SherpaASR) and the speaker's native ASR.