Xiaozhi Server

Xiaozhi Server is an open backend for Xiaozhi ESP32 voice devices, with dialogue, IoT, MCP and vision.

Xiaozhi Server (xiaozhi-esp32-server) is the open backend for the xiaozhi-esp32 hardware project, written in Python, Java and Vue and following the Xiaozhi communication protocol. It supports MCP endpoints, voiceprint recognition, vision, OTA and a console in the full install. Two Docker or source layouts are documented: a small file-based install, and a full module install that needs a database. A free-tier setup can run FunASR and ChatGLM locally; a streaming setup uses cloud ASR and LLMs and, from version 0.5.2, is about 2.5 seconds faster. The authors warn it is unfinished, not security-reviewed, and not for production. It comes from Professor Siyuan Liu's group at South China University of Technology.

Xiaozhi Server at a glance

Best for: Software engineer
Runs on: open source, self-hosted
Pricing: Free and open source; optional cloud ASR and LLM APIs are billed by their providers
Released: February 2025
Category: AI agent tool
Website: xiaozhi.biz

Key features

Backend for Xiaozhi ESP32 voice hardware
MCP, IoT, vision and optional voiceprint recognition
Minimal file-based install or a full console with a database
Local FunASR plus a free ChatGLM path
Streaming ASR and LLM path from v0.5.2
Open protocol docs and Docker images

Use it for

Point a Xiaozhi ESP32 board at a home server
Add MCP tools to a hardware voice agent
Try the free local ASR path before paying for APIs

Alternatives to Xiaozhi Server

py-xiaozhi — Connects LLMs to physical hardware and local IoT devices via async architecture.
Esp32 Thinks — Runs your own AI stack on low-cost hardware without vendor lock-in.
Zongzhu Bot — Connects an ESP32 display board to a Grok Bot for voice interactions.
Xiryn — Runs locally in Docker containers while keeping all data private on the user disk.
OpenHarness — Manages Claude Code, Codex, and Hermes from a single command center.
Home Assistant MCP — Connects AI assistants to Home Assistant using natural language.
MCP server — Connects directly to AI assistants without requiring an account, API key, or authentication.
Octop — Self-hosted multi-user assistant where data stays on own servers
Esp32 Iphone Agent — It uses standard AirPlay and a USB mouse bridge, requiring no Mac, jailbreak, or developer account.
Wx Bot — It prevents message loss with local SQLite storage and automatic redelivery.
Helper — Provides a Python plugin framework for WeChat text messaging and bot features on Windows.
Ming Voice — It runs fully local with low latency on your own GPU without sending audio elsewhere.

All Xiaozhi Server alternatives