OpenNextion XiaoZhi Ai is an open‑source voice AI solution that turns your OpenNextion screen into a conversational AI terminal. It integrates with Qwen, DeepSeek, and Doubao for natural dialogues. Offline wake‑up and local recognition ensure low latency and privacy.
| Feature | Description |
|---|---|
| Voice Conversation | Speak to the device and it will respond to you in real time. |
| AI Large Model Driven | Powered by large models such as Qwen and DeepSeek, delivering natural and intelligent responses. |
| Offline Voice Wake‑Up | Wake the device without an internet connection – fast response and privacy protection. |
| Voice or Touch Wake‑Up | Use the wake word "Hey Jarvis" or tap the screen to wake the XiaoZhi AI assistant for voice interaction. |
| Emoji Display | The screen shows emojis and status indicators, making interaction more engaging. |
| Multi‑Language Support | Supports Chinese, English, Japanese, and more. |
| Customization | Theme, character language, voice, AI agent, and extensions can all be modified online via the xiaozhi.me platform. |
Operation Instructions
Note: The ESP32 device provides microphone and speaker interfaces, but these modules are not included and must be obtained separately. You may refer to officially compatible accessories or use other compatible hardware.
Before starting, ensure the USB cable supports data transfer and maintain a stable connection between the board and your computer.
After successful flashing, the board will automatically restart and boot into the Xiaozhi AI.
After the device successfully connects to a 2.4 GHz Wi‑Fi network:
Xiaozhi-E4D5 (open)http://192.168.4.1
Click "Configure" on the model to set the device's theme background color, the Agent's character language, voice, large model, and extension capabilities. The main features depend on the xiaozhi.me platform.
*After making changes, press the REST button on the device to update the configuration.
Device Theme Configuration
.binfile must be generated. You can download this.binfile or flash it online.
Click the "Devices" button on the model to delete associated devices, add new devices, or perform OTA updates.
XiaoZhi AI is an open-source project that turns your OpenNext screen into an AI voice chatbot. It connects to large language models like Qwen and DeepSeek, delivering smarter and more natural responses. It also supports offline wake-up, voiceprint recognition, and other features, making it more intelligent than ordinary voice assistants.
No. The project provides ready-to-flash firmware that doesn't require a development environment – beginners can download and flash it directly. You only need to set up a development environment if you want to customize features or modify the code.
Personal users can register an account and use real‑time models like Qwen for free. If you prefer to deploy your own server, the project documentation also provides third‑party open‑source solutions.
Supported chip platforms include ESP32‑C3, ESP32‑S3, and ESP32‑P4. This project is specifically adapted for the OpenNextion G‑series 2.8‑inch and 3.5‑inch capacitive touch screens.
Yes. Through the MCP protocol, XiaoZhi AI can enable smart home control, PC desktop operations, knowledge search, email sending/receiving, and more. For details, refer to the MCP protocol documentation in the official project.
If you want to move the device to another xiaozhi platform account, you need to delete the device from the originally bound xiaozhi account first. If you cannot reach the original account owner, you can request unbinding via email: send to xiaozhi.ai@tenclass.com with the subject line "Device Unbinding Request, Device ID XXX, MAC Address XXX".
Device ID – If your device is working properly, try asking it verbally during a conversation. MAC Address – You will need to use a serial tool to read the serial output and locate the MAC address information.
No. The device communicates with the xiaozhi server over the internet, not through the local network. Even the initial connection between the device and the xiaozhi platform goes through the cloud, so they do not need to be on the same local network.
No. Offline voice wake‑up uses ESP‑SR technology and is performed entirely locally without an internet connection. However, this only wakes the device – it will not respond to any queries, as all conversations rely on cloud‑based large language models.