July 2026
XiaoZhi AI
A self-hosted backend for XiaoZhi ESP32 voice companions: wake word on device, then ASR → LLM → TTS and MCP tools through a server I control. Fork of xiaozhi-esp32-server, paired with the open-source xiaozhi-esp32 firmware (hands-on build logged separately as sihah-esp32).
My running fork of the XiaoZhi ESP32 voice-assistant backend — the server half of the xiaozhi.me stack.
My running fork of the XiaoZhi ESP32 voice-assistant backend — the server half of the xiaozhi.me stack.
My fork and running copy of xiaozhi-esp32-server — the open-source backend that pairs with xiaozhi-esp32 firmware so an ESP32 can be a real voice companion: wake word on device, then streaming ASR → LLM → TTS through a server you control.
Two stacks, don't mix them. This is the XiaoZhi / xiaozhi.me path (firmware + server). The Claude-quota desk duck is a separate path: quack-quack + ohmyclawd. They share a desk and Waveshare-era hardware curiosity; they do not share the quota wire protocol.
What this project is
| Layer | Upstream | Mine |
|---|---|---|
| Firmware | 78/xiaozhi-esp32 | Hands-on build/customisation logged as sihah-esp32 |
| Backend | xinnan-tech/xiaozhi-esp32-server (Python / Java / Vue; MQTT+UDP, WebSocket, MCP, knowledge base) | This repo: amirahnasihah/xiaozhi-esp32-server |
| Product surface | xiaozhi.me | Running and learning the stack end-to-end |
The server exists so the ESP32 stays a thin edge: audio and wake word on-device, heavy ASR/LLM/TTS and tool/MCP work on a machine you operate.
What the video shows
A desk demo (~98s) of a XiaoZhi-style companion in a clear acrylic case — blue OLED face (eyes / smile / clock + Wi-Fi icons), speaker, tactile buttons — sitting next to the round Waveshare AMOLED that runs quack-quack. Same session as the still photo: two IoT tracks on one desk. Exact board SKU for the clear-case unit is not labeled in the clip; treat it as XiaoZhi-class hardware paired with this server ecosystem, not as a claim about every PCB line.
Why it matters
sihah-esp32 taught the firmware side. xiaozhi-ai is the other half: owning the backend so voice AI on ESP32 is not a black-box cloud demo. Notes and protocol context I use while wiring this up live in a Feishu wiki.
Related
- Firmware upstream: 78/xiaozhi-esp32
- Product / docs hub: xiaozhi.me
- Hands-on firmware build: sihah-esp32
- Parallel (not the same) desk duck: quack-quack
More projects
view all →
sihah-esp32
A voice-controlled AI chatbot on ESP32 — offline wake word, streaming speech to an LLM and back, and device control over MCP. My build of the open-source XiaoZhi project.
ohmyclawd
A Claude Code usage monitor on a tiny ESP32 desk display — session and weekly usage as animated pixel art, with a retro clock. My macOS + cloud fork of an open-source project.
Quack prototype
A pixel-duck agent UI for the ESP32 Cheap Yellow Display — idle / listening / thinking / talking sprites, an I2S mic, and touch pages — prototyped in the Wokwi simulator. The precursor to ohmyclawd.