// the find
HeyWillow/willow
Open source, local, and self-hosted Amazon Echo/Google Home competitive Voice Assistant alternative
Willow is ESP32 firmware that turns Espressif dev boards into local voice assistants, offloading STT/TTS/LLM to a self-hosted inference server rather than cloud APIs. It targets privacy-conscious home automation users who want an Echo-like device without sending audio to Amazon or Google. Requires specific Espressif hardware (ESP-BOX series) and your own Willow Inference Server instance.
1. The architecture is genuinely well thought out — the ESP32 handles wake word detection and audio capture locally, then streams to your own inference server over WebRTC/REST, so nothing hits external APIs by default. 2. Home Assistant and OpenHAB integrations are first-class, not bolted on — separate endpoint files with proper abstractions. 3. Built on ESP-IDF with CMake and proper CI (build, format, container workflows), so it's not a hobby sketch — it follows embedded C conventions. 4. OTA update support built into the firmware means you can iterate without physically accessing deployed devices.
1. The README is basically empty — it punts all documentation to heywillow.io, a domain you don't control and that may or may not exist when you need it. Embedded projects with external-only docs are a long-term liability. 2. Last push was February 2026 and there are only 119 forks for 3k stars — this repo is either mature/stable or quietly abandoned; the discussion forums would tell you which but the code itself gives no signal. 3. Hard dependency on specific Espressif hardware (ESP-BOX, M5Stack CoreS3) — there's no abstraction layer, so porting to other ESP32 boards means patching ESP-ADF configs directly. 4. The inference server is a separate repo you also have to deploy and maintain; the combined operational burden (ESP32 firmware + GPU/CPU inference server) is non-trivial for someone expecting an Echo-like plug-and-play experience.