How to add voice control without Amazon or Google
Practical local voice control for Home Assistant and privacy-focused smart homes - Home Assistant Voice PE, Apple HomePod/Siri, and the open Matter route.
On this page
Voice control without Amazon or Google means leaning on local processing and accepting a few tradeoffs. Here’s how to do it right.
The Home Assistant Voice Preview Edition is the best pure-local pick - Wi-Fi hardware with on-device wake-word detection, feeding a speech pipeline that runs on your own server. Apple HomeKit plus a HomePod suits people already in Apple’s world, Aqara plus Matter is the most open foundation but needs a voice front end on top, and an ESPHome satellite you flash yourself reaches the same Assist pipeline for less money and more fiddling.
The Home Assistant voice path
The most capable local voice option is the Home Assistant Voice Preview Edition, which I run: dedicated voice-satellite hardware built around an ESP32-S3 with an XMOS audio chip and dual far-field microphones. It connects to your network over 2.4 GHz Wi-Fi and talks to your Home Assistant server, where the speech pipeline (speech-to-text, intent handling, text-to-speech) runs. You choose whether that pipeline runs fully locally on your server or uses Home Assistant Cloud - either way it never goes to Amazon or Google. If you want to run it fully on-device, my comparison of the Piper and Whisper local voice engines covers the speech-to-text and text-to-speech halves. If you still want Google Home as a backup front end without going all-in on it, see Home Assistant with Google Home fallback. If you are weighing it against the community-built FutureProofHomes Satellite1.1, see Satellite1 vs Home Assistant Voice Preview Edition.
Wake-word detection runs on the device using the microWakeWord engine, with “Okay Nabu,” “Hey Jarvis,” and “Hey Mycroft” available out of the box - the on-device counterpart to openWakeWord, which instead runs on your server for satellites that cannot do the listening themselves. (The board does carry Bluetooth 5.0, but it reaches Home Assistant over Wi-Fi, and it doesn’t run “Rhasspy” as an on-board engine.)
Setup is straightforward: power it over USB-C, join it to Wi-Fi, and adopt it in Home Assistant. The downsides versus Alexa: natural-language understanding is more literal, so clear commands like “turn on the living room lights” work better than vague ones, and you’ll get the best results with a local LLM or Home Assistant’s Assist intents configured.
Pair the satellite with a hub like Home Assistant Green running your Home Assistant server (the Home Assistant Yellow works too, though it was discontinued in October 2025).
Build your own satellite with ESPHome
The Voice Preview Edition is the finished product, but the same Assist pipeline accepts satellites you flash yourself. The boards below all run ESPHome, connect to your own Home Assistant server over the LAN, and never touch a cloud.
The Espressif ESP32-S3-BOX-3 is the one Home Assistant’s own voice documentation points at: an all-in-one dev puck with dual microphones, a speaker and a 2.4-inch touchscreen, running the wake word on-device. Setup is fiddlier than a finished product, which is the trade you are making for the price.
If far-field pickup is the problem, the Seeed Studio reSpeaker XVF3800 pairs an XMOS four-microphone array with a XIAO ESP32-S3 and does 360-degree pickup out to about five meters, with echo cancellation, beamforming and noise suppression. It ships as a bare board, so budget for the case as a separate part. If you would rather not build an enclosure, the FutureProofHomes Satellite1.1, which I also run, puts a 16-core XMOS XU316 DSP in a finished speaker puck that also runs ESPHome and carries Snapcast multiroom audio, at around $135 - when it is in stock, because it is frequently backordered.
At the cheap end, M5Stack sells thumb-sized pucks. The Atom VoiceS3R is an ESP32-S3 kit with a microphone, speaker and codec that M5Stack documents as a ready Assist satellite, and the older ATOM Voice, which the company sold as the ATOM Echo before renaming the line, is cheaper still. Both are limited by a single small microphone, so they suit a desk or a nightstand rather than a room you talk to from across.
There is also a salvage route worth knowing about: a first-generation Echo Show 8 can be wiped and reflashed with LineageOS, which turns it into a screen and a far-field microphone on the same Assist pipeline for the price of a used gadget. Reflashing an Echo Show 8 into a local Home Assistant satellite walks through what that gets you and where it falls short.
Apple HomePod as a local-leaning Siri
Apple’s HomePod devices are a middle ground. HomeKit runs on your local network, and Siri handles many requests on-device, though some still reach Apple’s servers. The trade-off is Apple ecosystem lock-in - everything flows through your Apple ID and HomeKit.
Add an Apple HomePod mini or a second-gen Apple HomePod, then expose Home Assistant entities to HomeKit using Home Assistant’s HomeKit Bridge. You can then control Home Assistant devices by voice through Siri. You’ll need an iOS device for setup. This works well if you’re already in Apple’s world; if not, the value drops.
Aqara and Matter: the open route
The Aqara Hub M3 runs Zigbee devices locally and also acts as a Thread border router and Matter controller. Pair it with Aqara gear like the Aqara Smart Lock U200 and Aqara Motion Sensor P2 for routines that keep working if the internet drops.
Matter is maturing as a local-first protocol. The Shelly Plus 1 and Shelly Plus 2PM integrate locally with Home Assistant over their native API (these Gen2 ‘Plus’ models have no native Matter - that arrived with Shelly’s Gen3 and Gen4 lines - and both are now discontinued, so a new build should use the Shelly 1 Gen4 and Shelly 2PM Gen4 instead). Run the Matter server on a Home Assistant Green with a Connect ZBT-2 stick (the successor to the discontinued ZBT-1, which is fine if you already own one) - or a Home Assistant Yellow (discontinued, but its Thread border router radio is built in) - for a vendor-agnostic setup.
The limitation: Matter does not itself provide a voice assistant. For local-only voice on a Matter setup, you still pair it with Home Assistant Voice or an Apple/HomeKit voice front end.
The Hubitat alternative
The Hubitat Elevation C-8 runs automations locally with no cloud dependency. Buy the standard C-8 Pro if you are buying new: Hubitat stopped making the plain C-8 in March 2026, so remaining plain stock is clearance or secondhand, and the standard Pro carries the same Zigbee 3.0 and Z-Wave 800 Long Range radios and acts as a Matter controller over Wi-Fi and Ethernet. Hubitat also sells a Thread Special Edition of the Pro whose Thread radio replaces the Zigbee one, so it is the wrong hub to buy if you have Zigbee gear. It has no built-in voice hardware, but it pairs with voice front ends that don’t phone home to Amazon/Google - namely Apple HomeKit and the Home Assistant voice path above. Bridge Hubitat into Home Assistant to combine Hubitat’s solid Z-Wave/Zigbee handling with Home Assistant’s voice options.
What you’ll sacrifice
Local voice means more literal language understanding than Alexa, fewer “skills,” and more setup. The upside is privacy and operation that survives an internet outage.