When it comes to smart speakers, Amazon has Alexa, Apple has the HomePod, and Google has Nest. If you want something private — that runs locally — to control your home, there weren’t many alternatives.
Or there weren’t until now. To fill that gap, Nabu Casa, the sponsor of the Home Assistant open source project, released the Home Assistant Voice Preview Edition.
I bought six of these to replace six HomePods I had scattered around the house. After using them for a while, the question is: can you trust this for everyday use, or is it better to wait for a release without “preview” in the name?
The Home Assistant Voice Preview Edition (Voice PE from now on) is a voice assistant device created by Nabu Casa, the commercial arm of the Home Assistant project. The idea is simple: a voice device you control, without depending on big tech companies.
The box is small and minimalist. Inside you’ll find the device itself, a quick-start guide and warranty information. As it’s 2025, it doesn’t include a USB-C cable or power adapter, so prepare to provide those yourself.
The design is understated: a semi-translucent white square, an LED ring on top for visual feedback, a central button, an iPod-like dial for volume, and a physical switch that cuts power to the microphones.
On the back there’s a USB-C power port and a headphone jack to connect external speakers.
Initial setup
Setup is straightforward. Power the device, open the Home Assistant app and it finds the device. A setup wizard guides you through the rest. Choose the wake word, the assistant’s voice, and you’re good to go.
Important: the device requires a Home Assistant server to work. If you don’t run Home Assistant at home and don’t plan to install it, this product is not for you.
On the plus side, Voice PE doesn’t require a cloud account (anywhere, really) and gives you a level of control unmatched by the big tech alternatives.
The microphones
I have six Voice PEs at home and they all work fine. I configured the wake word for local recognition and the devices generally understand my wife and me quickly.
The mic isn’t great, but it isn’t terrible either — it’s just… okay. Not as good as the HomePod mic, but good enough.
The audio processor and two microphones do a decent job of echo cancellation and noise reduction. If you’re nearby it will understand you without trouble. From across a large living room you’ll need to project a bit more.
Sensitivity can be adjusted in settings, which helps.
It doesn’t have speaker recognition like the HomePod, so anyone can give commands. If you’re watching something and make a request, it won’t distinguish between your voice and the TV’s audio.
You can use up to two wake words, each with its own personality and language — a neat perk. “Okay Nabu” can accept English commands while “Hey Jarvis” can handle Portuguese, for example.
Officially there are three possible wake words: Okay Nabu, Hey Jarvis, and Hey Mycroft. The community has trained many others, but using those requires manually reconfiguring firmware, which is tedious.
The power of AI
Request processing can happen via Home Assistant (locally), or through an AI model either local or remote (cloud).
This is where Voice PE truly shines. The flexibility is enormous. The advantages of using an AI to process requests, local or remote, are many.
For example, an LLM can infer what you mean even if you’re imprecise — something plain Home Assistant can’t do. “Play the punk album with the kid naked on the cover” will be understood by an LLM but not by vanilla Home Assistant.
You can also control the assistant’s personality. Want responses in rhyme? In a formal tone? Like a pirate? You can. That goes far beyond what Amazon or Google allow.
In the video below, GLaDOS answers a request in my house:
The level of control and possibilities are unmatched. Siri, Google Assistant or Alexa can control a limited set of smart devices, while Home Assistant offers virtually unlimited options thanks to the number of integrations already available and the project’s open source nature, which allows building integrations when something you need isn’t available.
For example, at home I can tell the assistant to run the robot vacuum in specific rooms, ask if it’s raining inside the house, or even ask when we last changed the baby’s diaper.
Having an AI connected to a search engine for everyday questions is useful too. It won’t change your life, but it’s handy from time to time.
Not everything is rosy, of course. Using remote AIs means higher response times, so commands can take longer to complete.
I don’t have hard data, but it seems Google Assistant, Siri and Alexa servers have higher uptime. I used Siri for a year and don’t recall outages. Claude, on the other hand, goes down every other day. It hasn’t affected me yet, but the risk is higher.
Privacy and control
The device offers privacy levels that let you decide what’s comfortable. You can connect to a remote AI if you want, but you can also use local AIs or let Home Assistant handle everything with no external connections.
The physical switch that cuts microphone power is a standout feature. It’s not a software mute — the microphone connection is physically disconnected.
The real cost
The device costs about USD 59, competitively priced with other assistants. But here’s the reality: although the device is cheap, the real investment to make it truly useful is higher. Privacy and control aren’t free.
Using a remote AI means paying per interaction with the assistant, which can be an issue. Depending on your setup, you may also pay to generate the speech voice.
That said, models like Claude Haiku or Gemini Flash are fast and very cheap. Depending on usage, monthly costs can be under USD 2.
Running local AIs with Ollama is the best option for privacy and latency, but for Brazilian standards it’s almost impossible due to the high cost of GPUs locally. It becomes viable if you have an Apple Silicon Mac Mini gathering dust.
Local AIs that can reliably use external tools require more resources, so again cost is the limiting factor. A mainstream Alexa is cheaper in the long run if you ignore the hidden privacy cost.
The music situation
This is where things get complicated. If you use streaming services for music, you’ll be limited.
Music support isn’t native, but you can add it using a Music Assistant server and a script in Home Assistant so an LLM can search your library and play tracks. Voice support isn’t official yet — it’s a hack that sometimes works and sometimes doesn’t, depending on how the AI fills required fields.
On the other hand, you can create automations to play random tracks or specific albums, which helps. For example, I have an automation to play a random album and another to play the Kpop Demon Hunters OST (my wife is addicted).
Audio quality
Another issue: connecting the assistant to your own speaker is almost mandatory, because the device’s built-in speaker is among the worst I’ve ever heard. Truly awful. The internal speaker is designed for voice, not music.
Being able to use your existing speakers is fantastic for playback. The 3.5mm output with a dedicated DAC allows lossless output to external speakers. If you already own good speakers, Voice PE’s audio shortcoming disappears.
Conclusion
If you can run AI models locally, Voice PE is the best voice assistant on the market. The combination of privacy, control and flexibility is unbeatable.
If you can only run everything remotely, I can’t recommend it right now. Interaction costs, latency and dependence on external services take away much of the device’s appeal.
That said, this is only a “preview” and things are likely to improve. New models are released daily and running them locally is getting easier. A second version with better microphones and cheaper hardware for local AI would be very welcome.
I used HomePods integrated with Home Assistant for a long time, but I bought Voice PE to have full voice control over the house. The reasons were many, but the main ones were greater device control, control over the assistant’s personality, and the ability to connect real speakers.
On a day-to-day basis, roughly 80% of what you’d ask a voice assistant to do is already supported by other assistants.
For my use, the downsides were acceptable for a “preview.” Your mileage may vary.
For smart home enthusiasts who already use Home Assistant and want to take part in developing an open source alternative to the big tech assistants, now is a great time to join. If you want something that “just works” like an Alexa or Google Home, wait for the final release.