Thirteen secondhand Echo Dots, Home Assistant, and a very different idea of what a home assistant could be.
By Odin · Aesir Reforge
Right now, my house has a Halloween cast.
The Crypt Keeper, a vampire, and a witch rotate through the assistant, each with a playful personality. They're part of a system with its own name, its own wake word, and a voice I can shape around the season. When December comes around, I'm planning to give Santa and an elf their turn.
That probably tells you something about the kind of project this became.
Underneath the Halloween fun is my own Home Assistant build, using EchoMuse and ElevenLabs. I've also implemented my own memory system so the assistant can remember my family. The interesting part is how those pieces come together: familiar hardware, useful home controls, conversation, and enough personality to make the whole thing feel like it belongs here.
And the speakers? I bought a used lot of thirteen second-generation Echo Dots for $100.
A second life for a box of Echo Dots
There's something satisfying about finding a new job for hardware that still has plenty to offer. A small speaker with microphones and buttons is a useful little device. Give it a different role, and that secondhand purchase starts to look a lot more interesting.
The $100 bought my lot of thirteen speakers. That gave me a collection of matching hardware to build around, instead of starting with a different device for every idea.
The project that makes this possible is EchoMuse. It repurposes second-generation Echo Dots as voice satellites for Home Assistant. Think of a satellite as a place where you can speak to the system and hear its answer. The supporting software does the work behind that conversation.
That arrangement is a big part of the appeal. You can reuse a familiar little piece of hardware while changing the experience around it. What it responds to, how it sounds, and what jobs it can help with become things you can work on.
For Aesir Reforge, that's a pretty good illustration of the name. There's useful life left in old technology, especially when you're willing to look beyond its original job.
How the pieces fit together
The easiest way to understand this build is to follow a conversation.
You say the wake word. The system hears the request, works out a response or an action, and speaks back through the speaker. Home Assistant sits at the center of the home-control side. EchoMuse connects the repurposed Dots to that voice setup. ElevenLabs provides the voices.
Home Assistant's ElevenLabs integration supports turning written responses into spoken audio, including selecting a voice. That matters when you're building something with character. The words are one part of the experience; the delivery is another.
The AI personality is its own piece. Home Assistant supports conversation agents with custom personality instructions, so you can shape the way the assistant replies. A concise, practical helper and an overdramatic Halloween host can approach the same question very differently.
Putting those pieces together gives me room to customize the experience. I can work on the useful jobs and the personality as parts of the same project. The voice becomes a way into the house's routines, and the character gives those interactions a little life.
It gets to have its own name
One of the nicest parts of this build is giving the assistant its own identity.
Mine has its own name and wake word. That small change makes a surprisingly good starting point for thinking about what you want a home assistant to be. Do you want something calm and brief? A friendly helper? A character that sounds like it escaped from a Halloween attraction?
Custom wake words are part of Home Assistant's voice ecosystem. Getting a good one involves choosing and testing a phrase the system can recognize. It's one of those places where the technical work supports a very simple everyday experience: call it what you want to call it, and have it answer.
The personality can go further than the name. You can think about how much it talks, the tone it uses, and how much humor belongs in an ordinary reply. A useful assistant ought to suit the people using it. Some homes will want the shortest answer possible. Apparently, mine has room for a vampire.
October has a cast of characters
For October, I've got the Crypt Keeper, a vampire, and a witch rotating with their personalities. That gives the same underlying assistant a different flavor depending on who's taking the role.
The appeal is easy to picture. A routine answer can arrive with a bit of theatrical flair. A familiar interaction can feel like part of the season. You get to carry the fun beyond the decorations and into something you actually use.
There's also room to keep developing it. One idea I want to explore is letting the characters banter with each other. I can see the entertainment value in a vampire and a witch having opinions about the same request. That's on the idea list, and it would need to earn its place without turning a simple question into an entire stage production.
December is the next seasonal plan: Santa and an elf. The same home-assistant project can take on another character as the year changes. That's a fun kind of customization because it gives you something fresh to do with equipment you already have.
A memory makes it more personal
The memory system is another part of what I've built. It lets the assistant remember my family, adding continuity to the experience.
That opens up a more interesting design question than simply choosing a voice: what is actually useful for an assistant to remember?
Think about preferences and shared context. In a personalized setup, a useful memory could be that someone prefers a short answer, or that a certain kind of suggestion has already been ruled out. Those are examples of the kind of continuity worth designing for. Everyday conversation becomes easier when you don't have to rebuild all the background every time.
There's a separate question about what's happening in the house right now. A remembered preference and a current device reading serve different purposes. Keeping those ideas clear is part of making an assistant useful: long-term context helps personalize the conversation, while current information helps it answer a question about the present.
For me, the family-memory piece is part of making this project feel like our assistant. The voice and the seasonal characters give it an identity. Remembering the people who use it gives the interaction somewhere to go beyond the first introduction.
How far can you take it?
Once you have a voice interface, the natural next question is what you want it to reach.
Home Assistant gives you a way to bring compatible devices and routines together. The voice assistant's reach is something you deliberately configure: you choose which entities are exposed to Assist. An entity might represent a light, a sensor reading, or another supported function.
From there, the interesting work is choosing useful combinations. A setup could let you ask about a room's temperature, control selected lights, or start a prepared routine. A movie-time routine, for example, could bring several lighting changes under one familiar request. These are possibilities to build around the equipment and routines in a particular home.
You can go deeper by connecting the information an assistant can read with the actions you've decided it should perform. That takes some thought. What should a request mean? Which room are you talking about? Would a single clear routine be easier than asking for several separate changes?
The depth comes from answering those questions well. A small setup that understands the job you want done can be more useful than a long list of connected devices that nobody enjoys using.
Simple to use is something you build toward
From the everyday side, the idea is wonderfully simple: speak, get a useful answer, carry on with your day.
Getting there is the project. There's hardware to prepare, software to connect, a wake word to test, and routines to make understandable. Speaker placement matters. Room names matter. So does deciding when a personality should keep its response short.
The easiest way to make progress is to give each part a clear job. Get a useful interaction working. Try it where you'll actually use it. Then add the next thing. A working light command is a good foundation for a more elaborate routine; one well-chosen personality is a good place to start before giving the whole house a seasonal cast.
That leaves room for both kinds of interest. You can enjoy the straightforward convenience of asking for something, or you can get deep into shaping the system behind it. This project happens to give me a reason to do both.
Want to build something that feels like yours?
This is the kind of hands-on technology work I do through Aesir Reforge: taking an idea, choosing how the pieces should fit together, and turning it into a setup people can actually use.
Maybe your starting point is a custom voice assistant. Maybe it's a few useful Home Assistant routines, a cleaner dashboard, or a better way to use equipment you already own. Maybe you've got an idea that gets more interesting every time you describe it.
Bring the idea. We can talk through the setup, customization, and practical work involved in making it yours. The best starting point is usually a real job you want help with and a clear picture of how you'd like it to feel.
Mine started with a lot of old speakers. Now it has a name, a personality, a memory, and a Halloween cast.
I'm looking forward to finding out what Santa makes of the place.