Jens Kastensson, Author at Hey Aragon

Aragon AI Chat: A Local Assistant That Already Knows Your Home

Your Aragon Maestro has its own AI assistant built in — one that understands and responds to you using a language model that runs on your own hardware, not a cloud AI service, with no subscription behind it. It’s grown into something with real range: six specialist tools it can call on to actually get things done, and, this release, noticeably faster at two of the things people ask it most.

What it is

Aragon AI Chat is a conversational assistant that lives on your Maestro hardware itself — the part that understands what you’re asking and decides how to answer runs locally, on the device. You interact with it by typing in the browser or speaking to it, in whichever language your Maestro is currently set to, and it answers in plain language.

That local processing covers understanding and responding to you — it doesn’t mean every tool it uses stays offline. A few of them do reach out to the internet to get real, current information: the weather tool checks a weather service, and Wikipedia lookups and web searches go out to fetch what they’re asked for, the same way any lookup of live information has to. Others, like the calculator and the date/time tool, work entirely on their own without needing anything outside your home.

This is a different thing from Aragon MCP, which lets you connect an external AI assistant like Claude or ChatGPT to your Maestro. Aragon AI Chat is the one that’s already there, understanding and responding to you on your own hardware.

Six specialists, one assistant

A general-purpose language model is only as useful as what it can actually go and check. Aragon AI Chat has six purpose-built tools behind it, each handling one kind of request precisely rather than guessing:

Weather — current conditions and forecasts for your home location or anywhere you name, with “today” and “tomorrow” correctly told apart and a retry built in so a slow network moment doesn’t turn into a wrong or missing answer.
Calculator — unit conversions in the words people actually use (“fifty centilitres into millilitres,” not just abbreviations), plus live currency conversion across 31 currencies using real European Central Bank exchange rates.
Date & time — countdowns and date questions, including holiday names and phrases like “next Friday,” answered correctly without the assistant ever having to guess today’s date itself.
Wikipedia — fast factual lookups, kept away from the questions it isn’t suited for (a commercial product name, a brand) so it doesn’t hand back the wrong answer for those — the assistant switches to web search instead.
Web search — search through a self-hosted SearXNG instance, covering general questions, current events, and news from more than one source, so a single outage elsewhere can’t take news search down entirely.
Aragon Agent — the coordinator behind all of the above, also available to voice commands, scripts, and automations, not just the browser chat window — the same tool-calling loop, everywhere you’d want to use it.

Built for speed

Two concrete improvements worth calling out by the numbers:

Spotify voice control used to take around 26 seconds to understand a request like “play X on Spotify.” Switching the model behind it now gets that down to 3–9 seconds — plus fixes so it no longer plays the wrong track when an artist name is ambiguous, and no longer fails outright when it mishears which room asked.

Home automation rules got the same model change under the hood, along with an engine rewrite — roughly twice as fast at turning a spoken rule into an action, with corrected handling of French and German solar-time phrasing along the way.

Why it matters

None of this needs an account or a monthly fee. It’s already running, it already speaks your language, and each release has made it either capable of more or faster at what it already did. If you haven’t tried it lately, it’s worth another conversation.

Aragon MCP: Control Your Smart Home From the AI Assistant You Choose

AI assistants have become part of daily life for a great many people — answering questions, drafting documents, planning a trip, or just helping think something through. Increasingly, those assistants don’t just talk anymore. They act: booking a reservation, editing a file, checking a calendar. A new open standard called the Model Context Protocol (MCP) is what makes that possible — it lets an AI assistant discover a set of “tools” a service exposes, understand in plain language what each one does, and call them on your behalf.

Until now, your Aragon Maestro wasn’t one of the services an AI assistant could reach. Aragon MCP changes that.

What it is

Aragon MCP exposes Aragon Maestro’s home-automation controls — lights, shutters, thermostat, scenes — as a small set of clearly defined tools an AI assistant can use. This isn’t a future promise: it’s working today, and you can connect an AI assistant to your Maestro right now, with more AI platforms joining as they adopt the standard. That includes today’s popular cloud-based AI assistants, and local, self-hosted AI assistants running on your own hardware — MCP is an open standard, so it doesn’t matter where the AI itself lives, only that it speaks MCP.

Once connected, you can simply say what you want:

“Turn off the living room lights.”
“What’s the temperature set to in the office?”
“Close the shutters everywhere.”

Your AI assistant figures out which tool to call and what to say, and Aragon Maestro carries it out immediately — through the exact same natural-language pipeline that already powers your voice control today. (Scheduling a command for later is a separate Maestro feature; Aragon MCP itself always acts right away, the moment you ask.)

Speak however you actually speak

You don’t need to learn a command syntax, and you don’t need to speak English. Your AI assistant translates whatever language you’re comfortable in into the phrasing your Maestro understands, on the fly. Say “Tänd i vardagsrummet” in Swedish, and what reaches your Maestro is simply “turn on the living room” — no setup, no language setting to change, just ask in your own words.

The same flexibility covers small mistakes. Say “turn off living,” or misspell it as “livingrooom” — your AI assistant already knows the actual rooms your Maestro has, so it recognises what you mean and sends your Maestro the exact right name before the command goes out. You don’t have to get the wording exactly right; your AI assistant closes that gap for you.

How it works

Connecting is standard, secure, and doesn’t ask you to manage any secrets by hand. Your Maestro exposes an MCP server at a single URL; your AI assistant discovers it and authenticates using OAuth 2.1 with PKCE — the authorisation approach the MCP standard itself recommends. There’s no API key or client secret to generate ahead of time: your AI assistant registers itself automatically the first time you connect (via OAuth Dynamic Client Registration), then takes you through a normal browser sign-in — except here, you’re signing in with your own Maestro owner password, not handing credentials to any third party.

Reaching your Maestro from outside your home network doesn’t require opening a port on your router. A private, end-to-end encrypted tunnel connects your AI assistant to your Maestro and nothing else on your network.

Optional — and the core of Maestro still needs neither the internet nor a subscription

This is worth being precise about: Aragon MCP is an opt-in feature you turn on if and when you want it. It’s not a requirement, and it changes nothing about how your Maestro behaves if you leave it off.

That matters because of what Aragon MCP is layered on top of. Voice control on Maestro runs 100% on-device — speech recognition, language understanding, and text-to-speech all happen locally, with no internet connection and no subscription required for any of it. If you choose to connect a cloud-based AI assistant through Aragon MCP, that specific request naturally passes through that assistant’s own cloud service to be understood — that’s simply how a cloud AI assistant works, on any platform, and it’s true the moment you decide to use one. Choose a local, self-hosted AI assistant instead, and even that step can stay on your own network. Either way, what doesn’t change is what happens next: the instant your request reaches your Maestro, it runs through the same local, on-device pipeline that has always executed every home command. No separate cloud “smart home” account. No home hub or subscription service sitting in the middle. No company beyond the one you already chose to talk to.

Why it matters

Smart home platforms have a habit of picking a favourite assistant for you and locking the door behind it. Aragon MCP does the opposite: it meets you wherever you already are, in whatever language and phrasing feels natural, using whatever AI you already trust — without asking you to adopt a new app, pay a new subscription, or hand your home to a new company. As the AI landscape keeps shifting, your home doesn’t have to keep up with it — it just has to stay open, and that’s what MCP gives it.

Home Automation rules! Especially with the new Aragon Maestro feature

Home Automation really does rule, now that the Home Automation Rules feature has been incorporated into the Aragon Maestro toolkit. This feature lets users create recurring automations using natural language input. Aragon Maestro can be controlled in real-time with on-the-spot voice commands. Now, with the new Home Automation Rules feature, actions can also be scheduled to work automatically during a period of the user’s choosing, or in relation to day/night natural light changes.

How It Works

The Home Automation Rules feature provides a web-based interface where users type commands in plain language to create automation rules. The system accepts input in English, French, or German. An on-device Large Language Model (LLM) processes these text commands and converts them into structured schedules that the Maestro then carries out automatically.

Examples of supported rules include:

“Turn on the kitchen lights 15 minutes before sunset every day.”
“Activate cozy scene at sunset on weekends.”
“Set the bedroom temperature to 22 degrees on weekday mornings.”
“Close all shutters at 10 PM on weekends.”

The web interface displays all active rules and allows users to add, edit, enable, disable, or delete them as needed.

Complete Privacy and Offline Operation

The Home Automation Rules feature operates entirely on-device with no cloud dependency. What’s said at home, stays at home!

On-device processing: Natural language parsing, rule scheduling, and command execution all happen locally on the Aragon Maestro. The system requires no internet connection for its core functions.

No data transmission: Automation rules are stored in a local database on the device. Nothing is transmitted to external servers or analyzed by third parties.

Reliable operation: Because the feature does not depend on internet connectivity, scheduled automations execute reliably regardless of network status.

What Can Be Automated

Any action that can be controlled through voice commands can also be scheduled as an automation rule. This includes:

Scenes: Activate predefined scenes such as “Movie Night” or “Good Morning” at specified times.
Lighting: Control lights to turn on at sunrise, dim to a specific percentage, or turn off at bedtime.
Shutters and blinds: Schedule blinds to open in the morning or close at sunset.
Heating: Set temperature targets for specific rooms at designated times.

Technical Implementation

The Home Automation Rules feature consists of three main components working together:

Natural Language Processing Engine: An on-board LLM interprets user input and extracts the command to execute along with scheduling details (time, days, solar events, and offsets). The system uses language-specific prompts to accurately parse commands in English, French, or German.

Scheduling System: A scheduling engine converts parsed rules into scheduled jobs. The system handles both fixed-time rules and solar-based triggers. For solar events like sunrise and sunset, the engine calculates precise times based on the user’s geographic location and updates these calculations daily to account for seasonal changes.

Execution Layer: When a scheduled time arrives, the system publishes the command to the same MQTT endpoint used by Aragon’s voice control system. This ensures that any command that works via voice will also function as a scheduled rule.

Multi-Language Support

The system provides native support for three languages. Users can write rules in their preferred language, and the LLM correctly interprets scheduling expressions such as:

English: “every weekday”, “at sunset”, “on weekends”
French: “les jours de semaine”, “au coucher du soleil”, “le week-end”
German: “an Wochentagen”, “bei Sonnenuntergang”, “am Wochenende”

The backend normalizes these expressions into a standardized format for consistent scheduling across languages.

User Interface

The web-based interface provides a straightforward workflow:

Users enter a rule in natural language
The system parses and validates the rule, showing the interpreted command and schedule
Users confirm and create the rule
All active rules are displayed in a list with options to edit, enable/disable, or delete

The interface also allows users to set their location (city or coordinates) to ensure accurate calculation of solar events.

Availability

The Home Automation Rules feature is available as a software update for all Aragon Maestro users. The update can be installed through the Maestro’s web interface. For more information about Aragon Maestro and its capabilities, visit heyaragon.com.

Sound Event Detection – a new Aragon Maestro feature!

Aragon Maestro’s latest update introduces Sound Event Detection (SED). This is a sophisticated feature that enables real-time monitoring and classification of various acoustic events in the home environment. YAMNet neural network technology catagorises sounds into Alert, Human, Household, and Animal categories.

Sound Event Detection Overview

The Sound Event Detection (SED) feature in Aragon Maestro 2.1.0 processes audio streams from Aragon Satellites using UDP protocol, analyzing them for both wake words and sound events. This system employs YAMNet for sound classification, which therefore enables real-time detection and categorization of various acoustic events in the home environment. While the Maestro can handle multiple satellites simultaneously, it is recommended to configure only one satellite for SED during the initial deployment phase. To utilize SED, the satellite’s wake word system must be set to ‘Centralized’. This therefore allows the Maestro to manage wake word detection and enabling audio streaming for sound event analysis.

Configuring Sound Event Detection in Maestro

To set up SED in Aragon Maestro, follow these steps:

Update the system to Aragon Maestro version 2.1.0 and Wakeword (Central) to version 1.1.1 from the Maestro About page
Select the desired satellite and set its Wakeword system to ‘Centralized’
Save the configuration
Enable SED through the “Step 5. Configure ‘Sound Events’” dialog
Access the SED user interface at MAESTRO-IP:1880/dashboard/all

For optimal performance in this early release, we recommend that only one satellite should be configured for Sound Event Detection. The ‘Centralized’ Wakeword setting is crucial as it allows the Maestro to handle wake word detection and enables audio streaming for SED functionality.

Monitoring Interface Features

The SED monitoring interface is accessible at MAESTRO-IP:1880/dashboard/all. It offers a comprehensive view of detected sound events through five tabs: All Sounds, Alert, Human, Household, and Animal.

Each tab features:

A time chart with color-coded dots representing detected sounds by category
A Sound Event Gauge displaying instant detection values
A Sound Activity time chart for temporal visualization
A Sound Event list showing the latest events in text format

The ‘All Sounds’ tab provides an overview of all categories, while the other tabs offer detailed information specific to each sound type. This intuitive interface allows users to easily monitor and analyze various acoustic events in their environment, enhancing the overall functionality of the Aragon Maestro system.

Sound Categories and Notifications

The SED system categorizes sounds into four main groups: Alert, Human, Household, and Animal, each with specific detection thresholds and aliases. For instance, Alert sounds include alarms and glass breaking, while Human sounds encompass speech and physical noises. Users can configure custom notifications through Node-RED, integrating with pushover.net for remote alerts. In addition, for off-site monitoring, Tailscale VPN enables secure connections to the Maestro, allowing users to access the SED dashboard and examine detailed sound event charts from anywhere. This comprehensive approach ensures users stay informed about significant acoustic events in their environment, even when away from home.

What do ProKNX smart speakers offer that Big Tech speakers cannot?

Smart speakers produced by Big Tech are very cheap and, as well as controlling your smart appliances, allow you to access all sorts of information. However, in order to have access to this information world, the end-user must pay in an alternative way; namely, with the harvesting and use of their personal data. The ProKNX smart speakers are first-and-foremost, made so that the end-user can maintain their privacy, but they also have a number of important advantages that ProKNX smart speakers over the voice assistants offered by Big Tech. Read on to find out the benefits of using an ethical smart speaker.

Privacy:

What’s said at home, stays at home. Everything the user says is private. Privacy of the user is valued highly by ProKNX, therefore no data is collected. In fact, voice recognition runs entirely on the device.

Operational safety:

The voice control works without an Internet connection, therefore is more reliable. The smart speakers continue to function without a hitch even with a poor Internet connection.

Investment security:

The technology continues to work as it did on the first day it was installed, even after years of use, as it is independent of automatic “silent” updates. As a result, this also means that even if the company stops selling or supporting the device, it will continue to work as it did out of the box.

Adaptability:

Commands and verbal interactions can be set in any variation according to the requirements. The choice of words and sentences can be said in thousands of different ways. This flexibility of choice also means that each device can be adapted to the unique needs of individual customers.

Connectivity and Interoperability:

The assistant controls lights, shutters, scenes, heating, music, TV, and much more. Many home server systems and gateways are supported out of the box. “Plug and Talk” functions with the worldwide KNX home automation system standard, or with gateways from Philips HUE and IKEA Tradfri. Also, SONOS Music systems are supported without additional configuration. ProKNX products are highly compatible with as many systems as possible, allowing integrators to add a large number of other IoT products using the installed NodeRed IoT editor.

Aging in Place: Anomaly Detection:

For elderly or handicapped people voice control of home appliances facilitates daily routines. Although ProKNX smart speakers work without need for the Internet, ARAGON COMPANION has a special feature that allows it to call a nominated carer for help when an untoward event occurs. Additionally, ARAGON can unobtrusively monitor actions and movements of a person, and analyse the data on-device. Movement detectors, door contacts, and switch operations record activity of the individual. Consequently, if the pattern seen diverges from the regular routine, ARAGON will ask if help is needed. No response, or an affirmative answer, will result in the smart speaker calling a carer.