On-Device Assistant for Denon & Marantz

Where it is

The Assistant tab

Type a question, or hold the microphone and say it. There is nothing to sign into, no key to paste, and no account. If your device can run an on-device model, the assistant works the moment you open it.

Ask it in your own words. "How loud is it", "what am I watching this on", "which Dirac slot is loaded", "is Dynamic EQ on" - it resolves what you meant against the things your receiver actually reports, so you do not have to know what the setting is called.

The assistant's opening screen with suggested questions and a microphone button — Open it and ask, or tap one of the suggestions

Four questions answered with the subsystems each answer read, then a request to raise the volume held behind a Yes or No — Every answer names what it read. Anything that would raise the room asks first

What it can tell you

Everything the app reads off your receiver, which is everything on the screens in this guide. Not a fixed list: a chassis that reports IMAX gets asked about IMAX, and one that does not is never offered it.

Ask about	And it knows
Power and volume	On or standby, the level in dB, mute, eco mode, and your receiver's own volume limit
Source and sound	The selected input under the name you gave it, the sound mode, and the format being decoded
Calibration	Audyssey MultEQ, Dynamic EQ, Dynamic Volume, reference level offset, LFC, and which Dirac Live slot is loaded
Speakers	The layout, per-channel trims, subwoofer levels and how many subs are configured
Video	Resolution and HDR in and out, picture mode, video processing, input mode, HDMI audio output, auto lip sync
Zones	Each zone's power, input, volume, channel trims, sleep timer and quick select
The chassis	Front panel brightness, auto standby, Bluetooth transmitter, and each trigger by number
What is playing	The track on a network source and its transport state

The Signal Info card showing audio in and out, video in and out, sample rate, color space, colorimetry and pixel depth — The readings it answers from are the ones on your screens

Under each answer it prints what it read to produce it. That line is recorded, not written by the model, so it cannot be invented - and it is the fastest way to tell a real answer from a confident-sounding wrong one.

What it will do for you

Ask it to change something and it changes it. What differs is how much ceremony each action gets, and the asymmetry is deliberate.

Ask for	What happens
Power on or off, mute, unmute	Done immediately, then it tells you what it sent
Turn the volume down	Done immediately. The moment somebody wants it quieter is the moment they least want a dialog
Turn the volume up	Asks first. It is the one action whose worst case - a damaged driver - cannot be undone by the person standing in the room
Switch the input	Resolved against your own source list, under the name you renamed it to. Two possible matches and it asks which

The amount is yours to give. "Turn it down by 5 dB" moves 5 dB; "turn it down" with no number moves one step, the same as one press of the button. It will not decide on a figure you did not say.

Changing the sound mode from here is not supported yet. Your receiver organizes sound modes by category and selects within the active one, so it is not a single name to send - the Sound tab does it properly.

Device requirements

Which devices can run it

The assistant runs a language model on your own device. No account, no server, nothing to sign into - which also means the device has to have a model on it. That is not something the app can install for you, and it is the one requirement we cannot work around.

Platform	What it needs
iPhone, iPad, Mac	iOS 26 or macOS 26, on a device where Apple Intelligence is available and switched on. If it is off, the app says so and points you at Settings rather than hiding the feature.
Android	Gemini Nano, through Google AICore. The app reaches it with Google's ML Kit Prompt API, and that API has its own supported-device list - see below.
Windows, Linux	No on-device model exists to run, so the assistant is not shown. The rest of the app is unaffected.

Android: Google's supported devices

This is Google's list, not ours. It is reproduced here because it is the single thing that decides whether the assistant appears on an Android phone, and because Google publish it in a place most people never look.

Model	Devices
Gemini Nano v3	Google Pixel 10, 10 Pro, 10 Pro XL, 10 Pro Fold, Pixel 9, 9 Pro, 9 Pro XL, 9 Pro Fold · Samsung Galaxy S26, S26+, S26 Ultra · Honor Magic 8 Pro · iQOO 15 · Lenovo Idea Tab Pro Gen 2, Legion Tab Gen 5 · Motorola Signature · OnePlus 15, 15R · OPPO Find X9, X9 Pro, Find X8, X8 Pro, Reno 14 Pro 5G, Reno 15 Pro 5G, Reno 15 Pro Mini 5G, Reno 15 Pro Max 5G · realme GT 7T · Sharp AQUOS R11 · Sony Xperia 1 VIII · vivo X200T, X200, X200 Pro, X300, X300 Pro
Gemini Nano v2	Samsung Galaxy Z Fold7, Z TriFold · Honor Magic V5, Magic 7, Magic 7 Pro · iQOO 13 · Motorola Razr 60 Ultra, Razr Ultra 2025 · OnePlus 13, 13s · OPPO Find N5 · POCO F7 Ultra, F8 Pro, F8 Ultra, X7 Pro, X8 Pro · realme GT 7 Pro · vivo X200 FE, T4 Ultra · Xiaomi 14T Pro, 15, 15T, 15T Pro, 15 Ultra, 17, 17 Ultra, Pad Mini

Model

Devices

Gemini Nano v3

Google Pixel 10, 10 Pro, 10 Pro XL, 10 Pro Fold, Pixel 9, 9 Pro, 9 Pro XL, 9 Pro Fold · Samsung Galaxy S26, S26+, S26 Ultra · Honor Magic 8 Pro · iQOO 15 · Lenovo Idea Tab Pro Gen 2, Legion Tab Gen 5 · Motorola Signature · OnePlus 15, 15R · OPPO Find X9, X9 Pro, Find X8, X8 Pro, Reno 14 Pro 5G, Reno 15 Pro 5G, Reno 15 Pro Mini 5G, Reno 15 Pro Max 5G · realme GT 7T · Sharp AQUOS R11 · Sony Xperia 1 VIII · vivo X200T, X200, X200 Pro, X300, X300 Pro

Gemini Nano v2

Samsung Galaxy Z Fold7, Z TriFold · Honor Magic V5, Magic 7, Magic 7 Pro · iQOO 13 · Motorola Razr 60 Ultra, Razr Ultra 2025 · OnePlus 13, 13s · OPPO Find N5 · POCO F7 Ultra, F8 Pro, F8 Ultra, X7 Pro, X8 Pro · realme GT 7 Pro · vivo X200 FE, T4 Ultra · Xiaomi 14T Pro, 15, 15T, 15T Pro, 15 Ultra, 17, 17 Ultra, Pad Mini

Google's published support list for the ML Kit GenAI Prompt API, read on 20 August 2026. Google add devices over time and this page is a copy, so check theirs if your device is newer than this list.

If your phone is not on that list but you were sure it was supported: Google publish more than one list. The Galaxy S25 and S25 Ultra, for instance, run Gemini Nano perfectly well and appear on Google's list for their other on-device features - summarizing, proofreading, rewriting. The assistant needs the Prompt API, which is what lets it ask a free-form question about your receiver rather than reshape text you already have, and that API reaches fewer devices. The S25 is not on it yet; the S26 is. There is an open request with Google to add the S25 series, and if they do, the assistant will appear on it with no update from us.

What we have run it on ourselves

Separate from Google's list, and shorter, because it only contains hardware we have actually held.

Device	Result
Pixel 10 Pro XL	Tested against a Marantz Cinema 30, on the receiver's own readings

If your device is on Google's list above, the assistant will appear and work; this second table is only about what has passed through our hands. We would rather name one device honestly than imply we have a lab.

Troubleshooting

When it says it cannot help

Each of these is a different sentence on purpose, because each has a different answer.

"This device cannot run the on-device model." The assistant needs Apple Intelligence on iPhone, iPad or Mac, or Gemini Nano through Google AICore on Android. Plenty of otherwise capable devices are not covered, including some current tablets and some flagship phones that run Gemini Nano for other things. See which devices can run it above. Nothing you can change from the app.
The model needs downloading. Android fetches it in the background the first time. It is a one-off, it is large, and it wants Wi-Fi.
"Your receiver has that, but there is nothing to report on it right now." The feature exists and is simply idle - asking what is playing while nothing plays gets this, not a claim your receiver lacks HEOS.
"This receiver does not report that." A genuine answer about your hardware: the chassis has no such feature and the app is not pretending otherwise.
A subsystem could not be read. Unknown, not absent. It will say so rather than treat a failed read as a feature you do not have.
"I cannot change that one from here yet." The assistant's limitation, not the receiver's - the screen for it can.

The assistant is free. Home Automation is the paid part, and the two are separate: you do not need Pro to ask your receiver anything.

Common questions

The assistant, answered

Does the assistant send my data anywhere?

The language model runs on your device and your receiver is on your own network, so a typed question and its answer do not leave your hardware. Voice input is the exception: turning speech into text is done by your phone's own recognizer, and on most phones that means the audio goes to the platform vendor to be transcribed. Typing avoids that entirely.

Why does the assistant say my device cannot run it?

The assistant needs an on-device model: Apple Intelligence on iPhone, iPad and Mac, or Gemini Nano through Google AICore on Android. Plenty of otherwise capable devices do not have one. The app reports exactly which state it is in rather than hiding the feature, so you know whether it is a download, a setting, or hardware.

Will the assistant change things without asking?

It acts directly on power, mute, input and turning the volume down, because all of those are trivially undone by the person in the room. Turning the volume UP always asks first, because that is the one action whose worst case is a damaged driver.

Why does it sometimes say it cannot tell me something?

Because it will not guess. Every figure it gives you comes from a reading taken off your receiver, so if a reading did not arrive it says the subsystem could not be read rather than inventing a plausible number. A wrong answer that sounds right is worse than no answer.

Ask the assistant - the answer comes from your receiver