On-device inference
Managed from your phone
Updates itself, rolls back on failure
No open ports, no login shell
Answers that cost no water

Ask all you want. It doesn't cost a drop.

Cloud AI is thirsty. Large data centers cool themselves by evaporating clean water, and a share of every prompt sent to them is paid for that way. Ask the Hub instead and none of that water is spent: the only cost of a question is the electricity it draws while it thinks. Your questions don't add load to a hyperscale facility, and they don't help justify building the next one.

Not one drop

Water it never takes

A thousand questions, a thousand answers, and not a single drop spent cooling a server farm. It keeps the overall environmental impact of using AI far smaller.

Yours

Power

A desktop-class appliance, not a rack in a warehouse. It idles quietly, runs on your own meter, and you can switch it off whenever you like.

Years

Hardware life

New capabilities arrive as signed software, not as a new box. Fewer machines built, fewer machines thrown away.

Anyone can run a model on a mini PC. Keeping a sealed box healthy for years is the product.

Signed updates that roll back on their own, capabilities delivered as signed containers, and a device with no shell and no open ports, all managed from a phone. The intelligence runs on hardware you bought. The operations are ours.

01

It runs on your hardware

Models load and answer on the box you own. No per-token meter, no rate limit. Your assistant works on your own data at your own pace.

02

Nothing to administer

Signed system updates install themselves and roll back automatically if the assistant stops answering. New capabilities arrive as signed containers.

03

Sealed by design

No SSH, no login shell, no inbound ports, nothing to port-forward. The Hub opens an outbound, mutually authenticated connection and accepts work only over it.

04

Set up from a phone

Power it on, enter the pairing code shown on the Hub's display, confirm, and it's yours. Everything after that is one tap.

The current lines.

Every current line runs the same sealed operating system, the same phone experience, and the same update path. The line sets how much the Hub can do at once and how quickly it answers. New capabilities reach every line as signed containers, and final configurations are announced at launch.

Hub Lite
The household Hub

Everyday use, for the whole household.

  • Fast everyday chat, drafting and summarizing
  • Reads your documents
  • Quick assistants that answer the moment you ask
Hub Max
The studio Hub

Studio and small-team workloads.

  • The largest open models available
  • High-resolution image generation
  • Several models loaded and working at once

Set up in three steps.

No accounts to configure on the device, no network settings, no port forwarding, no command line. If someone in your family can set up a wireless router, they can set up an AIRIS Hub, and this is the easier of the two. You do not have to be an engineer. We handle the engineering; you get the assistant.

01

Plug it in

Power and network. The Hub starts itself and shows a pairing code on its own small display.

02

Pair from your phone

Enter the code, confirm the second code the Hub shows back, and the Hub is yours. Nobody else can claim it.

03

Start talking

Pick an assistant from the library and ask it something. Everything after that is one tap.

Assistants, models, and full ownership.

Assistants

Streaming chat, on device.

Custom assistantsWeb search & page readingHistory across devices

Models

A curated library, one tap to install.

Bring any open modelAutomatic GPU accelerationImage generation (Pro & Max)

Ownership

Your device, your data, your call.

Live health & speed readoutsExport or erase anytimeFactory reset from the HubYears of updates

Nothing connects in.

The Hub opens the connection, so there is no port to forward and no service to expose. Your phone talks to the Hub through an encrypted control plane. Models and inference stay in your home.

Encrypted · outbound only · the Hub dials out

Your inputs never train a model

There is no training pipeline in the product and nothing is collected to build one.

No third-party AI provider

The model runs on the Hub, so no outside AI provider ever receives your prompts, your documents, or your clients' names.

The device is sealed

No SSH, no login shell, no inbound ports, nothing exposed to the internet.

Every Hub has its own identity

Each device holds a private key it generated itself and speaks over a mutually authenticated, encrypted channel.

Updates are signed and verified

Checked against a key burned in at manufacture; anything that misbehaves is reversed.

You can take it all back

Export or erase conversations on your own schedule, and reset the Hub to factory state from the device itself.

For practices that can't send work to someone else's servers.

Professional practices have the most to gain from AI and the most to lose from sending their work away. A Hub sits in the office, on the practice's own network, and answers from hardware the practice owns.

The firm's documents stay the firm's.

Draft, revise, and compare documents; summarize long records and filings; search the firm's own document history, with client material under the firm's control.

Built for offices that handle protected information.

Documentation and correspondence, literature and guideline summaries, and patient-facing explanations, with inference on hardware the office owns.

Analysis runs on the practice's own hardware.

Analysis over the firm's own numbers, proposals, memos, reporting, and research assistants that cite their sources.

Come and ask it something.

The fastest way to understand AIRIS is to stand in front of one. Reserve a Hub, or try the owner experience right now in your browser.