oneira
OverviewFeaturesCompareSpecsPreorder a Founding Unit · $80
Oneira guides/Physical AI interface

Architecture guide / device to agent

What is a physical interface for AI agents?

Understand the physical AI agent interface pattern: a device starts a request, an app enforces permission, and a configured model, agent, or connected tool does the work.

Preorder a Founding Unit · $80 →See current features →

Product status

What is available now?

  • Working today

    Working today means the expressive face has been demonstrated on a real Oneira device.

  • Connected beta

    Connected beta features require the Oneira app, internet access, a compatible service, and permissions you approve. They can change and may not work with every service.

What you need

Connected beta features need the Oneira app, internet access, a supported service, and permissions you approve. A separate account or fee may also be required.

The short answer

A physical AI interface gives an agent a deliberate place in the real world without pretending the agent lives entirely inside the object. The layers and permissions remain visible.

Real Oneira product resting in an open palm with its exterior visible
Real working product / exterior view
Physical layerInput + output
App layerIdentity + policy
Agent layerConfigured work
Human layerIntent + review

Oneira Dream Big+

A local AI powerhouse in a star.

Small enough to look playful. Built with a dual AI chip, OLED screen, and 1K camera, plus optional connected experiences you control.

Each $80 Founding Unit includes one Oneira device with a dual AI chip, OLED screen, and 1K camera, plus the selected color and included eye set. The 1K camera is included; its final production module and other non-material production details may change before fulfillment. Optional connected services, provider accounts, usage, and third-party fees are separate.

On-device

On-device reactions

Oneira's expressive reactions run on the device, so its face can respond without relying on a cloud request.

See the local core ↗
Working prototype

Dual AI chip, OLED screen, and 1K camera

The working prototype includes a dual ai chip, oled screen, and 1k camera. The 1K camera is included in the $80 Founding Unit; its final production module may change before fulfillment.

Inspect the hardware ↗
Connected beta

Optional connected services

Optional compatible connected services can extend Oneira when you choose. Setup, permissions, internet access, provider terms, and usage costs remain separate. Connected experiences work only with compatible services the owner deliberately configures. Oneira does not connect automatically to every service.

See compatible connections ↗
Connected beta

Compatible connected tools

Compatible connected tools are optional and owner-configured. Nothing is preconfigured, and universal compatibility is not promised.

See connected features ↗
Web beta

Dream Big+ photo generation

The 1K camera supports deliberate, owner-triggered experiences. Optional Dream creation can transform a photo when the owner chooses it.

Open Oneira Create ↗
Developer preview

Focused helpers

Optional helper experiences have narrow, visible limits. Broader research, coding, and workspace actions remain in development.

Understand the agent path ↗

A simple flow

How it works in three steps.

  1. 01
    Oneira

    Captures an intentional input or shows a brief output

    →
  2. 02
    Oneira app

    Checks permission, context, service, and network

    →
  3. 03
    Chosen AI or tool

    Handles the configured task and returns a result

Why physical

A button can be a better boundary than a wake word.

A physical action can clarify when the user is asking, which object is involved, and when the system should stop. That is useful for companions, controllers, accessibility surfaces, and tools used away from a full computer.

The interface should communicate state rather than hide autonomy behind a friendly shape.

Why an app

Secrets, destinations, and consent need a larger screen.

Pairing, authentication, provider selection, permissions, logs, and revocation are difficult to explain on a tiny display. The app is the appropriate place for those durable decisions.

The device can remain focused on quick intent and concise feedback.

Why modular

Change the tool without replacing the character.

When the physical interface is separated from the configured model, agent, or connected service, the owner can choose a different compatible service without turning the hardware into a disposable shell for one provider.

Compatibility still requires testing, and provider rules, accounts, data handling, reliability, and fees remain separate.

Interface design / let the object clarify intent

A physical AI-agent interface should make its boundaries felt before it makes its intelligence felt.

Physical AI interfaces matter when they give an abstract request a clear beginning. A touch, press, movement, or concise on-screen state can make it obvious that the owner is asking a tool to do something. Oneira contributes a current on-device layer—motion-reactive animated expression—then uses the Oneira app for real Connected beta access to optional compatible connected tools and services. The product is not claiming that the agent lives inside the device or that every external service is supported.

This separation is a design feature as much as a technical one. The small device can remain focused on intent and a short response, while the app presents setup, permissions, activity, and revocation controls. Connected services retain their own account rules, availability, costs, and behavior changes. Sight has its own consent-led Connected beta boundary: an owner triggers one visual frame. Everyone present must agree before an owner-triggered Sight capture. Translation and creator assistance are still in development, not an automatic extension of the interface.

Decision framework

Four design tests for a physical agent interface

  1. 01

    Can a bystander tell when the system has been asked to act?

    The activation moment should be legible to the owner and the people nearby. A physical interface is strongest when it avoids ambient, surprising, or invisible requests. The object should communicate that a connected action has started, what kind of action it is, and whether it awaits input, confirmation, a result, or recovery.

Questions, answered

The details that decide a purchase.

Does a physical AI interface run the model locally?+

It can, but it does not have to. Oneira runs its expression layer locally while optional compatible connected requests route through the app to separate services.

Why not use a phone button?+

A dedicated object can make the interaction faster, more legible, and more characterful. The phone remains important for setup, permissions, account management, and complex review.

Can the tool be changed later?+

Potentially, when another service is compatible with the Oneira app and the owner configures it. Universal compatibility is not promised.

What makes the interface trustworthy?+

Clear activation, visible destination, narrow permissions, confirmation for consequential actions, reviewable output, logs, and easy revocation.

Keep exploring

Understand the product before you choose it.

Agent guide / Connected beta, intent before autonomyOptional connected workflows—with bigger work ahead.→Connected beta / optional toolsOptional tools, connected on purpose.→Category guide / portable agentsA portable AI agent interface—not a phone replacement.→

Oneira / current availability

Preorder a Founding Unit · $80

500 Drop 01 units remain across 5 currently available colors. Availability is confirmed when secure checkout begins.

Preorder a Founding Unit · $80 →
oneira

Dual AI chip. OLED screen. 1K camera.

ComparePrivacyPreorder termsSupport
The honest boundary

Oneira's focused helpers have narrow, visible limits. Broader research, coding, and approved-workspace actions remain in development.

  • 02

    Does the app own durable policy decisions?

    A small screen cannot safely explain account choices, service permissions, long output, or revocation options. Oneira uses the app to present Connected beta setup and controls. Evaluate whether that larger-screen setup is clear enough to show what each connection may access.

  • 03

    Can services change without misleading the owner?

    The advantage of a modular interface is that an optional compatible connected tool or service can be selected for a defined job. Compatibility is not universal and can change with the app, credentials, service terms, or network. Make the exact connection testable and replaceable.

  • 04

    Does physical delight preserve review?

    Expression can make a device welcoming, but it must not disguise consequential work. Require explicit confirmation for actions that send, spend, publish, delete, modify, or disclose. Prefer an interface that can say it is blocked or uncertain over one that turns every agent response into an unqualified success animation.

    • Good fit

    Good fit if…

    • You want hardware that makes an intentional request feel more tangible than a generic software button.
    • You value an app-mediated permission and setup layer instead of hidden credentials or blanket tool access.
    • You prefer a small expressive interface for brief tasks while keeping detailed work on a larger screen.

    Choose differently

    Choose something else if…

    • Choose a voice-first system if hands-free ambient interaction is a higher priority than deliberate physical control.
    • Choose a full workstation interface if your agent work needs long context, many panels, or close operational oversight.
    • Everyone present must agree before an owner-triggered Sight capture. Avoid connected visual requests in spaces where that agreement cannot be given freely.