An astronaut at an ordinary garage workbench under an uncanny green light

NATIVE ANDROID APP / CLOSED ALPHA

Your phone
runs the agent.

Atlas is an agent runtime that lives on your Android phone.

It keeps an AI session running across voice, camera, movement, interruptions, and time. The model can change. Your context stays with Atlas.

SEE HOW ATLAS WORKSPLAIN ENGLISH ↓

ANDROID APP.
PERSISTENT SESSION.
REPLACEABLE MODELS.

A—01 / WHAT THE APP DOES

One agent session.
Running through
your actual day.

01 / PERCEIVE

It can look and listen.

Atlas can use the phone’s camera, microphone, and device state when a task calls for them.

02 / REMEMBER

It keeps the thread.

Observations, decisions, interruptions, and useful context stay attached to the ongoing session.

03 / ACT

It can use tools.

Atlas routes requests through explicit capabilities and policies instead of letting a model quietly do whatever it wants.

It is a runtime, not a chatbot skin.

ONE EXAMPLE / ONE CONTINUOUS SESSIONPHONE MOVED. CONTEXT DIDN’T.
You

Remind me which bag has the charger.

Atlas

The canvas one. You put it beside the door.

You

Wait—what was the score?

Atlas

Phillies are up 4–2. Back to packing?

You

Yeah. Where were we?

Atlas

You still need the charger and rain shell. The rest is in the car.

THE APP OWNS THE THREAD. THE MODEL ANSWERS INSIDE IT.

ATLAS CORE / DEVICE OWNED

The model is
not Atlas.

The native app owns memory, camera and voice input, session state, confidence and freshness checks, tool permissions, retries, and an event history you can inspect. It sends the right job to an AI model, then keeps the useful result. You can swap inference providers without replacing the agent.

Atlas Core diagram showing voice, memory, vision, tools, freshness, and policy converging into an active, device-owned core

THE RUNTIME STAYS.
THE MODEL IS REPLACEABLE.

An Android phone.
A model endpoint.
Something real to do.

Atlas is useful when you treat it as a persistent agent, not a question box. Start a live context session, give it access only to the capabilities the task needs, and let it follow a project, errand, repair, walk, or routine over time.

  1. 01

    Install Atlas

    The runtime runs natively on Android and stays tied to the phone you already carry.

  2. 02

    Connect inference

    Use a compatible model on the phone, on your local network, or through a cloud API. Managed Atlas inference is coming later.

  3. 03

    Choose its access

    Enable camera, microphone, and tools only when they make sense for what you want Atlas to follow or do.

  4. 04

    Keep one session alive

    Talk naturally, point the camera, change subjects, correct it, and come back later. Continuity is the point.

Different jobs can use
different models.
Atlas stays the same.

A quick response, image analysis, and careful reasoning do not need the same model. Atlas routes each job to a suitable endpoint: local, a machine on your LAN, or a cloud provider. The provider returns an answer. Atlas keeps the session.

MANAGED ATLAS CLOUD INFERENCE / ON THE WAY

FAST / 01VISION / 02REASON / 03FALLBACK / 04

Try it on
something that
takes time.

Atlas is being built for work and life that do not fit inside one prompt. If you want to test the native app, tell us what you would want an agent on your phone to follow, remember, or help you do.

Small Android cohort · bring your own inference · direct founder contact