Skip to content

Local + cloud

On-device first. Cloud when it earns it.

The question every serious user asks is where their work goes. Nova's answer: on-device by default, with optional cloud reasoning you enable per task. Here's how the pieces fit — and where your data does and doesn't travel.

  • Local models

    In preview

    On-device models run the everyday loop — perceiving, retrieving memory, and reasoning about routine work — so your data can be understood without leaving the machine.

  • Optional cloud reasoning

    In development

    For the harder problems, Nova can call a larger cloud model — opt-in, per task, encrypted in transit, used once, and never retained or trained on.

  • The router

    In development

    Nova decides whether a task can be handled locally or genuinely needs the cloud — and tells you before it reaches out, so the choice is never hidden.

  • Bring your own models

    Planned

    Running your own local models, or connecting your own provider and keys, is part of the model-orchestration direction.

The decision

How Nova picks local or cloud.

Every step of reasoning runs through the same choice. Local is the default; the cloud is reached for only when a step genuinely needs it — and only with your permission.

The routing decision
A task arrives
from you, or noticed
Router
can this stay local?
Local model
the default — on-device
Cloud, if needed
opt-in · named · once
The default path never leaves your machine. The cloud path is opt-in, per task, and named before it happens.

Where your data goes

Follow the data.

Two paths. Most work never takes the second one.

On-device · the default
  1. Nova perceives your work locally.
  2. It retrieves relevant memory from the local index.
  3. A local model reasons and plans.
  4. It acts — and nothing left the machine.
With cloud · opt-in, per task
  1. A task exceeds what local reasoning can do.
  2. Nova tells you it wants to use the cloud.
  3. Only what the task needs is sent, encrypted.
  4. The result comes back; nothing is stored or trained on.

Local reasoning is in preview today; the router and cloud handoff are in active development, and bring-your-own models are planned. See Architecture for how orchestration fits the stack, and Privacy for the guarantees.

Questions

Local + cloud, answered plainly.

Which model does Nova use by default?
A local, on-device model — always, for the everyday loop. The cloud is never the default; it's a path Nova asks to take only when a task needs it.
Will Nova ever use the cloud without asking?
No. Cloud reasoning is opt-in, per task, and named before the call is made. If you don't allow it, it doesn't happen.
Can I turn the cloud off entirely?
Yes — you can keep Nova fully local. Some of the hardest tasks may be beyond what an on-device model can do alone, and Nova will tell you that plainly rather than reach out behind your back.
Is anything sent to the cloud used for training?
Never. An opt-in call is used once to produce a result, encrypted in transit, and not retained. The full guarantees are on privacy & security.
Can I bring my own models or provider keys?
That's planned, not shipped. Running your own local models, or connecting your own provider and keys, is part of the model-orchestration direction — we'll mark it clearly when it's real.

See it do real work.

Nova is in private preview. Request access, and we’ll write when it’s your turn.

Request access