Enia on device

Personal AI should become more useful without becoming less yours

Enia on device is a personal AI designed to run on a device you control, learn only from what you choose to teach it and carry useful context across time. The working prototype proves the core interaction. Persistent memory, teachable skills and broader distribution are being built in clearly labelled stages.

Enia on device features

Working prototype · Local-first design · No conversation-data training · Roadmap shown honestly

What you can expect today

What works today, and what is still being built

This is the honest picture before any feature claims below. We use the same status words on every Kvantia page.

  • Working prototype

    Core local conversation

    The core runs and can be demonstrated on supported test hardware.

  • Working prototype

    Grounded response process

    The system can use its knowledge structure and express uncertainty. This does not guarantee every answer is correct.

  • In development

    Persistent memory across sessions

    The long-term memory system is being built and tested. We do not promise unlimited memory today.

  • In development

    User teaching and reusable skills

    The teaching flow is being developed after persistent memory.

  • Planned

    Skill marketplace

    Catalogue examples are product direction, not products for sale today.

  • In development

    Desktop, mobile and server distribution

    Each platform has its own status — see the device section below. Nothing is a finished download yet.

Your device is the home

Your personal AI should not require your private life to become cloud input

Benefit. Enia on device is designed to run its personal AI workload on a device you choose. The aim is simple: your conversations and personal memory stay in your environment instead of becoming a remote service’s conversation history.

Example. Teach Enia on device how you organise a project, what decisions were made and which details matter. That personal context stays associated with the instance on your device.

How it works

  • the core runs locally on supported hardware;
  • conversation and memory data are stored locally;
  • account and licence services are separate from conversation processing;
  • a periodic licence check may require a connection;
  • optional backup or external integrations are described separately when they exist.

Boundary. Local-first does not mean the device never connects to anything. Account, licence and future optional services may use the internet. The product shows what is sent before those services are enabled.

Data boundary

What stays on your device, and what the account service holds

On your device

  • Conversation
  • Local AI core
  • Personal memory
  • Teaching data
  • Local settings

Not sent as account data

  • Conversation text
  • Personal memory content
  • Local teaching material

Persistent memory

Keep useful context without pretending that every detail deserves forever

Benefit. Enia on device is being built to remember selected facts, preferences, projects and decisions across sessions. You should be able to inspect, correct and remove what it remembers.

Example. Instead of explaining your project, preferred working style and previous decision every week, you continue from the context you chose to keep.

Boundary. Persistent memory is not perfect recall. It can be incomplete, outdated or misinterpreted. Important decisions still need source material and human verification.

Before this feature can be called available, it must ship its controls: a memory viewer; edit and correct; delete one item; delete all memory; a source or date for each memory where possible; a clear export and backup policy; and defined behaviour when a subscription or licence ends.

Memory lifecycle

How a memory is made, kept and controlled

Persistent memory is in development. This is how it is being built to work.

  1. A conversation produces possible memories

    Candidate memories are proposed, not silently stored.

  2. Temporary context is separated from what is worth keeping

    Not every sentence deserves to be permanent.

  3. Saved memories connect to topics, people, projects or decisions

    Links make later retrieval meaningful.

  4. Relevant memories are retrieved when a new request needs them

    Context returns when it is useful.

  5. You review, correct, pin or delete

    The person stays in control of what is kept.

The last stage feeds back into the first, and the loop continues.

Controls at every stage: inspect · correct · pin · forget.

Your context, by choice

Teach it what matters to you — without handing over everything

Benefit. Enia on device is designed to learn from deliberate teaching: a correction, an example, a document or a repeated preference you choose to keep. The goal is a personal system shaped by use, not a global model trained on private conversations.

Example. Show Enia on device the structure you use for trip planning, the constraints you always apply and a good previous result. The next plan starts closer to your way of working.

Boundary. “Learns from you” must never mean every sentence silently becomes permanent. The interface keeps chat context, saved memory and an installed skill visibly separate. Teaching is in development.

Teaching loop

How teaching is meant to flow

  1. 01

    Show an example

    Give Enia on device a concrete instance of what you want.

  2. 02

    Explain what matters

    Point out the constraints and details that count.

  3. 03

    Enia on device proposes what it learned

    It shows back its understanding before keeping it.

  4. 04

    You confirm, correct or discard

    Nothing is saved without your decision.

  5. 05

    It applies in a future task

    The next result starts from your confirmed way of working.

Confidence should be earned

A better answer can also be an honest “I don’t know”

Benefit. Enia on device is being designed to connect an answer to what it currently knows and to expose uncertainty when that knowledge is weak or conflicting.

Example. Ask about a project decision. Enia on device should distinguish a remembered fact, a conclusion and a missing source instead of presenting all three with the same confidence.

What the interface should show

  • known from saved memory;
  • inferred from current context;
  • based on a user-provided source;
  • uncertain or conflicting;
  • not known.

Boundary. Grounding reduces unsupported answers; it does not make errors impossible. Enia on device can still misunderstand, retrieve the wrong context or reason incorrectly. High-impact decisions require verification.

Choose where it lives

Desktop, mobile or home server — only when that build is ready

One personal AI, on a device you control. Each option shows its own status. None is a finished download yet.

Reusable skills

Start with a capability, then adapt it to your way of doing things

Status: Planned. A skill is intended to package a defined body of knowledge, examples and behaviour so Enia on device does not need to be taught from zero. After installation, the skill can continue adapting to your confirmed preferences.

A skill card will eventually show

  • what the skill knows and does;
  • what data it can read;
  • whether it can connect externally;
  • author and version;
  • evaluation and limitations;
  • licence;
  • uninstall and delete behaviour.

How a skill layers up: Enia on device core + an installed skill (defined knowledge and behaviour) + your confirmed preferences and examples = a capability adapted to you.

Boundary. Marketplace examples are not current offers while the marketplace is planned. When it opens, skills carry their own status and licence — there is nothing to purchase here today.

Useful within local limits

Local AI should be honest about the trade-off

Enia on device is not trying to be the largest general-purpose cloud model. It is designed around a smaller personal system that can live close to its user and become useful through context, memory and teaching.

Local personal AI Large cloud model
Personal data can remain on owned hardware Requests are processed in provider infrastructure
Capability is constrained by the device Access to very large general models
Can be shaped around one person Same base service used by many people
Works offline for confirmed local features Usually depends on a connection
You manage hardware, storage and backup The provider manages infrastructure

Boundary. Enia on device may not match a frontier model on broad coding, writing or world knowledge. The proposition is personal continuity and control, not winning every benchmark.

Personal outcomes

Start with continuity, not a list of AI tricks

Each outcome is labelled by the capability it depends on, so nothing here reads as available before it is.

Continue a long-running project

Keep approved decisions, vocabulary and working preferences connected over time. Requires persistent memory — in development.

Build a private knowledge companion

Teach selected personal material and ask questions against that context, without it becoming part of a central conversation service. Exact source and file support will be confirmed.

Practise a skill over time

Use repeated examples and corrections so the experience develops with you. Depends on teaching and skills — in development and planned.

Use a personal AI where connectivity is limited

Continue confirmed local features without a constant cloud-model connection. A periodic licence renewal still needs an occasional connection.

Different by design

Different by design, not by a longer feature list

Question Typical cloud assistant Enia on device direction
Where is inference performed? Provider infrastructure On supported user-controlled hardware
Who holds conversation memory? Provider account or service The local Enia on device instance
Does every conversation become permanent? Depends on service settings No — memory should be visible and controllable
Can the user inspect learned context? Often limited Memory controls are a product requirement
Does it require internet? Usually Local features are offline-first; licence and optional services may connect
Is it the largest general model? Often a frontier model No — it optimises for personal continuity and control

This compares a typical cloud assistant; it does not claim every provider trains on all user conversations.

Honest limits

A personal AI is more trustworthy when its boundaries are visible

  • It is not guaranteed to be correct.
  • It is not a search engine or a continuously updated copy of the web.
  • It is not yet a finished marketplace of expert skills.
  • It is not the largest cloud model running on a small device.
  • It does not remember perfectly or understand intent perfectly.
  • Local operation does not remove your responsibility for device security and backup.
  • A working prototype is not the same as a generally available product.

Questions people ask

Features FAQ

Does Enia on device send my conversations to Kvantia?

The intended architecture keeps conversation and personal-memory processing on the device. Account, licence and billing data use the account service. This answer is updated if crash reports, support exports, backup or optional integrations transmit content.

Does “grounded” mean Enia on device never makes mistakes?

No. Grounding and uncertainty states are designed to reduce unsupported answers and make limits clearer. Errors remain possible.

Can it remember everything?

No product should promise perfect or unlimited memory. Enia on device is being built to retain useful, user-controlled context across time.

Can I teach it without coding?

That is the intended experience: examples, corrections and explicit memory or skill controls. The exact teaching capabilities carry their current status.

Is Enia on device the same model as Enia?

No. Both belong to the Enia direction, but they are different architectures. Enia is a classic transformer model rebuilt from an open-source model base and designed for further training. Enia on device follows a separate personal, local design.

More about Enia

Is Kvantia AI Office part of the same company?

Yes. Kvantia AI Office is the company’s business product for supervising AI work with rules, approvals and an audit trail.

Follow the product as it becomes real

Join early access with the right expectations

Tell us which device you want to use and what you want your personal AI to remember or learn. We invite people in stages as each capability meets the status described on this page.

Early access to the working prototype — not a finished product. No sensitive personal material needed to join.