← Back to home AI-native interaction · A research architecture

Coordination precedes representation.

Nura is a new foundation for AI that works with your attention instead of against it. The intelligence lives in one place — and can appear anywhere: calm, explainable, and always under your control. A working reference prototype — not a product. Yet.

Role
Concept, design & prototype
Project
Nura · a research architecture
Focus
AI-native interaction
Status
Working reference prototype

The premiseSeparate the thinking from the surface — and everything changes.

Most systems treat the screen as the product. Nura treats it as a projection. What endures is the coordination beneath it: your goals, the people involved, the lines that must never be crossed, and what deserves to be remembered. The surface is disposable. The understanding is not.

Because the intelligence is no longer trapped inside a single interface, it can meet you wherever you are — and stay coherent as the moment, the device, and the medium change around it.

The frictionThe interface era optimized for the machine, not for you.

Today's AI inherits four habits worth designing away:

  • It fights for your attentionAI arrives as interruptions — pings, badges, banners that demand the center of the screen whether or not the moment deserves it.
  • It quietly takes overAutomation acts on your behalf and then makes itself hard to question or undo. Agency erodes one convenient default at a time.
  • It's a black boxYou can't ask why it decided what it decided, and you can't see what it remembers about you. Trust is requested, never earned.
  • It's trapped in one shapeThe intelligence is welded to a single chat window or app. Step into another context and the understanding simply disappears.

The shiftThe same problems, answered from the other side.

DimensionToday's AINura
AttentionInterruption on the system's schedulePeripheral awareness on yours
TrustOpaque decisions you must acceptLegible, reversible ones you can question
ContinuityIntelligence tied to one appUnderstanding that persists as surfaces come and go

The ideaOne mind, many surfaces.

The coordination core stays constant; the representation adapts. The same understanding can meet you as a quiet ambient cue, a conversation, or a full dashboard — without rebuilding the intelligence underneath.

Interactive · one core, many surfaces
Ambient
Chat
Graphical
Dashboard
Core

One mind, many surfaces. Four different representations resolve around a single coordination core — the same intelligence, projected wherever the moment calls for it, without rebuilding anything underneath.

The principlesPrinciples you can feel, not features you configure.

Six claims carry the model — each written so it can be tested and, if wrong, fail honestly.

01

Calm by default

Nura lives at the edge of your attention and earns the center only when something genuinely matters. Awareness without interruption.

02

Agency is recoverable

You can always inspect, override, and correct. And when Nura isn't sure, it doesn't decide for you — it hands the choice back.

03

Safety is structural

Some rules can never be broken — not even to produce a "better" result. Guardrails are part of the foundation, not a patch bolted on later.

04

Explainable & private

Ask "why" and get a real answer. What Nura remembers is always visible, and always erasable. Nothing about you is hidden from you.

05

One mind, many surfaces

The same understanding can meet you as a quiet ambient cue, a conversation, or a full dashboard — without rebuilding the intelligence underneath.

06

Built as science

Nura is defined by claims it could fail, then tested against them. An explanatory model held to evidence — not a demo dressed as a promise.

The proofWe handed a real job to an AI — and stayed in control.

To find out whether the foundation actually holds, it was pointed at a domain it was never built for: software development. A developer delegates a repository-wide accessibility fix — alt text, labels, contrast across dozens of files — to an AI coding agent, and every principle had to earn its place, live. It runs as five moments of coordination, human-in-the-loop from the hand-off to the merge gate.

Case study · Delegate → Merge gate
01 · Delegate the workHuman stays the decision-maker
02 · See the plan — and its doubtPer-step confidence, shown not hidden
03 · Inspect, don't just trustChanges + evidence, all checkable
04 · Correct one assumptionOverride in place, no starting over
05 · Decide whether to mergeApproval + green checks required

Five moments of coordination. The work flows down the connected wire from an explicit hand-off to a hard merge gate — and at every rung the human can see, inspect, correct, and decide.

Revised

The coordination held. The theory grew — carefully.

The full workflow ran end to end with no changes to the core, in a domain it was never designed for — real cross-domain robustness. It also surfaced one honest limit, and rather than paper over it, the evidence set the agenda for what comes next. That discipline is the whole point: Nura is a foundation held to claims it could fail, not a promise dressed as a demo.

See it liveBuilding the calm layer for AI.

The prototype runs this exact coordination live — and swaps between surfaces without rebuilding the intelligence underneath. Try it, run the study, or start a conversation about what comes next.

One-line takeawayCoordination precedes representation.

Nura is a calm, explainable, always-under-your-control foundation for AI — one coordination core that can appear as any surface, held to claims it could fail rather than a demo dressed as a promise.

Explore more work

Onboarding & Adoption →