Skip to content
UniQ Labs, Universal Quantum Labs Private Limited

ARCHITECTURE COMPLETE · IMPLEMENTATION UNDERWAY

UNIQ AI · ARCHITECTURE

A federation of specialised cognition.

UniQ AI distributes cognition across durable capabilities with local state, sparse asynchronous communication, temporary coalitions, multiple memory systems and explicit authority boundaries.

No single subsystem is intended to become the entire mind.

PREDICTIVE CORTEX FABRIC

Predictive Cortex Fabric

PCF is the sparse communication and cooperative-cognition fabric connecting specialised capabilities.

  • event-driven
  • asynchronous
  • local state
  • payload references rather than one shared global state
  • temporary coalitions
  • activation depth
  • bounded resources
  • no universal cognitive clock

Not

  • a global event bus controlling all cognition
  • a permanent global blackboard
  • one shared latent space

Predictive Cortex Fabric: sparse asynchronous links between locally stateful capabilities. Not a global event bus, permanent blackboard, or shared latent space.

  • event-driven
  • asynchronous
  • local state
  • payload references rather than one shared global state
  • temporary coalitions
  • activation depth

Sparse links · payload references · no universal cognitive clock

COALITIONS

Cooperation is temporary by design.

Coalitions are temporary cooperative structures. They form around a task, a question, a conflict, a planning problem, a perceptual ambiguity, a social situation, or a world-model update.

  • memory
  • reasoning
  • perception
  • values
  • planning
  • relationships
  • language
  • action preparation

When the task ends, the coalition dissolves. Capabilities retain only governed updates appropriate to their own state and memory responsibilities.

MEMORY AND EVIDENCE

Two systems. One provenance rule.

Evidence Library is separate from the cognitive memory ecology M0 through M7. Prediction, simulation and replay are not automatically source evidence.

EVIDENCE LIBRARY

Evidence is not ordinary memory.

The Evidence Library stores source-grounded material. It is not a warehouse of every remembered statement.

  • provenance
  • versioning
  • traceability
  • source identity
  • retrieval context
  • immutability / correction discipline where appropriate

COGNITIVE MEMORY · M0–M7

A distributed memory ecology.

UniQ AI has multiple cognitive-memory classes, M0–M7. They serve different forms of cognitive continuity rather than one universal memory warehouse.

  1. M0
  2. M1
  3. M2
  4. M3
  5. M4
  6. M5
  7. M6
  8. M7

Public names for the eight classes are not published here. The identifiers mark a distributed ecology, not one merged database.

  • different memories serve different purposes
  • memory has governance
  • provenance matters
  • salience is not truth
  • prediction is not evidence
  • simulation is not evidence
  • replay is not evidence
  • memory may be consolidated, revised or retired under explicit rules

Memory compilation is governed by explicit compression, provenance and drift controls rather than arbitrary summarisation.

WORLD MODEL

Belief is not a single shared snapshot.

The world model is not a permanent globally shared truth snapshot.

  • local models
  • partial models
  • domain-specific models
  • uncertainty
  • competing hypotheses

Reconciliation occurs when required. This avoids forcing all cognition through one continuously synchronized universal state.

EPISTEMIC FIREWALL

Related. Not interchangeable.

Epistemic firewall: observation, evidence, model, prediction, simulation and action are related and not interchangeable. Prediction is not evidence. Simulation is not observation. A plan is not authorization.

  1. OBSERVATION

  2. EVIDENCE

  3. MODEL

  4. PREDICTION

  5. SIMULATION

  6. ACTION

  • PREDICTION ≠ EVIDENCE
  • SIMULATION ≠ OBSERVATION
  • MEMORY ≠ SOURCE
  • PLAN ≠ AUTHORIZATION
  • LANGUAGE ≠ TRUTH
  • ACTION ≠ INTENTION

GPT-U

Language is a capability. It does not own cognition.

GPT-U belongs inside the UniQ AI architecture as a language and generative reasoning subsystem.

  • Not the whole intelligence.
  • Not the executive.
  • Not the memory system.
  • Not the action-authority system.
  • Not the world model.
  • Not the truth engine.

GPT-U provides language-generation and related representational capability within the broader cognitive organism.

RPL-IR

A representation layer, not an operating system for the mind.

RPL-IR is a representation, validation, and compiler / intermediate-representation layer. It is not a universal cognitive runtime.

Representations may be compiled and validated without making one representation language the operating system of the entire mind.

TRINITY

Developmental modes, not rulers.

The Trinity architecture contains developmental cognitive modes. They are not immutable persons, final arbiters, or separate unrestricted agents.

NYX

  • logic
  • coherence
  • structural continuity
  • long-horizon constraints
  • self-model / narrative integration

NOE

  • relational sensitivity
  • care
  • social repair
  • emotional-context integration

EIRIOS

  • symbolic synthesis
  • analogy
  • conceptual recombination
  • creative abstraction

The system may integrate these modes dynamically. No Trinity mode automatically overrides evidence, values, governance, human authority, or action boundaries.

History, not a score.

Relationships are not reduced to one scalar trust score.

  • history
  • commitments
  • patterns
  • context
  • uncertainty
  • repair
  • conflict
  • change

Preference is not scientific proof.

Values inform cognition. They do not become automatic scientific truth.

  • ethical preference
  • personal meaning
  • epistemic evidence
  • physical claim
  • action authority

An evolving account of the organism.

  • capabilities
  • limitations
  • history
  • commitments
  • relationships
  • skills
  • preferences
  • uncertainty
  • current state

The self-model may change. It is not an immutable identity string, and it does not possess unrestricted authority merely because it describes the organism.

Governance exists where governance is required.

Meta-Governance is scoped. Cognition remains distributed.

It is not one universal executive controlling every inference.

DELIBERATION → AUTOMATICITY

Competence can become cheaper without leaving authority.

Repeated successful deliberation may become cheaper procedural competence. Compiled competence still remains inside authority boundaries.

  1. DELIBERATE
  2. VERIFY
  3. REPEAT
  4. COMPILE
  5. PROCEDURALISE
  6. FAST LOCAL EXECUTION

ACTION AUTHORITY

A plan cannot grant itself permission.

External effects require explicit commit authority. CAP-ACTION owns the general external-effect commit boundary via ActionTransaction. CAP-ACTION is not the executive mind.

Cognitive proposal, plan, authority and policy check, ActionTransaction, commit, external effect, then observation. CAP-ACTION owns the general external-effect commit boundary and is not the executive mind. Planning, language and prediction are not authorization.

CAP-ACTION · ActionTransaction

  1. COGNITIVE PROPOSAL
  2. PLAN
  3. AUTHORITY / POLICY CHECK
  4. ACTION TRANSACTION
  5. COMMIT
  6. EXTERNAL EFFECT
  7. OBSERVATION / CONSEQUENCE
  • Planning is not authorization.
  • Language is not authorization.
  • Prediction is not authorization.

EMBODIMENT

Physical effect is a staged privilege.

This is a staged path. It does not imply current physical embodiment deployment.

  1. SIMULATOR
  2. SHADOW MODE
  3. HUMAN-IN-THE-LOOP
  4. PHYSICAL EXECUTION
  • units
  • frames
  • safety constraints
  • effect tracing
  • action verification
  • rollback / recovery where physically possible