Native cognitive memory of NQAi

The database that remembers why.

NQAD records every trigger, weighted internal state and decision of NQAi as a generative JSON synapse. Any outcome can be replayed step by step, from the first event to the final choice.

Replay a decision How it works

Native: Lives inside NQAi, not beside it
Generative: Stores the recipe, not the snapshot
Auditable: Every state points to its parent

The gap

Today's AI decides, but cannot show why.

Reasoning lives in transient memory and opaque weights, and is lost when a computation ends. Conventional databases sit outside the AI and were never built to hold cognition.

RelationalFacts in fixed tablesNot the thinking about those facts
VectorEmbeddings for similarityNot cause and effect
GraphRelationshipsNot probabilistic states and collapse
Event logsThat something happenedNot the logic that made it happen

NQAD closes that gap by making cognition itself the stored object, where it matters most: algorithmic trading, banking compliance and risk management.

Core concept

Like a vector drawing, not a photograph.

AutoCAD stores the instructions that draw an image, not the pixels. A synapse stores the generative recipe of an AI state, so it is compact, exact and can be rebuilt at any time.

Synapse: The smallest unit of stored cognition: one JSON file with the algorithm, parameters, weighted candidates and a pointer to the state it evolved from.
Trigger: Any event that changes a status or result: a new input, a market tick, a rule firing. Nothing is written on a schedule, only because something changed.
Brain logic: The chain of linked synapses behind a decision. A complete, replayable path from the first trigger to the final outcome.
Internal states: Confidence, urgency, risk aversion, priority, caution. Stored as numeric weights so they can be measured and audited like any parameter.

Brain logic of an action

Replay one gold trade, synapse by synapse.

A month later, an auditor asks why ATOS bought XAUUSD. NQAD walks the parent chain and re-runs every generator. Illustrative values.

Trigger · step 1 of 4

XAUUSD breaks above pivot

The trend q-cell forks into two candidate paths.

q-cell trend

continuation0.55
false_breakout0.45

confidence: 0.52
risk_aversion: 0.70

syn_7c01.json

{
  "id": "syn_7c01",
  "parent": null,
  "trigger": "XAUUSD breaks above pivot",
  "internal_state": {
    "confidence": 0.52,
    "risk_aversion": 0.7
  },
  "states": [
    {
      "action": "continuation",
      "weight": 0.55
    },
    {
      "action": "false_breakout",
      "weight": 0.45
    }
  ]
}

Synapse lifecycle

Seven stages, from event to archive.

Writing one synapse can re-weight linked synapses and fire new triggers. That bounded feedback loop is what makes the Synapses matrix a living structure, not a stored archive.

01
TriggerAn event changes a status or result in the AI’s environment.
02
ForkAffected q-cells branch into parallel candidate states.
03
WeightEach branch is scored and adjusted by internal states.
04
CollapseBranches resolve into a single recorded outcome.
05
RecordWritten as JSON with generator, branches and *parent pointer.
06
Index & auditClassified, linked into the graph and sealed with a hash.
07
Prune & archiveWeak branches discarded, old chains compressed, lineage kept.

Quantum-inspired

Physics as a model, not qubits.

NQAi borrows three ideas from quantum physics and runs them on classical hardware. A q-cell is a simulated qubit, named so to make clear it is not physical.

Superposition: A q-cell holds several candidate states at once, each with a probability weight.
Entanglement: Linked synapses influence each other: a change in one re-weights its neighbours.
Collapse: A trigger forces the q-cell to resolve into one outcome, which is recorded.

Adaptive Grid

One record, three views.

Each synapse is a JSON file in the AI’s own environment. The index holds pointers; the grid reorganises its layout as usage changes.

Relational: Fields, filters, joins — Reporting and compliance queries
Vector: Embeddings for similarity — Find past situations like this one
Graph: Edges and pointer chains — Lineage and replay

Hardware foundation

Open silicon: RISC-V from the ground up.

No ARM or x86 licences, so the stack stays under NQAi’s control. The Vector extension updates many q-cells in parallel, and a physical quantum co-processor can be added later without changing architecture.

Proposed instruction extension · concept

syn.writeCommit a synapse and its parent pointer
syn.forkBranch a q-cell into candidate states
syn.collapseResolve candidates into one outcome by weight
syn.pruneDiscard branches below a weight threshold
syn.linkUpdate index pointers and graph edges

Positioning

Not a replacement for Oracle. The database for the reasoning about it.

Accounts, ledgers and transactions still belong in conventional systems. NQAD combines event sourcing, knowledge graphs, vector search and lineage tracking into generative records of an AI’s own cognition.

Comparison of NQAD with conventional database types
Capability	Relational	Vector	Graph	Event sourcing	NQAD
Stores	Facts about the world	Embeddings	Entities and relations	Events that occurred	The AI’s cognitive states
Location	External service	External service	External service	External service	Inside the AI
Record type	Values	Vectors	Nodes and edges	Event payloads	Generative recipe
Probabilistic states	No	No	No	No	Yes, q-cells
Writes driven by	Applications	Applications	Applications	Applications	AI triggers
Reproduces a decision	No	No	Partly	Partly	Yes, by replay

Swipe the table sideways to compare.

Deployment in NQAi.cloud

Comes with the platform. Sealed by default.

NQAD is not sold separately. ATOS/NAFTAhub and MCC Bank use NQAi through the cloud, and NQAD records the cognition behind every result, under GDPR and the Swiss Federal Act on Data Protection.

Tenancy: Each client or AI instance gets an isolated synapse space with its own index and keys. No cross-tenant links.
Residency: Synapses stay in the jurisdiction of the RISC-V nodes holding them: Switzerland, EU or Hong Kong.
Access: Applications receive decisions through the NQAi.cloud API. Raw synapses stay sealed inside the AI.
Audit export: A regulator receives the brain logic of one decision as a signed, self-contained package.

MCC Bank: explainable compliance

A payment is screened before release. NQAi forks release, hold for review and reject, weighted by sanctions matches, transaction patterns and counterparty risk. When an auditor asks why, the bank exports that decision's brain logic as a signed package, without reconstructing it from logs.

Growth, bounded by design

Branch explosionMax branches per q-cell, pruning, collapse on deadline
Runaway propagationDepth limits and damping on re-weighting
Replay determinismPinned algorithm versions, recorded random seeds
PrivacyPer-tenant keys, field-level redaction in exports

Roadmap

Six phases to silicon.

Phase 1SpecificationSynapse schema, trigger taxonomy, internal-state model.
Phase 2Reference engineSoftware on off-the-shelf RISC-V boards with Adaptive Grid, index and audit.
Phase 3ATOS pilotRecord live trading decisions in a sandbox; validate replay and export.
Phase 4Hardware accelerationDefine and prototype the NQAD custom instruction extension.
Phase 5NQAi.cloud releaseMulti-tenant deployment with regional residency and audit portals.
Phase 6Quantum co-processorOptional: offload optimisation and sampling to physical quantum hardware.