Deterministic by construction.
Defensible by design.
QGI ships into regulated workflows where a decision has to be replayed, signed, and defended. Every architectural choice in the stack — from quantum-structured encoding to time-aware memory — is made in service of that bar.
This page documents how we operate: the principles the stack is built around, the deployment models available today, how we handle customer data, and the public compliance roadmap.
Four non-negotiables.
Principle 01
Deterministic by construction
Every decision the QGI stack ships is reproducible. Given the same inputs and the same data versions, the reasoning graph replays identically. There is no 'temperature' knob that changes the chain you are defending.
Principle 02
Inspectable before generation
The QAG Engine exposes seven Hilbert-Space Compacting signals — Relevance, Conflict, Overlap, Redundancy, Coverage, Coherence, Topology — as first-class artifacts of the reasoning pass. Conflicts are shown to the reviewer before a decision letter is written, not buried in a paraphrase.
Principle 03
Provenance preserved end-to-end
Every fact, rule, and document carries its source, version, and temporal scope through the stack. Quantum Graph Memory (QGM) extends this across sessions and regulatory replays — a decision made a year ago is replayable against the rule versions in force on that date.
Principle 04
License-safe data
Training data, licensed corpora, and customer data are tracked with explicit licensing metadata. Qualtron specialists are trained on regulated-industry-clean corpora — no uncertain web-scrape provenance, no commercial-content contamination.
Three ways to run the stack.
Regulated customers choose a deployment model up-front. The control plane and data plane separation is explicit so data residency, key ownership, and regulatory residency are all defensible at audit.
Deployment 01
QGI Managed
QGI-hosted service for evaluation, pilots, and workloads where the regulated data itself is not residency-constrained. Single-tenant isolation at the workspace level.
- US region; Canada region on roadmap
- Encryption in transit and at rest
- Per-workspace encryption keys
- Audit log export (JSON / SIEM)
Deployment 02
Customer VPC
QGI control plane; data plane inside the customer VPC on AWS, Azure, or GCP. No regulated data leaves the tenant boundary.
- Customer-owned keys (BYOK)
- Private networking; no public data plane
- Region-pinned to customer policy
- SSO, SCIM, and workspace-level RBAC
Deployment 03
On-premise / Air-gapped
Full-stack deployment inside the customer environment for the most restricted workloads — classified, HIPAA-critical, or sovereign data. Control plane and data plane both customer-hosted.
- Full data sovereignty
- Upgrade cadence driven by customer change window
- Offline model cards and replay tooling
- Supports air-gapped inference
What we do with your data.
Training
Training
Customer data is never used to train QGI models without an explicit, written agreement. Qualtron specialists are trained on domain corpora QGI licenses or owns — not on customer prompts.
Inference
Inference
Customer prompts and responses stay inside the tenant boundary under VPC and on-premise deployments. Under Managed, they are retained only for the workspace's declared retention window and are not pooled across customers.
Replay & audit
Replay & audit
Reasoning graphs are retained per the workspace's audit policy. Export is first-class — JSON-LD, SIEM, or a signed archive — so you can move the audit trail into your own GRC platform.
Deletion
Deletion
Customer-initiated deletion removes prompt / response records and associated reasoning graphs inside the agreed SLA. Model cards and training artifacts live in separate retention per the training agreement.
Detailed policy lives in the current Privacy Policy and Terms of Service. Enterprise engagements are governed by a Master Service Agreement and Data Processing Addendum available on request.
What is committed. What is next.
We publish the roadmap deliberately. Customers evaluating QGI under an AI Governance framework should have a clear, dated view of where compliance artifacts are today and where they will be by the time a pilot moves to production.
Committed 2026
- SOC 2 Type II audit in progress (report target: Q4 2026)
- Model card and audit export standardization across the stack
- Data Processing Addendum and EU SCCs available on request
- Customer VPC deployment (AWS first, Azure next)
2027 roadmap
- HIPAA readiness for healthcare-vertical engagements
- FedRAMP Moderate path for government-vertical engagements
- ISO 27001 alignment
- On-premise / air-gapped reference deployment
Founder statement · On the record
I hold a PhD in formal verification — the discipline of proving, not hoping, that a system behaves correctly. When we say QGI is deterministic by construction, that is not a marketing line. It is the architectural choice I made as the first principle of the company: a regulated decision that cannot be replayed and defended has no business being made by an AI system at all.
— Dr. Sam Sammane, Co-Founder & Chief Scientist, Quantum General Intelligence Inc.
Evaluating QGI under an AI Governance framework?