QGI Logo QGI
Coming soon Platform · Memory

Quantum Graph Memory

QGM · Time-aware graph memory for QAG

QGM is the memory substrate for QAG and Neural Symbolic Agents. Every fact, rule, and decision carries its temporal scope and provenance — so reasoning stays consistent across sessions, audits, and regulatory replays.

Coming soon

Quantum Graph Memory (QGM) — waitlist open

Target: later in 2026

The waitlist opens early access for teams already running Q-Prime and QAG pilots. We prioritize workflows that have an explicit 'replay-as-of' requirement from a regulator or auditor.

Notify me or email sam@qgi.dev
Why time-aware, graph-native memory

"Replay as of March 1" is a memory question, not a retrieval question

Every regulated industry has some version of this question. What version of the rulebook applied? Which overlays were in force? What data was present and what was missing? Classical vector memory answers "what is closest to this query now." QGM answers "what was true then."

Classical memory

Flat store. Present tense.

  • Vector similarity against "now" — no explicit time axis
  • Provenance as metadata — losable, stampable, brittle
  • No cross-fact reasoning — dependencies implicit at best

QGM

Graph. Time-aware. Auditable.

  • Validity intervals on every fact — temporal queries are first-class
  • Provenance is a graph edge, not a metadata tag
  • Cross-fact dependencies reasoned by the QAG Engine directly
What QGM delivers

The memory layer, regulatory-grade

Model

Time-aware by construction

Most vector stores know only "now". QGM carries a first-class temporal axis: every fact, rule, decision, and revision is stamped with its validity interval. An overlay effective on March 1 and rescinded on August 1 is not a conflict — it is a temporal fact the memory distinguishes from a rule in force today.

Audit

Provenance preserved end-to-end

Every memory item carries its source, author, and chain of revisions. When a decision is replayed two years later, you see exactly which version of which document and rule produced it — not a best-effort reconstruction from a current snapshot.

Architecture

Graph, not key-value

Memory is a graph over quantum-structured encodings, not a flat vector store with metadata. The QAG Engine reasons over the graph directly, so cross-fact dependencies and contradictions are first-class — exactly the properties a regulator expects.

Agents

Cross-session continuity

Agents built on Neural Symbolic Agents retain consistent memory across sessions without the summarize-and-forget pattern that breaks auditability. You can replay an investigator workflow from a month ago and get the same reasoning graph — not a paraphrased fragment.

Compliance

Regulatory replay

Designed to answer the question a regulator will ask: "Replay the decision as of that date." QGM does not answer from today's memory — it answers from the memory as-of the date in question, including the rule versions, document versions, and conflict states in force then.

Integration

QAG-native

QGM speaks the same seven HSC signals as the QAG Engine. Coverage gaps, conflicts, and coherence breaks propagate naturally through the memory — which means an audit trail is not retrofitted; it is how the memory is built.

Frequently asked

Questions regulated teams ask about graph memory.

What is Quantum Graph Memory (QGM)?
QGM is the memory layer of the QGI deterministic stack: a time-aware graph memory substrate that lets agents, reasoning pipelines, and decision workflows retain consistent state across sessions — with provenance and temporal structure preserved for audit and regulatory replay.
How is QGM different from a vector database or a standard graph database?
Vector databases know only "now" — they collapse all history into a current snapshot. Standard graph databases add relationships but still treat time as metadata. QGM carries a first-class temporal axis: every fact, rule, decision, and revision is stamped with its validity interval. An overlay effective on one date and rescinded on another is not a conflict — it is a temporal fact the memory distinguishes from a rule in force today.
What is "regulatory replay" and why does it matter?
Regulatory replay is the ability to answer the question a regulator, auditor, or investigator will ask two years after a decision: "Replay this as of that date." QGM answers from the memory as-of the date in question — including the rule versions, document versions, and conflict states in force then — rather than reconstructing from today's snapshot. That is the property an adverse-action letter, CFPB inquiry, or FOIA request requires.
How does QGM work with Neural Symbolic Agents?
Agents built on Neural Symbolic Agents (Layer 1) use QGM as their long-term memory. Because QGM preserves provenance and temporal structure, agents retain consistent memory across sessions without the summarize-and-forget pattern that breaks auditability. Replay an investigator workflow from a month ago and you get the same reasoning graph — not a paraphrased fragment.
How does QGM connect to the QAG Engine and Q-Prime?
QGM speaks the same seven Hilbert-Space Compacting signals as the QAG Engine — Relevance, Conflict, Overlap, Redundancy, Coverage, Coherence, Topology. Coverage gaps, conflicts, and coherence breaks propagate naturally through the memory. Under the hood, QGM is a graph over Q-Prime's quantum-structured encodings, not a flat vector store with metadata.
When will QGM be available?
Later in 2026. The waitlist is open now. Teams already piloting the QAG Engine or planning an agent-grade workflow get priority access, and QGM's schema is tuned to the specific workflow before each engagement.

Need "replay as of" in your workflow?

Join the waitlist and tell us the regulatory or audit question you need to answer. We prioritize teams already running Q-Prime or QAG pilots.

Join the waitlist → See Neural Symbolic Agents
Partner with QGI