🧠 Dual Memory Layer in Vitruvyan

PostgreSQL + Qdrant coordinated by governance orders

🧭 1) What it is

The Dual Memory Layer is the memory substrate of Vitruvyan:

🧱 Archivarium (PostgreSQL) stores canonical, structured truth
🌐 Mnemosyne (Qdrant) stores semantic/vector representations
⚖️ Memory Orders keeps the two layers coherent over time
🏛️ Vault Keepers records audit evidence for traceability
🛡️ Orthodoxy Wardens records epistemic verdicts for governance

In short: one memory for deterministic structure, one memory for semantic retrieval, and a governance layer that keeps them aligned.

🧱 2) What each layer does

Archivarium — structured memory

PostgreSQL is the authoritative source for entities, versions, and relational state. It supports deterministic queries, consistency checks, and governance-grade audit linkage.

Mnemosyne — semantic memory

Qdrant stores vectors and semantic payloads used by retrieval and pattern correlation. It enables similarity search and context expansion across multilingual and unstructured inputs.

⚖️ 3) Role of Memory Orders

Memory Orders is the coordination and integrity order between the two memories. It performs three core functions:

Drift calculation: detects divergence between canonical and semantic state
Health aggregation: computes a unified health view across memory dependencies
Sync/reconciliation planning: produces and, when allowed, executes repair operations

Memory Orders does not define business truth. It protects alignment between truth and representation.

🛡️ 4) Reconciliation and enforcement model

Reconciliation follows a controlled policy model:

dry_run: generate plan only, no data mutation
assisted / autonomous: allow execution when requested

Execution safety is enforced by:

Idempotency key (idempotency_key) to avoid duplicate runs
Run lock per source pair (table + collection) to avoid concurrent collisions
Mass-delete guardrail (threshold policy, explicit override required)
Retry + dead-letter for failed operations (memory.reconciliation.dead_letter)

The design keeps PostgreSQL authoritative and reconciles derived Qdrant state.

🔗 5) Governance coupling

The orders are connected but not merged:

Memory Orders
  ├─ computes coherence/health/reconciliation
  └─ emits audit.vault.requested

Vault Keepers
  └─ persists audit evidence in vault_audit_log

Orthodoxy Wardens
  └─ persists epistemic verdicts in audit_findings

This separation preserves clear responsibilities:

Memory integrity (Memory Orders)
Accountability (Vault Keepers)
Epistemic discipline (Orthodoxy Wardens)

🚀 6) Why this is important

The Dual Memory Layer is not “just Postgres + vectors.”
Its value is architectural:

It keeps semantic retrieval tied to canonical truth
It makes drift visible and governable
It enforces safe reconciliation instead of ad-hoc repairs
It provides auditability and epistemic accountability for regulated contexts

This is what makes Vitruvyan memory governed cognitive infrastructure, not a generic RAG storage stack.

Vitruvyan Docs