Skip to content

Memory Engine Index

Document ID: MEM-INDEX-001
File Path: docs/06-memory-engine/index.md
Version: 1.0.0
Status: Active
Owner: AI Platform Team
Last Updated: 2026-06-27


This document is the central navigation and architecture index for the Memory Engine in the Wovyr AI Platform.

The Memory Engine is the deployable service that stores, indexes, retrieves, and governs all agent memory. It operates the memory abstraction defined for agents in the Memory System and exposes it as a shared, multi-tenant platform container backed by PostgreSQL, Qdrant, Redis, and object storage.


As with the LLM Gateway and the Provider SDK, the platform separates the abstraction from the operated service.

ConcernMemory System (04-agent-framework)Memory Engine (06-memory-engine)
FormIn-agent abstraction / library viewDeployable service / container
AudienceAgent Runtime authorsAny service over REST / gRPC
ScopeA single agent’s view of memoryAll tenants, agents, and workflows
StateDescribes record shapes & strategiesOwns the durable stores and indexes
GovernanceConceptual permissionsEnforced isolation, RBAC/ABAC, audit
StorageStorage-agnosticConcrete PostgreSQL / Qdrant / Redis / object store

The Memory System defines what a memory is and how agents reason about it. The Memory Engine is where memories actually live and how they are served at scale. See C4 Container §4.4.


Memory Engine
├── Memory API (store / retrieve / update / delete / query)
├── Ingestion Pipeline (validate → embed → index → persist)
├── Retrieval Engine (vector + keyword + graph hybrid search)
├── Ranking Engine (relevance, recency, importance scoring)
├── Knowledge Graph (entities + relationships)
├── Compression Engine (summarization, dedup, token optimization)
├── Storage Layer (Postgres, Qdrant, Redis, object store)
└── Governance (tenant isolation, RBAC/ABAC, audit, retention)

Caller (Agent Runtime / Workflow / Service)
│ REST / gRPC
Memory API ──► AuthN/Z + tenant resolution
├── write ──► Ingestion (validate → embed → index → persist)
└── read ──► Retrieval (hybrid search)
Ranking (score + filter by policy)
Compression (fit token budget)
Return ranked memory set

A detailed lifecycle appears in Overview §6.


DocumentResponsibility
overview.mdService responsibilities, architecture, lifecycle, NFRs
memory-api.mdExternal store/retrieve/query contract (REST + gRPC)
storage-architecture.mdTiered storage across Postgres/Qdrant/Redis/object store
retrieval.mdHybrid retrieval pipeline and strategies
ranking.mdRelevance, recency, and importance scoring
semantic-memory.mdEmbeddings and semantic memory
knowledge-graph.mdEntity/relationship graph and traversal
compression.mdSummarization, deduplication, token optimization

  1. One memory plane — all subsystems read/write through the Engine.
  2. Storage-tiered — hot, warm, and cold data live in the right backend.
  3. Hybrid retrieval — vector, keyword, and graph combine for recall + precision.
  4. Governed by default — isolation, RBAC/ABAC, and audit on every access.
  5. Deterministic & versioned — every memory is versioned and reproducible.
  6. Token-aware — retrieved context is compressed to fit model budgets.
  7. Observable — every read/write emits logs, metrics, and traces.



VersionDateDescription
1.0.02026-06-27Initial Memory Engine Index