Docs hub for onboarding, architecture, retrieval, and operations.
Start here: ../README.md
../README.md— product story + quick startfeatures/update-guide.md— update MemSensefeatures/no-docker-quickstart.md— advanced macOS / Linux setup without Dockerfeatures/local-bge-oneclick.md— local embedding setupfeatures/api-smoke-test.md— verify the system is actually working
features/architecture-overview.md— the full system flowfeatures/retrieval-algorithm.md— ranking and selection logicfeatures/embedding-search.md— compact retrieval/storage summary
features/dashboard-rbac.md— dashboard surfaces and access controlfeatures/worker-retry-dlq.md— async jobs, retry, DLQfeatures/api-smoke-test.md— runtime verification
MemSense can be understood in four layers:
- capture — online interaction becomes memory chunks
- enrichment — embeddings, tags, and memory semantics are added
- retrieval — candidates are recalled with semantic + lexical signals
- selection — results are reranked for relevance, time-awareness, and diversity
That is the system meaning of:
From agent history to living memory.
| Area | Page | What it covers |
|---|---|---|
| Product | ../README.md |
Vision, story, quick start |
| Setup | features/update-guide.md |
Update MemSense |
| Setup | features/no-docker-quickstart.md |
Advanced macOS / Linux startup without Docker |
| Setup | features/local-bge-oneclick.md |
Local embedding service |
| Architecture | features/architecture-overview.md |
System flow and data movement |
| Retrieval | features/retrieval-algorithm.md |
Dual recall, scoring, MMR-style selection |
| Retrieval | features/embedding-search.md |
Compact search/storage summary |
| Ops | features/dashboard-rbac.md |
Dashboard surfaces and RBAC |
| Ops | features/worker-retry-dlq.md |
Worker reliability and job lifecycle |
| Validation | features/api-smoke-test.md |
Runtime verification |
README.mdsells the vision and gets people to try it.docs/README.mdtells people where to go next.docs/features/*.mdexplains one concrete part of the system at a time.