seek://map/dsh-memoria
dsh-memoria
#vector #knowledge-graph #memory
TL;DR
A retrieval-engineer's memory: HNSW vectors, FTS5 keywords and knowledge-graph signals fused by reciprocal rank fusion, persisted automatically at turn end and namespaced per agent.
Memory built like a search engine — vectors, keywords and graph edges, ranked together and saved every turn.
Facts
- license
- BSD-3-Clause
- note
- 本次复核：仓库实测 BSD-3-Clause
Key points
- Three retrieval signals fused via RRF ranking
- Automatic turn-end persistence — nothing depends on manual saves
- Per-agent namespace isolation on a local memoria server
FAQ
Why fuse three retrieval signals?
Semantic, lexical and structural recall each fail differently; RRF combination is more robust than any single one.
Do I need to run a server?
It talks to a local memoria server; setup is documented in the repository.
Is there a companion plugin?
Yes — dsh-memoria-extra adds health reports, relation graphs and profile injection.
Official references