dearbld/dsh-living-memory

dsh-living-memory stores every conversation and lesson in a local SQLite knowledge base and then actively tends it - deduplicating, merging, decaying unused entries, detecting conflicts and building a knowledge graph - with a nightly patrol and a seven-signal hybrid recall (full-text, optional vector KNN, decay, relevancy, co-occurrence, PageRank and hop) fused by reciprocal rank fusion.

dsh-living-memory captures every conversation, lesson and decision into a local SQLite knowledge base and then actively tends it like a hippocampus rather than a log file. A nightly patrol deduplicates and merges near-duplicate and high-similarity entries, applies temporal decay so unused memories lose retrieval weight (stale todos sink), detects conflicts where a new memory contradicts an old one (flagging both for review instead of overwriting history), extracts entities and co-occurrence edges into a knowledge graph with Personalized PageRank propagation, and writes daily snapshots and an audit trail. Queries fan out into seven signals - SQLite FTS5 full text, optional vector KNN, decay, per-entry relevancy, co-occurrence statistics, PPR graph proximity and hop edge distance - fused by reciprocal rank fusion into one ranking.

Memory ★ 2
View on GitHub ↗

Install

dsh plugin --profile web add dsh-living-memory

Copied verbatim from the README's install line; the README documents the command twice (main and configuration sections).

Compatibility

A DeepSeek Harness plugin whose store and nightly patrol run entirely on the local machine - no telemetry leaves the machine - with an on-disk audit trail, daily snapshots and telemetry panels in the DSH web GUI. Chinese text is first-class via jieba tokenization; the optional vector signal uses local KNN over embeddings only when configured.

Details

Alternatives

863683348/dsh-plugin-focus · akslcw/dsh-negative-ledger · alcheme-labs/dsh-experience-map

More plugins in Memory

Browse more in Memory

Guides for Memory plugins