KLRSL/dsh-biomemory
Biomimetic memory system for DeepSeek Harness: plain-Markdown data layer, memory tools, frozen-snapshot injection, graded approval gating, structured audits, memory metabolism (dream), memory pins, semantic recall, /memory command, and cross-session retrieval.
Biomimetic cross-session memory for DSH, designed like a human brain: layered memory (L2 structured + L3 semantic), graded human approval for important memories, memory metabolism (half-life decay, reference consolidation, conflict arbitration, cold archiving), SQLite data layer with WAL, offline semantic retrieval (bge-small-zh-v1.5 via transformers.js), frozen snapshot injection at session start, editable in-place entries with conflict surfacing (conflicts pinned to top of list + Knowledge tab red badge), single-entry rollback, deduplication by content fingerprint, and a /memory command with list/query/add/edit/remove/pin/unpin/dream/audit.
Install
dsh plugin add dsh-biomemoryLocal-bundle install per README (no npm package published): dsh plugin add dsh-biomemory (as a local bundle in a DSH profile) or pnpm add link:./dsh-biomemory, then add dsh-biomemory to dsh.profile.bundles. v0.5.2 (2026-08-20): editable memories + conflict surfacing; v0.5.1 extracted the DSH↔desktop-pet bridge into dsh-whale-pet-bridge; v0.5 (2026-08-19) added SQLite data layer + offline embedding (bge-small-zh-v1.5, ~24MB ONNX) + semantic retrieval.
Compatibility
DeepSeek Harness (any profile). Node 24 built-in node:sqlite; transformers.js for offline embeddings — zero external native modules.
Details
- Repo: KLRSL/dsh-biomemory
- Category: Memory
- Stars: 5
- Version: GitHub source v0.5.2 (2026-08-20; no npm package)
- Last push: 2026-08-20
- First seen: 2026-08-15
Recent updates
v0.5.2: edit-in-place + conflict surfacing + single-entry rollback; v0.5: SQLite + semantic retrieval + audit aggregation + dream checkpointing; v0.4: graded approval gate, memory_recall tool, metabolism.
FAQ
- How is memory stored?
- SQLite at ~/.dsh/biomemory/biomemory.db (WAL mode) for structured entries + vector blobs + audit log; legacy ~/.dsh/memory Markdown is imported once and kept as a read-only backup.
- Does it save everything automatically?
- No — ordinary facts auto-save, but important memories (preferences/decisions/lessons) require human approval, and it fails closed when no approval channel is available.
- How does retrieval work?
- Three modes: exact (keyword), semantic (vector), and hybrid (default, Reciprocal Rank Fusion). If the embedding model is unavailable it degrades gracefully to keyword-only.
Alternatives
Aik358/dsh-auto-memory · csyangwen/dsh-memory-evolve · Scorp1o117/dsh-tdai-memory