Skip to content

v0.6.0 — procedural type, bi-temporal supersede, hybrid search & link suggestions

Choose a tag to compare

View all tags
@cremenescu cremenescu released this 15 Jun 16:56
v0.6.0
e09621c

Six things land in 0.6.0 — sharper records, smarter search, and human-confirmed link suggestions. All optional, all keep the zero-dependency default.

New

  • procedural memory type — reusable multi-step runbooks/workflows (release steps, a recovery drill), distinct from one-shot commands.
  • Bi-temporal supersede — superseding now keeps when a fact stopped being true (invalidated) and why (invalid-reason), separate from created (valid-from). Full "what did we believe, and when" history instead of a bare tombstone. mem.py supersede <old> --by <new> --reason "...".
  • files field — attach the paths a memory is about (--files "a.ts, b.ts"); indexed for search and shown as chips.
  • Recency-weighted ranking — mem.py search nudges fresher memories up among near-ties without overriding a stronger keyword match (MEM_RECENCY_WEIGHT, default 1.5).
  • Optional hybrid semantic search — with Ollama + a tiny embedder (all-minilm, ~40 MB), search fuses keyword scores with cosine similarity over locally-stored vectors, so "auth token expiry" finds "JWT TTL". Retrieval-only — it never decides or writes, so it stays clear of the trust gate. No Ollama → silent keyword-only fallback. mem.py embed builds the vectors into store/.embed.db.
  • Semantic link suggestions — the Links page now offers the closest unlinked memory pairs as suggested related-to edges; you confirm or dismiss each one. Nothing is linked automatically (MEM_SUGGEST_THRESHOLD, default 0.62).

Notes

  • Vectors live in a separate derived file (store/.embed.db) — delete any time, mem.py embed rebuilds.
  • Embedder is gated cleanly: small models (all-minilm ~256-token context) return HTTP 500 on long text, so embedding truncates + retries on shorter slices instead of silently skipping records.
  • The two-parser contract (mem.py ↔ PHP web UI) still holds: conformance passes on 396 records.

Install: claude plugin marketplace add cremenescu/mem0ry4ai then claude plugin install mem0ry4ai@mem0ry4ai. Full docs in the README.

Assets 2
Loading