Skip to content

LAMF 3.0.0 — encrypted metadata and keyed retrieval

Choose a tag to compare

@AI-LUCI AI-LUCI released this 15 Aug 13:23
1be51c5

LAMF 3.0 hardens the local memory authority and restores 2.x-class retrieval performance without changing frozen MemoryBench quality.

Highlights:

  • encrypted bodies, titles, tags, and entities at rest
  • keyed 96-bit HMAC contentless FTS5 index with native BM25 ranking
  • explicit, backed-up v2 and legacy-v3 migration plus automatic derived-index recovery
  • enforced context budgets, bounded work queues, HTTP auth throttling, sanitizer/ACL/dependency/installer hardening
  • six independently switchable optimization modules; all 64 subsets benchmarked with confirmation repeats
  • portable Python 3.10/3.12 behavior across Linux, macOS, and Windows

MemoryBench (three fresh 19,195-document stores, 500 frozen queries): quality exactly matches the established 2.x reference; ingest is +32.4%, disk amplification -27.7%, and p50/p95/p99 query latency -36.5%/-37.1%/-36.9%.

Validation: 41/41 protocol/runtime tests, 14/14 smoke stages, 5/5 final deployment tests, all 8 package checks, and the complete public 12-job OS/Python matrix. See docs/RELEASE_3_0.md for methods, security boundaries, and qualifications.