LAMF 3.0.1

LAMF 3.0.1 repairs the clean Codex Desktop installation path.

• Uses one consistent lamf-memory MCP registration.
• Anchors startup to the permanent runtime directory.
• Marks LAMF enabled and required while leaving operator-only tools approval-gated.
• Installs profile-level startup guidance and the lamf-memory skill without replacing unrelated guidance.
• Performs a live MCP initialize/tools-list check during installation.
• Preserves optimization state; optional modules are not silently enabled.
• Publishes a curated source installer ZIP without the obsolete unsigned beta EXE.

Verified before release: package validation, isolated source install, strict Codex activation, 5/5 final tests, 14/14 smoke stages, 8/8 harness matrix, 10/10 GUI tests, 25 Protocol 3/resource/accessibility tests, and all 12 GitHub Actions jobs across Windows, macOS, Ubuntu, Python 3.10, and Python 3.12.

SHA-256: 4a63d52153999da1418fe6dda554e680d34207fb960300593fbc7e8dcba887d8 LAMF-3.0.1.zip

LAMF 3.0.0 — encrypted metadata and keyed retrieval

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.

LAMF 2.0.0 — Ledgered Agent Memory Fabric

First stable public release of LAMF — Ledgered Agent Memory Fabric, a governed, local-first memory authority for compatible AI agents.

Highlights

• Revisioned memory records with scope and provenance
• Append-only integrity spine and deep verification
• Secret exclusion and fixed security profiles
• MCP/harness registrations for Codex, Claude, Kimi, Grok, OpenClaw, Hermes, and generic MCP, subject to documented capability levels
• Cross-platform installers for Windows, macOS, and Linux
• Core-only operation with no optimization dependency
• Synthetic final, smoke, harness, installer-boundary, and recovery tests

Important limitations

LAMF 2.0.0 is a reference implementation for evaluation and development, not a production-hardening claim. It assumes a trusted host, defaults to localhost/stdio, and does not protect against an attacker with equivalent local access. Keep live data, keys, tokens, logs, vaults, configuration, and exports outside the Git checkout.

LAMF Optimizations is a separate optional product and release. It is not included in core and does not own durable memory.

See installation, demo, threat boundaries, and attribution.