Skip to content

MentisDB 0.10.6.51

Choose a tag to compare

View all tags
@gubatron gubatron released this 16 Aug 00:48
· 1 commit to master since this release
0.10.6.51
0293e96

MentisDB 0.10.6.51

Permanent skill delete, Skills summary counts, vector sidecar WAL, and search persist wins.

Date: August 15, 2026

Upgrade

cargo install mentisdb --locked --force

Or let the daemon self-updater install tag 0.10.6.51. Restart the daemon after install.

Existing vector sidecars keep working. The next append starts a sibling *.json.wal. Older binaries that only read the JSON snapshot treat a sidecar with a pending WAL as stale and rebuild.

Skill delete

Revoke still keeps every version for audit. Permanent delete is the explicit exception to the registry’s append-only lifecycle: SkillRegistry::delete_skill removes the skill and all versions from mentisdb-skills.bin. After delete, the same skill_id can be uploaded again as Active.

Surface How
Crate SkillRegistry::delete_skill(skill_id)
REST POST /v1/skills/delete { "skill_id": "…" }
MCP mentisdb_delete_skill
Dashboard DELETE /dashboard/api/skills/{id}

Crate, REST, and MCP delete any existing skill. The dashboard UI matches bearer-token delete:

  1. Revoke on active/deprecated skills (audit row stays).
  2. Delete on the revoked list row, or Delete Permanently… on the detail pane (type the skill id to confirm).

The Skills page summary bar shows total / active / revoked / deprecated (SkillRegistry::counts()). Deleted skills are absent and are not counted.

Vector sidecar WAL

Each append writes one integrity-chained WAL record (MDBVWAL1, digest SHA-256(prev \|\| entry)) instead of rewriting the full JSON snapshot. Load verifies the snapshot, then replays *.json.wal. After 32 records the snapshot is compacted and the WAL is removed.

The thought log remains canonical. A corrupt WAL fails verification and the sidecar is rebuilt.

Persist and search

  • MCP/REST append releases the chain write lock before sidecar WAL and overlay I/O. Crate append_thought still flushes derived files before return.
  • Implicit-edge first build: pairwise for N ≤ 128; temporary HNSW top-k above that.
  • Dashboard reopens a cached chain only when on-disk thought count is ahead of the live handle.
  • query_ranked uses search_filtered on the cached Exact/HNSW backend.
  • Exact cosine top-k uses select_nth_unstable_by.
  • Caches: adjacency by thought count, process-wide Porter stemmer, thesaurus &[String], HNSW ef_search, webhook reqwest::Client, bearer-token registry mtime.
  • Skill persist encodes the live map by reference; SkillVersion.schema_version is stored.

Verification

  • make clippy (-D warnings)
  • cargo test --all-features
  • cargo package --features local-embeddings
  • Criterion search_ranked smoke
  • Full LoCoMo / LongMemEval not re-run: fusion weights unchanged. Overlay first-build is approximate only for N > 128.

Links

Assets 6
Loading