Back to Signal Feed
CodeTracked since May 19, 2026

Persistent HNSW checkpoints make memory restarts much faster

The release makes @claude-flow/memory recover ANN state from disk snapshots: reopening the same database path now restores entries and the HNSW index instead of rebuilding, removing restart-time index reconstruction overhead.

@claude-flow/memoryMemoryServiceHNSWdbPath sidecar snapshot

What Happened

  • The release makes @claude-flow/memory recover ANN state from disk snapshots: reopening the same database path now restores entries and the HNSW index instead of rebuilding, removing restart-time index reconstruction overhead.
  • The release makes @claude-flow/memory recover ANN state from disk snapshots: reopening the same database path now restores entries and the HNSW index instead of rebuilding, removing restart-time index reconstruction overhead.
  • 1 evidence item attached for review.

What is Different

Before

Scattered source updates, isolated context, and manual follow-up across multiple feeds.

Now

Added durable persistence for the HNSW index in MemoryService (including auto-snapshoting and restore on reopen), which changes restart behavior from rebuild-on-start to snapshot restore.

Why Track This

Why It Matters

Developers and operators can restart services that depend on the memory service and get vector search usable almost immediately, instead of losing startup time to rebuilding index state from all stored entries. This reduces restart/incident recovery delays and improves operational readiness after node recycle; continue monitoring snapshot write/read integrity, recovery failures when storage is interrupted, and sidecar file growth relative to write rate.

Impact

Developers and operators can restart services that depend on the memory service and get vector search usable almost immediately, instead of losing startup time to rebuilding index state from all stored entries. This reduces restart/incident recovery delays and improves operational readiness after node recycle; continue monitoring snapshot write/read integrity, recovery failures when storage is interrupted, and sidecar file growth relative to write rate.

What To Watch Next

  • Watch whether @claude-flow/memory becomes a repeated pattern.
  • Track follow-up changes around Hybrid Search.
  • Compare future signals against this evidence trail.
  • Re-check risk flags: snapshot_corruption_on_crash, meta_hnsw_version_mismatch.
Open Topic TimelineOpen Technical EventOpen Original Sourcesnapshot_corruption_on_crash / meta_hnsw_version_mismatch / autosnapshot_disk_growth / restore_consistency_regression

Supporting Evidence

GITHUB RELEASEHigh Trust

v3.7.0-alpha.71 — ADR-125 Memory Consolidation (full delivery)

The note states `MemoryService.close()` snapshots HNSW to `<dbPath>.hnsw` with `<dbPath>.meta.json`, and the same path reopen restores entries and HNSW so a cold-start rebuild is no longer required.