doobidoo/mcp-memory-service v10.64.0

v10.64.0 Incremental memory consolidation

on GitHub

Highlights

Incremental memory consolidation (#985, closes #983, contributed by @filhocf)

New time_horizon="incremental" mode for memory_consolidate that processes only memories created since the last consolidation run — safe to invoke from session Stop hooks without blocking the agent.

• DB-based atomic lock (BEGIN IMMEDIATE + locked column) — multi-process safe
• 10-second timeout enforced via asyncio.wait_for at the handler level
• Skips decay and forgetting phases (those stay on monthly/yearly horizons); runs clustering, compression, and association discovery on the incremental window
• Bootstraps with a 24-hour window on first run, then advances created_at > last_run_at
• New consolidation/run_tracker.py module + 15 unit tests
• Relates to RFC #732

Other changes

Fixed

• GET /api/quality/trends AttributeError on all backends (#982, reported by @TonbiLX): the endpoint was calling two non-existent storage methods (recall_by_timeframe, search_all_memories) and returned HTTP 500 on every backend. Fixed by using get_memories_by_time_range with a DB-side BETWEEN filter.

Documentation

• docs/research/contradiction-resolution-approaches.md (#984, contributed by @rudi193-cmd / Sean Campbell): system-neutral comparison of invalidation models, detection mechanisms, and decision guide for RFC #732 implementors.

Maintenance

• Binaries migrated to Git LFS; generated statistics removed from version control
• Stale archive directories cleaned up
• Dead documentation links removed

Known followup

• #986 — on incremental timeout, last_run_at is not advanced; same window can be retried on next run. Two fix shapes outlined in the issue; not blocking for typical usage.

Install / Upgrade

pip install -U mcp-memory-service==10.64.0

Docker images published to ghcr.io/doobidoo/mcp-memory-service:10.64.0 and Docker Hub.

Full changelog: https://github.com/doobidoo/mcp-memory-service/blob/main/CHANGELOG.md