Special Thanks
Special thanks to @henry201605 for continuing to deliver high-quality, well-tested Milvus backend improvements. PR #970 completes the full native search method set from issue #888 — a comprehensive body of work spanning multiple releases.
What's New
feat(milvus): native search_memories, retrieve_with_quality_boost, recall_memory
PR #970 by @henry201605 — completes the medium-priority methods from issue #888.
All three methods now push filters to the Milvus server side:
search_memories— semantic ANN search with tag/type/date filter pushdown in one callretrieve_with_quality_boost— re-ranks results by blending embedding similarity with stored quality scoresrecall_memory— adds temporal recency weighting
Eliminates N round-trip fallbacks to the SQLite base class. 18 new mock-based unit tests in tests/storage/test_milvus_search_methods.py.
fix(hooks): parse Claude Code transcripts as JSONL with nested message
PR #971 — Claude Code transcript format changed from bare JSON objects to JSONL with a message wrapper. The auto-capture hook now parses each line as JSON.parse(line).message ?? line before extracting tool calls, restoring transcript-based memory capture for current Claude Code versions.
Dependencies
- #973 — bump actions/github-script from 7 to 9
- #974 — bump snok/container-retention-policy from 1 to 2
- #975 — bump actions/setup-python from 4 to 6
- #976 — bump uv group with 13 updates
Documentation
- #969 — modernize server commands to
memoryCLI across 23 docs files
Full Changelog
See CHANGELOG.md for the complete entry.
Full diff: v10.61.0...v10.62.0