v1.0.3 v3.42.4
tobychui/zoraxySep 17, 2026 → Sep 17, 20260 days between releases
What's New
memory_search now exposes raw scores in similarity and adds rankingScore for fused results
Breaking Changes
- Code expecting similarity to contain RRF fusion scores will receive raw relevance scores instead.
New Features
- New rankingScore field exposes the fused SmartRetrieval ranking score.
Improvements
- Added 14 new regression tests for SmartRetrieval package and MCP handler.
Bug Fixes
- Fixed memory_search reporting RRF fusion score in similarity instead of raw relevance.
Upgrade Notes
Applications parsing the similarity field must be updated to expect raw scores; use rankingScore for fused scores. Existing logic relying on the previous similarity value will need adjustment to avoid incorrect results.
The Two Releases
v1.0.3
Sep 17, 2026· 1 month ago
This release introduces configurable TLS controls and process group cleanup fixes in the sandbox daemon, while significantly enhancing the Python SDK with strict type safety and adding comprehensive enterprise blueprints for fleet management and network policies.
Full release notesv3.42.4 — smart search score semantics fix
Sep 17, 2026· 1 month ago
Fixes a bug in `memory_search` where the `similarity` field was incorrectly reporting an internal RRF fusion score instead of the raw retrieval relevance score. The `similarity` field now correctly holds the raw score, and the fused score is exposed as a new `rankingScore` field.
Full release notes