Ships Phase 2 of the ROADMAP (memory namespaces #16, cold tier search + restoration #14, memory budgeting, adaptive sleep advisory) and closes out related tracker issues. Also fixes #95: schema.sql no longer issues failing CREATE INDEX statements for HNSW on bare halfvec columns (pgvector requires an explicit dimension spec). The two affected indexes are now documented in a new schema/hnsw-indexes.example.sql template that operators apply AFTER choosing their embedding provider's dimension. Fresh installs without embeddings emit no errors; installs with embeddings get a clear, idempotent path to the HNSW indexes. This is the first release publishing via npm Trusted Publishing (OIDC) instead of a long-lived token — validates the infrastructure landed in Sprint A and the Phase 2 follow-up (PR #97). Changes bundled: - package.json: 3.0.0-beta.3 → 3.0.0-beta.4 - CHANGELOG.md: new v3.0.0-beta.4 entry with per-feature and known-gap breakdown - ROADMAP.md: "Where We Are Today" updated; Phase 1 "Remaining" section collapsed (all shipped); Phase 2 marked COMPLETE with per-item PR references; #13 two-pass triage flagged as the one partial item - schema/schema.sql: remove two failing HNSW CREATE INDEX (with documented why-comment pointing at the template) - schema/hnsw-indexes.example.sql: new — template for operators to apply HNSW indexes after pinning halfvec dimension

Cold tier querying and restoration #14

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions