Skip to content

Releases

v2.4.19

Choose a tag to compare

@github-actions github-actions released this 09 Oct 12:08

Changelog

  • 4e8f3fc feat(backup): portable logical snapshot — cortex backup --out / restore --from (#165)
  • 9bbae51 feat(embedding): optional near-duplicate detection at ingest (duplicates_of, default OFF) (#167)
  • 356601d perf(embedding): provider-native MRL dimensions (CORTEX_EMBEDDING_DIMENSIONS) — unlock pgvector HNSW for dims <= 2000 (#169)

v2.4.18

Choose a tag to compare

@github-actions github-actions released this 09 Oct 08:50
adfa5e0

Changelog

  • adfa5e0 feat(bench): offline recall regression gate with pinned thresholds + advisory CI job (P8) (#163)
  • 8a00c89 feat(doctor): cortex doctor --deep active diagnostics (P1) (#159)

v2.4.17

Choose a tag to compare

@github-actions github-actions released this 09 Oct 06:17
712a985

Changelog

  • dfa89a0 feat(budget): shared provider rate budget for rerank, embedding and chat (P6, #150) (#151)
  • 712a985 feat(server): admin-gated Prometheus /metrics exposition (P9) (#157)
  • dc59852 feat(webkey): CORTEX_WEB_KEY env pin for ephemeral hosts (P7, #152) (#153)

v2.4.16

Choose a tag to compare

@github-actions github-actions released this 09 Oct 03:21
79668e0

Changelog

  • d982227 fix(store): merge-projects case-insensitive matching and global doctor orphan count

v2.4.15

Choose a tag to compare

@github-actions github-actions released this 08 Oct 22:33
5366ba1

Changelog

  • f4ef078 feat(bench): add fusion A/B decision-gate harness (RRF k=60 vs calibrated convex combination) (#137)
  • 7bfeff4 feat(retrieval): add tier-gated cached HyDE generation, default-off (#139)
  • da2fe54 perf(embedding): make worker LeaseBatch configurable at composition (#136)
  • 31c6d83 perf(mcp): route tiered search query embeddings through shared cache (#142)
  • 5366ba1 perf(pgvector): replace per-point upsert with chunked multi-row INSERT (#144)
  • 3cfbad4 perf(retrieval): rework rerank pacer to per-query token-bucket budget limiter (#138)
  • 9389ed3 perf(server): route agent retriever query embedding through shared cache (#141)
  • 0f6e2da refactor(server): delegate queryEmbeddingCache to retrieval.QueryEmbeddingCache (#140)

v2.4.14

Choose a tag to compare

@github-actions github-actions released this 08 Oct 18:36
0148639

Changelog

  • 0148639 fix(server): close idle provider connections before the provider path kills them (#133)
  • cd25322 perf(pgvector): add config-gated exact-scan tuning (parallel gather + optional normalized IP) (#132)
  • 3ed6af8 perf(retrieval): add shared QueryEmbeddingCache helper for query embeddings (#131)

v2.4.13

Choose a tag to compare

@github-actions github-actions released this 08 Oct 15:19
4269212

Changelog

  • b38563d fix(server): retry agent chat on transient provider failures and make the answer deadline configurable (#127)
  • 4269212 perf(server): cache hybrid search query embeddings to skip repeated provider round-trips (#128)

v2.4.12

Choose a tag to compare

@github-actions github-actions released this 08 Oct 08:14

Changelog

  • 0ce0a8c fix(retrieval): treat version-inclusive rerank base URLs correctly (#124)

v2.4.11

Choose a tag to compare

@github-actions github-actions released this 08 Oct 06:47
c0a7a0c

Changelog

  • d81f76a fix(server): persist per-observation embedding state for honest rag/stats coverage (#120)
  • c0a7a0c fix(server): wire rerank gate post-fusion with honest observability on hybrid search (#122)

v2.4.10

Choose a tag to compare

@github-actions github-actions released this 08 Oct 04:21

Changelog

  • 5dd4e6e fix(server): resolve project_key by label or public UUID in embedding worker query (#118)