Changelog

What shipped, what changed.

Every release of ManyVector — new capabilities, performance improvements, and new object-storage backends. Subscribe for updates.

Book a Demo Talk to us

0.9.0

July 2026

Feature
  • Hybrid search: RRF fusion of vector + BM25 full-text in a single query call
  • Namespace branching: copy-on-write forks for eval, testing, and rollback
  • Metadata pre-filter mode: apply attribute conditions before HNSW traversal
  • New storage backend: Cloudflare R2 (zero egress, S3-compatible)
  • Python SDK 0.9: unified query() interface for vector, full-text, and hybrid

0.8.0

May 2026

Feature
  • Full-text search: BM25 ranking across stored attributes, no separate cluster
  • Google Cloud Storage backend: multi-region and dual-region support
  • Azure Blob Storage backend: LRS, ZRS, and GRS redundancy
  • efSearch and M tunable per namespace for recall/latency trade-offs
  • Cosine, dot product, and L2 distance metrics supported

0.7.0

March 2026

Performance
  • Cold namespace warm-up latency reduced by 40% via prefetched graph segments
  • HNSW index write throughput improved 2× through batched object flushes
  • MinIO / self-hosted S3-compatible backend: air-gapped deployments now supported
  • Namespace-level cache TTL configuration
  • Improved recall at high efSearch values for large (1B+) namespaces

0.6.0

January 2026

Foundation
  • AWS S3 backend: Standard and Express One Zone storage classes
  • HNSW index stored as object-storage native format — no proprietary disk format
  • Vector upsert and delete with consistent read-after-write guarantees
  • Initial Python SDK with namespace, upsert(), query(), and delete() primitives
  • Metadata attribute storage alongside vectors — no separate document store needed
Stay updated

Get release notes by email.

No marketing. Just a short email when a new version ships.