Back to VectorScale
Changelog
Engine releases, SDK updates, and infrastructure improvements across VectorScale.
v2.4.0May 18, 2025
Single-Stage Hybrid Search & Asymmetric Binary Quantization
Unifies dense vector search and sparse BM25 lexical scoring in one gRPC call with automated Reciprocal Rank Fusion (RRF) and 32x RAM compression.
- FeatureAdded native hybrid query API combining dense HNSW, sparse SPLADE/BM25, and configurable RRF weights.
- FeatureLaunched zero-config binary quantization with asymmetric float16 rescoring from NVMe.
- ImprovementReduced cold-namespace wake-up latency from 42ms to 11ms via parallel S3 range prefetching.
- FixResolved edge-case deadlock during concurrent schema evolution on indexes exceeding 500M vectors.
v2.3.0April 09, 2025
Kafka CDC Connector & OpenTelemetry Native Exporter
Stream embeddings directly from Confluent Cloud or Redpanda and export per-namespace latency histograms to Datadog, Grafana, or Honeycomb.
- FeatureReleased managed Kafka & MSK Connect sink with automatic dead-letter queue retries.
- ImprovementAdded OpenTelemetry OTLP trace propagation across TypeScript and Python SDK v3.2.
- FixFixed roaring bitmap intersection memory spike on deeply nested $or metadata filters.
v2.2.0March 02, 2025
EU-Central Sovereign Region & Customer-Managed KMS
Expanded our global footprint to Frankfurt and Zurich with full BYOK envelope encryption and zero cross-border telemetry.
- FeatureAdded eu-central-1 (Frankfurt) and eu-central-2 (Zurich) serverless control planes.
- FeatureIntroduced AWS KMS and GCP Cloud KMS key rotation without index rebuilds.
- ImprovementUpgraded gRPC HTTP/2 connection multiplexing for 2.4x higher batch upsert throughput.