// CHANGELOG

What is new in GalaxDB.

Every shipped feature, improvement, and fix. All releases on GitHub.

v1.0.0-beta.1·

First public beta

GalaxDB v1.0.0-beta.1 is the first public release. All core v1 features are included.

  • ReleaseFirst public beta. Pre-built binaries for Linux x86_64, macOS Intel, macOS Apple Silicon.
  • FeatureEMBEDDING MODEL column type -- declare a HuggingFace model in DDL, embeddings computed automatically on INSERT via the local sidecar.
  • FeatureSEMANTIC_MATCH(col, 'query', threshold) -- vector similarity search in any WHERE clause, combinable with SQL filters.
  • FeatureCREATE VERSION TAG ... FOR TRAINING -- immutable training snapshots with Lance export and PyTorch zero-copy loading.
  • FeatureAT VERSION 'tag' -- time-travel queries against historical snapshots.
  • FeatureWHERE NOT DUPLICATE -- MinHash LSH near-deduplication, typically removes 15-30% of training data.
  • FeatureBULK INSERT -- multi-row insert with a single statement.
  • FeatureBACKUP TO / RESTORE FROM -- point-in-time backup and restore.
  • FeatureANALYZE -- background statistics collection for the adaptive query planner.
  • FeaturePostgreSQL wire protocol -- psycopg2, SQLAlchemy, tokio-postgres, JDBC all work unchanged.
  • FeatureEmbedded mode -- use as a Python library with no server, like SQLite.
  • FeatureDocker image -- harbi256/galaxdb:latest on Docker Hub.
  • FeatureHomebrew tap -- brew tap zentrix-innovative-labs/tap && brew install galaxdb.
  • FeaturePython wheel -- pip install galaxdb-client, pre-built for Linux, macOS, Windows.
  • FeatureObservability -- /health and /metrics (Prometheus) on port 9090.
  • FeaturePluggable key management -- local file, env var, AWS KMS, HashiCorp Vault, any shell command.
  • ImprovementHNSW recall@10 = 0.990 on SIFT-1M at ef=200. Build rate 15,114 vec/sec.
  • Improvement258,555 write TPS (16 threads, 1M rows, group commit WAL, NVMe).
  • Improvement4.49 GB/s scan throughput with PAX blocks and zone-map pruning.
  • Improvement740 Rust unit tests passing. 7 chaos scenarios in 10.91s.
  • Fixio_uring falls back to tokio gracefully when running inside Docker Desktop or VMs that restrict the syscall.