What is built. What is next.

LSM + PAX columnar storage, ART primary index, Monkey-optimal Bloom filters

NUMA-aware buffer pool, Lazy Leveling compaction with MVCC GC

WAL with crash recovery and write-stall mitigation

AES-256-GCM encryption at rest with pluggable key management (local file, env var, shell command, HashiCorp Vault)

Mutable HNSW with crash-safe delta buffer, SQ8/FP16/RaBitQ quantization, parallel construction

EMBEDDING MODEL columns with local HuggingFace sidecar (any model, no API key)

SEMANTIC_MATCH in any WHERE clause, combinable with SQL filters

AT VERSION time-travel queries and version tag snapshots

MinHash near-duplicate detection via WHERE NOT DUPLICATE

FOR TRAINING Lance export to PyTorch with zero-copy memory mapping

EU AI Act Article 13 data lineage

PostgreSQL wire protocol (psycopg2, SQLAlchemy, tokio-postgres, JDBC)

Python client: embedded mode (no server) and remote mode

pg_catalog compatibility

Local backup and restore

Observability: /health and /metrics (Prometheus) on port 9090

Pre-built binaries: Linux x86_64, macOS Intel, macOS Apple Silicon

Docker image, Homebrew tap, pip wheel

io_uring backend with graceful fallback to tokio

SCRAM-SHA-256 wire authentication

TLS transport encryption

Roles, privileges, and table-level GRANT / REVOKE

Native cloud KMS key providers: AWS KMS, GCP KMS, Azure Key Vault

Secondary (B-tree) indexes for non-primary-key lookups

Extended query protocol and prepared statements (ORM and connection-pooler support)

Faster single-row INSERT throughput

COPY protocol for bulk ingestion

Backup and restore to object storage (S3, GCS, Azure Blob, S3-compatible)

AT VERSION time-travel over on-disk data

Embedding columns included in FOR TRAINING exports

Automatic memory and configuration tuning

Semantic query caching via CREATE SEMANTIC CACHE

Gradient-driven adaptive storage (single-node)

Active-learning SQL: FEEDBACK, ORDER BY ACTIVE_LEARNING(), uncertainty and drift detection

Versioned vector-index snapshots via SEMANTIC_SNAPSHOT

Serializable Snapshot Isolation

Disk-resident ANN for larger-than-RAM vector sets

Distributed clustering with Raft replication and consistent-hash sharding

Distributed approximate nearest-neighbor search with correct global top-K

Cross-shard consistency and HTAP read replicas

Automated storage tiering: NVMe, object storage, cold archive

SSO (OIDC / SAML), fine-grained RBAC, and audit logging

Federated queries with differential privacy