◆pip install vantadb-py◆BM25 + HNSW via RRF◆WAL · CRC32C checksums◆1.2ms HNSW p50 · 10K◆Apache 2.0◆Rust 1.94.1+◆Zero network◆Local-first◆pip install vantadb-py◆BM25 + HNSW via RRF◆WAL · CRC32C checksums◆1.2ms HNSW p50 · 10K◆Apache 2.0◆Rust 1.94.1+◆Zero network◆Local-first

Changelog — VantaDB

§09

Changelog

Historial de versiones

Todos los releases
0
v0.5.0Latest2026-07-31

IVF Flat index + multi-level LSM compaction

  • IVF Flat index — inverted file with k-means clustering (no external deps)
  • New IndexType::Ivf on HnswConfig; lazy-built on first search, serialized in v8 format
  • ~50x faster than brute-force Flat on 1M vectors at ~90% recall
  • Multi-level LSM compaction (L0→L1→L2→L3): per-level VantaFiles with legacy migration
  • compact_level() promotes live nodes between tiers; write amplification reduced from O(all data) to O(L0 size)
  • New PipelineMode::CompactOnly / CompactL0Only variants
0
v0.4.0MVP2026-07-20

Initial public release

  • Embedded persistent vector/graph database engine (HNSW, configurable distance metrics)
  • WAL with automatic crash recovery; Arrow IPC zero-copy interchange
  • Hybrid Retrieval v1: BM25 + HNSW fused via RRF
  • Python SDK (PyO3), WASM build, TypeScript SDK (vantadb-ts)
  • CLI tool (vanta) and HTTP server with rate limiting and TLS support
  • AI framework adapters: LangChain, LlamaIndex, Haystack, CrewAI, DSPy, Litellm, OpenAI, Ollama, Mem0, Letta
  • MCP server (vantadb-mcp); Prometheus metrics and OpenTelemetry tracing
  • Encryption at rest (AES-GCM), PITR, WAL shipping for replication

Sigue la hoja de ruta

Mirá próximos milestones, RFCs y releases en el repo de GitHub.

Ver en GitHub