HelixDB — OLTP Graph-Vector Database Built in Rust

Introduction

HelixDB is an open-source database that unifies graph and vector storage in a single engine. Written in Rust and built on object storage, it is designed for AI-native applications that need both relationship traversal (graph queries) and similarity search (vector queries) without running separate databases. It targets OLTP workloads where low latency matters.

What HelixDB Does

• Stores and queries graph data with nodes, edges, and properties
• Provides vector similarity search for embeddings and feature vectors
• Combines graph traversal and vector search in unified queries
• Runs on object storage for cost-effective, scalable persistence
• Delivers low-latency OLTP performance from a Rust-native engine

Architecture Overview

HelixDB is written entirely in Rust for memory safety and performance. The storage layer sits on object storage (S3-compatible), which decouples compute from storage and enables cost-effective scaling. The query engine supports a custom query language for graph traversal operations (node/edge creation, path finding, pattern matching) and vector operations (nearest-neighbor search, cosine similarity). An in-memory cache layer handles hot data for sub-millisecond latency on frequently accessed nodes and vectors.

Self-Hosting & Configuration

• Install via Cargo or download pre-built binaries for Linux and macOS
• Configure data directory for local storage or S3-compatible object storage
• Tune cache size and index parameters for your workload profile
• Expose the query endpoint on a configurable port
• CLI tools included for database management, backup, and migration

Key Features

• Unified graph and vector storage in a single database engine
• Rust-native implementation for high performance and memory safety
• Object storage backend for scalable and cost-effective persistence
• Combined graph traversal and vector similarity queries
• Lightweight deployment with no JVM, Python, or external dependencies

Comparison with Similar Tools

• Neo4j — graph-only database without native vector support; HelixDB combines both
• Milvus — vector-only database without graph capabilities
• Qdrant — vector database; does not support graph traversal queries
• DGraph — distributed graph database but requires more infrastructure
• SurrealDB — multi-model database; HelixDB focuses on the graph-vector intersection with Rust performance

FAQ

Q: What query language does HelixDB use? A: It uses a custom query language designed for combined graph and vector operations.

Q: Can I use HelixDB for RAG applications? A: Yes. Store document embeddings as vectors and use graph edges to model document relationships and metadata.

Q: Does it support ACID transactions? A: Yes. HelixDB provides transactional guarantees for OLTP workloads.

Q: What object storage backends are supported? A: Any S3-compatible storage including AWS S3, MinIO, and Cloudflare R2.

Sources

• https://github.com/HelixDB/helix-db