Hey HN, we want to share HelixDB helix db, a fully native Graph-Vector Database implemented in Rust that we are working on. Why a hybrid? Standard vector databases are great for semantic similarity, but many AI-driven applications need both similarity and structural relationship queries to truly understand organizational data. When an AI agent is asked, 'who on the team has worked with X technology?', standard vector retrieval often fails to connect the dots. We built HelixDB to bridge this gap.

Use Cases

• Skills Mapping: Trace a path from a specific developer to a project timeline and the specific technology deployed, ensuring accurate personnel tracking.
• Complex Organizational Reasoning: Perform multi-hop queries that require linking disparate documents about system failures, team members, and specifications.
• Knowledge Graph RAG: Replace fragmented multi-tool stacks with a unified property graph engine, reducing synchronization errors and latency.

Technical Deep Dive

We implemented HelixDB in Rust to provide maximum memory safety and low CPU overhead. Many people are skeptical of “yet another database engine,” but we went ahead and did it anyway, because we believe the native fusion of graph and vector types is essential. By keeping relationships intact within the core engine—rather than forcing developers to query two disconnected data stores—we reduce the impedance mismatch between mapping relationship nodes and calculating embeddings.

Performance Benchmarking

Our current benchmarks show that HelixDB is on par with Pinecone and Qdrant for pure vector search, while our graph traversal performance is up to three orders of magnitude faster than Neo4j for complex multi-hop enterprise queries. Our new LSM-based storage engine, detailed here at introduction, handles concurrent writes to the writer node with significantly higher throughput than traditional sequential systems.

Get Involved

If you’d like to try it out in a simple RAG demo, you can follow this guide: examples. We are eager to see what you build. Many thanks! Comments and feedback are more than welcome.