Hey HN, we want to share HelixDB helix db, a native Graph-Vector Database we are building in Rust to solve complex RAG retrieval challenges.

Why a hybrid? Vector databases are excellent for semantic similarity queries, but they often struggle to 'connect the dots' when an AI needs to reason across fragmented documents or complex entity relationships. Many developers find themselves stitching together separate graph and vector stores, leading to massive synchronization headaches. We built HelixDB to provide both natively in a single, high-performance engine, allowing you to build RAG applications 10x faster.

Here are some ways our architecture helps:

• Knowledge Graph Construction: Map relationships between entities across millions of documents for precise multi-hop reasoning.
• Enterprise Document Analysis: Cross-reference incident logs, employee schedules, and technical documentation to diagnose system failures.
• Semantic Clustering: Combine vector embeddings with metadata properties to filter results at query time without maintaining external indexes.

Many developers are wary of learning yet another query language or adding another database component to their stack, but we went ahead and did it because the performance gains from a unified memory layer are too significant to ignore. By unifying the storage engine, we eliminate the latency penalty of cross-database API calls.

In terms of performance, our benchmarking shows that while we remain on par with established vector stores for pure similarity tasks, our graph traversal speed is up to three orders of magnitude faster than traditional Neo4j-based setups for multi-hop RAG queries. You can see a deep dive on our architectural choices in our documentation: introduction.

We would love to see what you build with it. If you’d like to try it out in a simple RAG demo, you can follow this guide: getting started. Many thanks! Comments and feedback welcome!