Hey HN, we want to share HelixDB helix db, a next-generation database technology we are working on that is implemented natively in Rust to help developers build RAG and AI applications 10x faster.

Why a hybrid? Vector databases are excellent for semantic similarity queries, but many production-grade AI agents require both similarity and precise relationship queries to maintain context. Often, agents fail because they lack a shared, structural memory layer. HelixDB addresses this by fusing a property graph engine with approximate vector search and BM25 full-text search, ensuring agents can reason over both context and truth.

Here are some ways you can use HelixDB to improve your agent workflows:

• Complex supply chain modeling: You can vectorize product descriptions while maintaining graph relationships between suppliers and parts to retrieve exact truth alongside semantic suggestions.
• Multi-agent memory: Agents can share a continuous memory state, storing episodic interactions as nodes in a graph that other agents can query for historical alignment.
• Enterprise RAG: By combining BM25, vector search, and graph traversal, you can ensure the system retrieves specific policy documents linked to an entity rather than just 'similar' text snippets.

Many people are skeptical of 'yet another database engine,' but we built this natively in Rust precisely to eliminate the operational nightmare of synchronizing separate graph and vector stores. By using a native Graph-Vector architecture and backing it with durable object storage, we provide a unified engine that avoids the data silos that plague legacy setups. Our benchmarks show that we are on par with dedicated vector stores for similarity search, while our graph traversal performance is up to three orders of magnitude faster than traditional standalone graph databases when integrated into RAG pipelines.

Check out this video runthrough for a quick look at the architecture: watch. You can also dive into our technical design documentation here: docs.helix db.com.

If you’d like to try it out in a simple RAG demo, you can follow this guide: examples. Many thanks! Comments and feedback welcome!