Hey HN, we want to share HelixDB helix db, a project we are building to fundamentally change how enterprise AI applications handle interconnected data. Why a hybrid? Standard vector databases are great at finding semantically similar text, but they fall apart when your AI assistant needs to perform multi-hop reasoning across an organization's specific relationships—like identifying which projects a specific engineer worked on or how documents are linked to business initiatives. HelixDB solves this by natively combining graph and vector capabilities into a single, unified engine.

Here are a few ways this changes the development landscape:

• Enterprise Knowledge Mapping: You can vectorize documentation while maintaining explicit edges between authors, teams, and product features to prevent hallucinations.
• Supply Chain Intelligence: You can index parts as vectors while querying the graph to trace multi-tier dependency chains, allowing for immediate impact analysis.
• Regulatory Compliance: You can map legal entities and their relationships as a graph to ensure that RAG-based search retrieves only legally relevant, authorized document versions.

Many developers are skeptical of 'yet another query language,' but we went ahead and implemented a custom DSL in Rust and TypeScript because it allows you to define complex traversal logic as an inline HTTP request, saving you from the headache of managing separate database clients. You can see how this works in our quick start guide: introduction.

Our benchmarking shows that we are on par with Pinecone and Qdrant for vector similarity, and our graph traversal latency is up to three orders of magnitude faster than Neo4j because we eliminate the need for application-side joins. Check out our technical deep dive and architecture video here: watch.

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