Hey HN, we want to share HelixDB, a native graph-vector database designed to replace the unmaintainable JSON blobs currently used for agent memory. Why a hybrid? Vector databases are excellent for similarity queries, but many modern AI-driven applications need both similarity and complex relationship queries to function reliably. Instead of hacking relational databases with brittle JSON, HelixDB provides a native environment for structured agent context. Here are a few ways this changes development:

• Complex Agent Context: You can model multi-hop relationships as first-class citizens, allowing agents to traverse facts rather than parsing strings.
• Dynamic RAG: You can combine semantic vector search with graph traversal to ensure agents retrieve 'ground truth' alongside similar passages. See our design paper for more.
• Multi-Agent Coordination: You can maintain a shared memory graph that allows disparate agents to read and write state without causing data collisions.

Many people are skeptical of 'yet another query language,' but we implemented a custom Rust and TypeScript DSL because it allows developers to send dynamic queries via HTTP, removing the need for constant schema migrations or separate deployment steps. Our benchmarking shows that while we are on par with Pinecone and Qdrant for vector operations, our graph traversal is up to three orders of magnitude faster than traditional approaches like Neo4j when handling concurrent agentic workloads. Check out our video runthrough to see this performance in action.

We provide extensive documentation and examples at docs.helix db.com. If you’d like to try it out in a simple RAG demo, you can follow our quick-start guide at examples. Many thanks! Comments and feedback welcome!