Hey HN, we want to share HelixDB helix db, a project we have been working on to solve the messy state of AI agent memory. Why a hybrid? While vector databases excel at semantic similarity, many AI-driven applications need both similarity and deterministic relationship queries to function reliably. By combining a property graph engine with vector search, HelixDB allows agents to query complex causal chains and multi-hop relationships without losing semantic context. Here are some ways you can use it: - Complex Retrieval-Augmented Generation (RAG): You can link documents, entities, and user feedback into a unified graph, ensuring the LLM retrieves only the relevant, factually accurate context. - Multi-Agent Orchestration: You can store cross-agent state as typed relationships, allowing agents to coordinate tasks by traversing causal chains instead of reading raw chat logs. - Supply Chain Mapping: You can model suppliers, contracts, and delivery entities to answer specific questions like 'Which contracts are affected by this specific supplier outage?' through multi-hop graph traversal. Many people have a thing against “yet another database”, but we went ahead and did it anyway, because we think it makes working with structured agent memory so much easier. By implementing the core in Rust, we have built an LSM-based storage engine that handles high concurrency, allowing our system to be 10x faster than traditional patchwork pipelines that require separate synchronization for graph and vector data. Our benchmarking shows that we are on par with Pinecone and Qdrant for raw vector performance, while our graph traversal is up to three orders of magnitude faster than Neo4j for specific multi-agent state queries. See our design docs here: architecture. We also have a quick start video guide here: helix intro. If you would like to try it out in a simple RAG demo, you can follow this guide: rag demo. Many thanks! Comments and feedback welcome!