HelixDB is an OLTP graph-vector database built in Rust that combines graph, vector, and full-text search capabilities. The project emerged from a college collaboration and addresses a common pain point in AI development: stitching together multiple disconnected database systems. Traditional approaches require application-level logic to coordinate queries across separate graph and vector databases.

The database uses object storage as its persistence layer, enabling unlimited graph size and relationships. This architecture avoids the scaling challenges that plague other graph databases, which typically duplicate entire datasets or struggle with sharding due to interconnected edges spanning partitions. By storing data in S3 and caching hot subsets on each node, HelixDB achieves approximately 100ms write latency and 50ms read latency from cold storage.

The system supports graph + vector data models alongside KV, documents, and relational data through Rust and TypeScript SDKs. Developers can deploy locally using the CLI or scaffold projects with the `helix chef` interactive bootstrapper. A managed cloud service with high availability and auto-scaling reader nodes is nearing general availability.

HelixDB targets AI memory systems, company-wide agent autonomy, and workloads requiring terabytes of searchable data. The approach consolidates what would typically require multiple database technologies into a single platform, potentially reducing operational complexity for teams building knowledge-intensive AI applications.