HelixDB is a graph database with native vector search and BM25 full-text search, written in Rust on object storage. It stores entities, relationships, embeddings, and text in one transactional engine, so an application queries all of them together instead of keeping a separate relational, vector, and graph store in sync.
A typical query traverses the graph first and then searches only the records it reached. For example, it walks from a service to the issues that affect it, then runs vector and BM25 search over those issues, in one request. Use cases include long-term agent memory, a connected company knowledge graph, and retrieval for AI apps that returns only the records each user may see.
Queries are written with SDKs for Rust, TypeScript, Python, and Go, and each SDK produces the same JSON request, with no build or deploy step. Source builds also accept Cypher for reads and CRUD. HelixDB runs as a local instance, embedded inside an application process, or on Helix Cloud, a managed deployment where object storage is the durable record, a single writer commits ACID transactions, and reader nodes scale with query load.
Features
- Labeled property graph: models entities, relationships, and typed properties
- Vector search: approximate nearest-neighbor search, prefiltered by graph traversal
- Full-text search: BM25 keyword search over node and edge properties
- Multi-language SDKs: Rust, TypeScript, Python, and Go clients that all emit the same JSON query request
- Cypher support: Cypher reads and CRUD through the HTTP API, embedded clients, and the CLI in source builds
- Embedded mode: runs inside your own process without a separate server
- Agent scaffolding: a CLI command installs HelixDB query skills and a docs MCP server, seeds example data, and hands the project to a coding agent
- Helix Cloud: managed deployment with ACID writes and auto-scaling reader nodes
