HelixDB

Key features

• One engine for graph, vector, and text Nodes, edges, embeddings, and text live in a single transactional store, so one query can combine connected entities, semantically similar content, and exact terms.[1][2]

• Labeled property graph model Nodes are entities, directed edges are relationships, and both carry typed properties, mapping naturally to users, documents, events, and their links.[1]

• Vector search with graph prefiltering Approximate nearest-neighbor search can be prefiltered by graph traversal, with over 90 percent recall documented.[1][2]

• BM25 full-text search Keyword search with BM25 relevance runs over node and edge properties for exact terms alongside semantic matching.[1][2]

• Searchable edges Indexes are scoped to a label and a property rather than a special node type, so relationships can be searched and filtered as well as entities.[1]

• ACID transactions Reads and writes across graph, vector, and text data share one consistency model, with Cloud commits serialized through a single MVCC writer.[1][2]

• Object-storage durability Object storage is the system of record while NVMe and memory serve as caches, letting datasets grow beyond serving-machine capacity.[1]

• Four SDKs and a CLI workflow Rust, TypeScript, Python, and Go SDKs produce the same JSON query request over HTTP, and the helix CLI covers local instances, embedded mode, and Cloud deployment.[1]

Who will love it

• Agent memory and RAG builders Documented uses are company brains, people search, and agent swarms, where relationships plus embeddings plus text match the retrieval problem directly.[1]

• Teams running three stores today If the current design synchronizes graph, vector, and search systems, the single-engine model removes an entire consistency burden.[1]

• Rust, TypeScript, Python, or Go shops All four SDKs emit the same wire request, so polyglot teams share one query model and one local development loop.[1]

Who should avoid it

• Teams needing no-code operations Local development needs the CLI plus Docker or Podman and queries are written in code; there is no visual builder or spreadsheet surface.[1]

• Buyers needing published pricing and SSO Cloud pricing has no public per-seat schedule on the inspected pages and SSO support is undocumented.[1]

• Teams needing at-scale references The Cloud architecture is young, BYOC is still coming, and the v1-to-current storage rewrite means older reports may not describe the present engine.[1]

Platforms, SDKs, and query interfaces