Skip to content

README.md

Latest commit

 

History

History
42 lines (29 loc) · 2.29 KB

README.md

File metadata and controls

42 lines (29 loc) · 2.29 KB

Dgraph

A distributed Apache 2.0 graph database with native float32vector and HNSW indexes queryable through DQL and GraphQL.
Website dgraph.io
Model Label Property Graph
Status Production
License Apache 2.0 (permissive)
Query DQL / GraphQL
Vector strategy Native
HA Yes
Deployment Self-hosted, managed by partners

Overview

Dgraph is a distributed, sharded graph database written in Go, with native GraphQL and DQL query interfaces. It was acquired by Hypermode in 2024. Dgraph v24 introduced the float32vector data type and an hnsw index, exposing similarity search through the similar_to() function in DQL. Vectors are stored as ordinary node properties and multiple embeddings per node are supported.

Hypermode positions Dgraph with a "Graph-and-Vector" (GAV) framing where graph and vector are both first-class primary indexes, not a graph layered on a vector store or vice versa. HA is provided through Raft replication. Native sharding is built into the architecture.

Vector strategy — Native

Dgraph documents float32vector plus HNSW indexes in DQL. Distance metrics include cosine, Euclidean, and dot product. Vectors are stored as node properties alongside other attribute types. Hybrid queries that combine graph traversal with vector similarity are supported in a single DQL statement. This is clear first-class vector support in current official documentation.

Best fit

  • Distributed graph plus vector workloads where a single Apache 2.0 engine is preferred
  • GraphQL-friendly workflows that need graph traversal and vector similarity in one API
  • Teams that want native HNSW-indexed similarity without an external vector service
  • Workloads requiring multiple embeddings per node

Considerations

  • Distributed architecture adds operational complexity relative to single-node alternatives
  • Maintainer changed to Hypermode in 2024; evaluate long-term roadmap continuity
  • DQL is a proprietary query language; Cypher or SPARQL familiarity does not transfer directly
  • Managed hosting options depend on third-party partners rather than a first-party cloud service

Sources