HelixDB: A Multi-Model Graph-Vector Database for AI Applications

HelixDB is a unified OLTP graph-vector database built in Rust, designed to eliminate the need for separate application, relational, vector, and graph databases when building AI agents. By providing a single platform for memory, company brains, and federated data access, HelixDB simplifies the storage layer for complex AI applications.

Unified Multi-Model Data Support

HelixDB primarily utilizes a graph and vector data model but extends support to key-value (KV), document, and relational data. This multi-model approach allows developers to manage diverse data types without the overhead of managing multiple storage locations or application layers.

Architecture and Deployment Options

HelixDB offers two primary deployment paths: local development and a managed cloud service.

Local Development

Local instances are managed via the Helix CLI. Developers can initialize projects using helix init and start instances with helix start dev. By default, local instances operate in-memory, meaning data is wiped upon stopping the instance; however, persistence can be enabled using the --disk flag.

HelixDB Cloud

HelixDB Cloud is a distributed, high-availability managed service. Its architecture is backed by object storage and includes the following features:

  • Integrated Search: Built-in vector and full-text search.
  • Transactionality: Full ACID transactions.
  • Scalability: A single-writer architecture with auto-scaling reader nodes.
  • Availability: High availability ensured by three or more gateways and database nodes.

Developer Experience and Tooling

HelixDB provides a streamlined onboarding process and strongly typed SDKs to reduce the friction of database interaction.

Rapid Prototyping with helix chef

The helix chef command is an interactive bootstrapper that scaffolds a project, installs query skills and documentation via MCP, seeds example data, and generates a HELIX_CHEF_PROMPT.md. This tool is designed to integrate with coding agents like Claude Code, Codex, or OpenCode to build full-stack applications from a single description.

Type-Safe Querying via SDKs

Queries are authored using Rust or TypeScript DSLs and sent as dynamic requests to the POST /v1/query endpoint. This architecture removes the need for a build or deploy step for queries.

  • Rust SDK: Uses the helix-db crate and allows queries to be defined as #[register] functions.
  • TypeScript SDK: Uses the @helix-db/helix-db package and provides a DSL for defining parameters and projections.

Community Insights and Technical Discussion

Following its announcement, the developer community raised several technical questions regarding the database's performance and internals:

  • Performance and Query Patterns: Users have inquired about p99 latency for multi-hop queries and the specific graph shapes or query patterns that represent worst-case scenarios for an object-storage-backed architecture.
  • Open Source Status: Some community members noted that the provided GitHub repository appears to contain the client SDKs rather than the source code for the database engine itself.
  • Feature Requests: There is active interest in the availability of a generalized AI memory layer and whether full-text search (FTS) or vector capabilities are supported specifically on the edges of the graph.
  • Comparisons: Users have compared HelixDB to other multi-model or object-storage-based databases such as Turbopuffer and SurrealDB.

Sources