HelixDB

HelixDB is a graph-vector database written entirely in Rust that combines graph traversal, vector similarity search, and keyword search in a single embedded store. It targets AI application backends — RAG pipelines, agent memory, semantic search — where developers would otherwise stitch together a relational database, a vector store, and a graph database. HelixDB replaces that stack with one system, one query language, and one deployment. The storage engine is LMDB, giving it extremely low read latency with full ACID guarantees.

Features

• Graph + vector data model — nodes, edges, and vector embeddings are first-class citizens in the same store; no synchronisation between separate systems required.
• HelixQL — a type-safe, compiled query language with a schema definition step; queries are checked at compile time before deployment, catching type errors before they reach production.
• Built-in embeddings — the Embed function vectorises text inline during a query, removing the need for a separate embedding service.
• Built-in MCP support — agents can discover data and walk the graph via the Model Context Protocol without writing human-readable queries.
• Vector, keyword, and graph search — combine approximate nearest-neighbour vector search, BM25-style keyword search, and multi-hop graph traversals in a single query.
• Secure by default — data is only accessible through compiled HelixQL queries; there is no general-purpose query surface exposed over the network.
• TypeScript and Python SDKs — official client libraries for calling compiled query endpoints from application code.
• Ultra-low latency — LMDB as the storage backend delivers sub-millisecond read latency suitable for interactive applications.

Installation

The Helix CLI is the primary way to install and manage HelixDB locally. The install script works on Debian, Ubuntu, Fedora, and all other Linux distributions, as well as macOS.

curl -sSL "https://install.helix-db.com" | bash

For Debian / Ubuntu and Fedora specifically, the curl script above is the recommended install path. Pre-built binaries for each release are also available on the releases page if you prefer to install without running a script.

# macOS (via the same install script)
curl -sSL "https://install.helix-db.com" | bash

Quick Start

1. Initialise a project

mkdir my-app && cd my-app
helix init

This creates a .hx schema and query file in the current directory.

2. Define a schema and queries

# schema.hx — define a node type with an indexed field
N::User {
    INDEX name: String,
    age: U32
}

# queries.hx — write typed queries against the schema
QUERY addUser(name: String, age: U32) =>
    user <- AddN<User>({name: name, age: age})
    RETURN user

QUERY getUser(user_name: String) =>
    user <- N<User>({name: user_name})
    RETURN user

3. Validate and deploy

# Check that queries compile against the schema
helix check

# Deploy queries as HTTP endpoints to the local dev instance
helix push dev

4. Call from application code

# TypeScript
npm install helix-ts

# Python
pip install helix-py

import HelixDB from "helix-ts";

const client = new HelixDB(); // default port: 6969

await client.query("addUser", { name: "Alice", age: 30 });

const user = await client.query("getUser", { user_name: "Alice" });
console.log(user);

Vector search example

HelixDB can store and search embeddings alongside graph data. The Embed function takes a string and returns a vector inline, so no external embedding step is needed:

N::Document {
    content: String,
    embedding: F64[1536]
}

QUERY addDocument(text: String) =>
    doc <- AddN<Document>({
        content: text,
        embedding: Embed(text)
    })
    RETURN doc

QUERY search(query: String) =>
    results <- N<Document>::Search(Embed(query), 10)
    RETURN results

HelixDB vs separate stacks