HelixDB/helix-db

HelixDB – a graph‑vector database built for AI applications

What it is – HelixDB is a Rust‑implemented database that combines graph, vector, key‑value, document and relational storage into a single engine. It is marketed as the “database for knowledge graphs and AI memory”, letting LLM‑powered agents query and mutate data without juggling separate DB products.

Why it matters for AI – Modern AI agents often need to retrieve structured knowledge (graphs) and semantic similarity (vectors) in the same request. HelixDB exposes a single HTTP POST /v2/query endpoint that accepts a JSON‑encoded abstract syntax tree (AST) generated by language‑specific SDKs. The server then executes the mixed graph‑vector query atomically, with ACID guarantees and full‑text search.

Core features

• Hybrid data model – Graph + vector as the primary model; also supports KV, document and relational tables.
• Multi‑language SDKs – Rust, TypeScript, Python and Go provide a DSL that builds the query AST; the same AST is sent over HTTP, so code is portable across languages.
• CLI (helix) – Installs locally, spins up a dev instance, scaffolds projects, and can hand off to supported AI agents (Claude, OpenAI Codex, etc.) via the helix chef wizard.
• HelixDB Cloud – Managed, HA deployment with object‑storage‑backed persistence, auto‑scaling readers, single‑writer transactions, and integrated vector/full‑text search.
• Versioned wire protocol – Current endpoint is POST /v2/query; the version refers to the API, not the product version.

Getting started (quick‑start)

1. Install the CLI

   # macOS / Linux
   curl -sSL https://install.helix-db.com | bash
   # Windows PowerShell
   irm https://raw.githubusercontent.com/HelixDB/helix-db/main/crates/cli/install.ps1 | iex
   

2. Run the bootstrapper

   helix chef   # answers prompts and creates a runnable app
   

   The wizard installs the query SDK, seeds example data, starts a local server on http://localhost:6969, and writes a HELIX_CHEF_PROMPT.md describing the generated app.
3. Or follow the manual quick‑start – see the docs for a step‑by‑step local setup.

Example query (Rust)

use helix_db::Client;
use helix_db::dsl::prelude::*;

#[query]
fn add_user(name: String) -> WriteBatch {
    write_batch()
        .var_as("user", g().add_n("User", vec![("name", name)])
            .value_map(None::<Vec<String>>))
        .returning(["user"])
}

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = Client::new(None)?; // defaults to localhost:6969
    let resp: serde_json::Value = client.query(add_user("John Doe".into())?).send().await?;
    println!("new user: {:#}", resp);
    Ok(())
}

Similar snippets are provided for TypeScript, Python and Go.

Typical use cases

• RAG / Retrieval‑Augmented Generation – Store documents as vectors, link them with graph relationships (e.g., topics, authors) and retrieve relevant chunks in a single query.
• Agent memory – Persist an LLM‑agent’s internal state, facts, and embeddings so the same agent can recall and reason over past interactions.
• Knowledge‑graph‑backed apps – Build recommendation, fraud‑detection, or supply‑chain systems that need both relational joins and semantic similarity.
• Hybrid AI pipelines – Combine structured business data (SQL‑like tables) with unstructured embeddings without moving data between services.

Ecosystem & community

• Docs & tutorials – https://docs.helix-db.com (query guide, SDK setup, cloud deployment).
• Discord – active community for troubleshooting and feature discussion.
• Y‑Combinator launch – early‑stage backing, indicating commercial interest.
• Open‑source repo – core engine and CLI are under the HelixDB/helix-db GitHub repository; the Rust crate is published as helix-db (v3.0.0).

License

The repository is open‑source (the README does not specify a license, but the code is published on GitHub and crates.io, so it is publicly available for use and contribution).

---

HelixDB aims to be the one‑stop store for AI‑centric workloads, letting developers write a single query that spans graph structure, vector similarity, and traditional data models.