Skip to content

Get Started

Velr is an embedded property-graph database for local-first, edge, and AI-native applications. It is implemented in Rust, stores data in a standard SQLite database file, and is queried with openCypher.

It also includes BM25 full-text search and vector approximate nearest neighbor (ANN) search for retrieval-heavy graph applications.

Note

Velr is currently in public alpha. The public drivers are usable today, but APIs may still evolve before 1.0.

Fastest Path

The quickest way to try Velr locally is with the released CLI.

Install Velr with Homebrew on macOS, Linux, or Windows Subsystem for Linux:

brew tap velr-ai/velr
brew trust --formula velr-ai/velr/velr
brew install velr

Verify the install and run a first query against an in-memory database:

velr --version
velr -e 'RETURN 1 AS n'

Use a database file when you want persistent state:

velr graph.db -e 'CREATE (:Person {name: "Ada"});'
velr graph.db -e 'MATCH (p:Person) RETURN p.name AS name;'

For direct binary downloads, Windows installs, manual setup, output formats, REPL commands, and scripting workflows, see Velr CLI.

First Embedded Query

If you want to embed Velr in an application, Python is the fastest driver path for AI, notebooks, pandas, Polars, and Arrow workflows.

Install the Python package:

python -m pip install velr

Create an in-memory graph, bind a parameter, and read a result:

from velr.driver import Velr

with Velr.open(None) as db:
    db.run(
        "CREATE (:Person {name: $name})",
        params={"name": "Ada Lovelace"},
    )

    with db.exec_one("MATCH (p:Person) RETURN p.name AS name") as table:
        rows = table.collect(lambda row: [cell.as_python() for cell in row])
        print(rows)

Use Velr.open("graph.db") instead of Velr.open(None) for a file-backed database.

Pick A Driver

All public drivers use the same embedded Velr runtime and SQLite-backed file format. Choose the driver that fits your application.

Best for AI, data science, notebooks, pandas, Polars, and Arrow workflows.

python -m pip install velr

Python 3.12 or newer is required.

Best for embedded systems, edge runtimes, high-performance services, and native applications.

[dependencies]
velr = "0.2"

Best for services, tools, agents, and infrastructure software that wants a small embedded graph layer.

go get github.com/velr-ai/velr-go-driver@latest

Go 1.22 or newer is required.

Best for Node.js applications, worker-thread workflows, tooling, and JavaScript/TypeScript agents.

npm install @velr-ai/velr

Node.js 22 or newer is required.

Best for JVM and Android applications that need embedded graph queries with native runtime performance.

repositories {
    mavenCentral()
}

dependencies {
    implementation("ai.velr:velr-java-driver:0.2.35")
}

For Android, use implementation("ai.velr:velr-java-driver-android:0.2.35").

Best for Kotlin JVM and Android applications. The Kotlin artifact includes the Java core classes plus Kotlin extension helpers.

repositories {
    mavenCentral()
}

dependencies {
    implementation("ai.velr:velr-kotlin-driver:0.2.35")
}

For Android, use implementation("ai.velr:velr-kotlin-driver-android:0.2.35").

The same query model is available from every driver. The language-specific driver pages show richer examples for transactions, parameters, Arrow, BM25 full-text search, and vector/ANN search.

Retrieval Built In

Velr alpha includes retrieval features directly in the embedded engine:

  • BM25 full-text indexes with CREATE FULLTEXT INDEX
  • Full-text queries with CALL db.index.fulltext.queryNodes(...)
  • Vector/ANN indexes with CREATE VECTOR INDEX
  • Vector queries with CALL db.index.vector.queryNodes(...)
  • Application-provided embedders for vector index maintenance and query text

The driver pages include examples for the public APIs.

Where Next