DEV Community

ArshTechPro
ArshTechPro

Posted on

Turso: A Rust Rewrite of SQLite. Setup Guide and Whether It's Worth Your Time

If you have spent any time in the SQLite-adjacent corner of GitHub lately, you have probably seen the name Turso pop up. The project Turso describes itself as an in-process SQL database written in Rust that is compatible with SQLite.

This post walks through what the project actually is, how to get it running locally and in the cloud, and an honest take on whether you should bother adopting it right now.

First, clear up the naming confusion

Before touching any code, it helps to know there are three related but different things, and mixing them up is the most common source of confusion in the community:

  • libSQL: Turso's original project, a fork of SQLite written in C. It is production ready today and currently powers Turso Cloud.
  • Turso (the database engine): The project at the GitHub link above. A clean-room rewrite of SQLite in Rust, started as a side project called "Limbo." It is currently in beta and is explicitly positioned as the successor to libSQL, not a side experiment.
  • Turso Cloud: The managed, hosted database-as-a-service product built by the same company. It currently runs on libSQL, with the Rust engine gradually being rolled into it.

If you just want a managed SQLite-compatible database in production today, libSQL plus Turso Cloud is the safer starting point. If you want to try the next-generation engine itself, that is the tursodatabase/turso repo this article focuses on.

What's actually in it for developers

This is the part most setup tutorials skip, so here is the practical pitch:

  • It still talks SQLite. Same SQL dialect, same file format, same C API surface where practical. If you know SQLite, you already mostly know Turso.
  • Concurrent writes. SQLite's biggest historical complaint is its single-writer lock. Turso ships an experimental MVCC engine with BEGIN CONCURRENT, letting multiple writers commit without blocking each other on simple cases.
  • Async by design. The engine is built around asynchronous I/O (including Linux io_uring), instead of bolting async behavior on top of a synchronous core the way most SQLite drivers have to.
  • Native vector search. Built-in approximate vector search for embeddings and RAG-style workloads, no extension required.
  • Change Data Capture (CDC). Real-time tracking of every change made to the database, which is the backbone of Turso's newer sync story.
  • Full-text search, powered by the tantivy library, again without an extension.
  • Encryption at rest and incremental view maintenance (via DBSP), both still marked experimental.
  • Multi-language support out of the box: official packages for Rust, JavaScript/TypeScript, Python, Go, with more community drivers (PHP/Laravel, .NET) appearing in the ecosystem.
  • A genuinely open contribution model. Unlike SQLite's core, which uses a closed test suite, Turso is built with public deterministic simulation testing and fuzzing against SQLite's own bytecode output, specifically so outside contributors can verify their changes.

In short: if you have ever wanted SQLite with safer memory handling, real concurrent writes, and vector search baked in, this project is aimed squarely at you.

Setting it up locally

1. Install the CLI

curl --proto '=https' --tlsv1.2 -LsSf \
  https://github.com/tursodatabase/turso/releases/latest/download/turso_cli-installer.sh | sh
Enter fullscreen mode Exit fullscreen mode

Windows users can use the PowerShell installer linked from the releases page instead.

2. Launch the shell

tursodb
Enter fullscreen mode Exit fullscreen mode

This drops you into an interactive shell connected to a transient in-memory database:

turso> CREATE TABLE users (id INT, username TEXT);
turso> INSERT INTO users VALUES (1, 'alice');
turso> INSERT INTO users VALUES (2, 'bob');
turso> SELECT * FROM users;
1|alice
2|bob
Enter fullscreen mode Exit fullscreen mode

Use .open FILENAME inside the shell if you want a persistent file instead of an in-memory database.

3. Prefer Docker or building from source?

# build and run the dev version
cargo run

# or with Docker
make docker-cli-build && make docker-cli-run
Enter fullscreen mode Exit fullscreen mode

4. Pick a language binding

Rust:

cargo add turso
Enter fullscreen mode Exit fullscreen mode
let db = Builder::new_local("sqlite.db").build().await?;
let conn = db.connect()?;
let res = conn.query("SELECT * FROM users", ()).await?;
Enter fullscreen mode Exit fullscreen mode

JavaScript/TypeScript:

npm install @tursodatabase/database
Enter fullscreen mode Exit fullscreen mode
import { connect } from '@tursodatabase/database';
const db = await connect('sqlite.db');
const stmt = db.prepare('SELECT * FROM users');
const users = stmt.all();
Enter fullscreen mode Exit fullscreen mode

Python:

uv pip install pyturso
Enter fullscreen mode Exit fullscreen mode
import turso
con = turso.connect("sqlite.db")
cur = con.cursor()
res = cur.execute("SELECT * FROM users")
print(res.fetchone())
Enter fullscreen mode Exit fullscreen mode

Go:

go get turso.tech/database/tursogo
Enter fullscreen mode Exit fullscreen mode
conn, _ := sql.Open("turso", "sqlite.db")
defer conn.Close()
Enter fullscreen mode Exit fullscreen mode

That's the entire local setup. No server, no daemon, no config file: it runs in-process exactly like SQLite always has.

Setting up sync to the cloud

If you want a local-first app that stays in sync with a remote primary, Turso's newer sync package is the relevant piece:

npm install @tursodatabase/sync
Enter fullscreen mode Exit fullscreen mode
import { connect } from "@tursodatabase/sync";

const db = await connect({
  path: "./my-app.db",
  url: "libsql://your-db.turso.io",
  authToken: "your-token",
});
Enter fullscreen mode Exit fullscreen mode

You will need a Turso Cloud database and auth token for the url and authToken fields, both available through the turso CLI or the Turso dashboard. This sync layer is what replaces libSQL's older "embedded replicas" feature, and the project's own team now recommends it over the legacy approach for any new sync work, citing faster replica bootstrap and lower bandwidth use thanks to the async, page-level sync protocol.

The honest caveats

The repository itself is upfront about this, and it is worth repeating rather than glossing over: this software is in beta. The maintainers explicitly warn that it may contain bugs and unexpected behavior, and they recommend caution with production data and proper backups.

A few specifics worth knowing before you commit to it:

  • MVCC has real limitations today. Indexes cannot currently be created on MVCC-enabled databases, and the entire dataset is eagerly loaded into memory on first access, so very large databases can be slow to start and memory-hungry.
  • Query result ordering isn't guaranteed to match SQLite in every case, which can matter if your tests or application logic implicitly depend on row order.
  • Several headline features are explicitly experimental: encryption at rest, incremental computation, and multi-process WAL coordination all carry that label in the project's own documentation.
  • libSQL is still the production recommendation for most things. The maintainers say plainly that libSQL is production ready while the Rust engine is not, even though it is evolving quickly.

None of this is a knock against the project. It is a young, fast-moving rewrite being built in the open, with public fuzzing against SQLite's own bytecode output as a compatibility check. That is a genuinely rigorous approach. It just means "beta" is not a formality here.

Is it worth a try?

For side projects, prototypes, and anything where you can tolerate some rough edges, yes, it is worth trying. The setup cost is close to zero: one shell script, no servers, and your existing SQLite knowledge mostly transfers directly. If your workload involves frequent concurrent writes, embeddings and vector search, or you simply want to watch where SQLite is heading next, this is a good time to kick the tires.

A reasonable way to think about it: Turso is not trying to be "yet another database." It is trying to answer a fairly specific question, whether SQLite's simplicity can be carried forward into a memory-safe, async-native, concurrent-write world without losing what made SQLite SQLite.

Top comments (0)