Reference - Turso

Turso offers three TypeScript packages:

	`@tursodatabase/database`	`@tursodatabase/sync`	`@tursodatabase/serverless`	`@libsql/client`
Use case	Local / embedded database	Local database + cloud sync	Remote access (servers, containers, serverless, edge)	ORM support (Drizzle, Prisma)
Engine	Turso Database (rewrite)	Turso Database (rewrite)	Turso Database	libSQL (SQLite fork)
Dependencies	Native (Node.js, WASM)	Native (Node.js)	`fetch` only — zero native deps	Requires Node.js or `/web` subpath
Concurrent writes	Yes (MVCC)	Yes (MVCC)	Planned	Not supported
Sync	—	push/pull (local-first)	—	Embedded Replicas (writes go to cloud primary)
ORM support	Drizzle (beta)	—	—	Drizzle, Prisma, and others

Starting a new project? Use @tursodatabase/database for local/embedded use, @tursodatabase/sync for local + cloud sync, or @tursodatabase/serverless for any application that connects to a remote Turso Cloud database (Node.js servers, Docker containers, serverless functions, edge runtimes). Using an ORM? Use @libsql/client — it’s production-ready and supported by Drizzle, Prisma, and others. Drizzle has beta support for @tursodatabase/database for local/embedded use. The following runtime environments are known to be compatible:

Node.js version 12 or later
Deno
CloudFlare Workers
Netlify & Vercel Edge Functions

@tursodatabase/database

For local and embedded use. Built on the Turso Database engine with concurrent writes (MVCC) and async I/O.

Installing

npm install @tursodatabase/database

Initializing

import { connect } from "@tursodatabase/database";

const db = await connect("app.db");

In-memory databases are also supported:

const db = await connect(":memory:");

Querying

const stmt = db.prepare("SELECT * FROM users");
const users = await stmt.all();

const insert = db.prepare("INSERT INTO users (username) VALUES (?)");
await insert.run("alice");

Encryption

Encrypt local databases at rest using the encryption option:

import { connect } from "@tursodatabase/database";

const db = await connect("encrypted.db", {
  encryption: {
    cipher: "aegis256",
    hexkey: "b1bbfda4f589dc9daaf004fe21111e00dc00c98237102f5c7002a5669fc76327",
  },
});

Supported ciphers: aegis256, aegis256x2, aegis128l, aegis128x2, aegis128x4, aes256gcm, aes128gcm. Encrypted databases cannot be read as standard SQLite databases — you must use the Turso Database engine to open them.

Turso Cloud databases can also be encrypted with bring-your-own-key — learn more.

@tursodatabase/sync

For local database with cloud sync. All reads and writes happen locally; use push() to send changes to the cloud and pull() to fetch remote changes.

Installing

npm install @tursodatabase/sync

Initializing

import { connect } from "@tursodatabase/sync";

const db = await connect({
  path: "./app.db",
  url: process.env.TURSO_DATABASE_URL,
  authToken: process.env.TURSO_AUTH_TOKEN,
});

On the first run, the local database is automatically bootstrapped from the remote. See Turso Sync for full details.

Push and Pull

// Push local writes to Turso Cloud
await db.push();

// Pull remote changes to local database
const changed = await db.pull();

Checkpoint

Compact the local WAL to bound disk usage while preserving sync state:

await db.checkpoint();

Stats

const s = await db.stats();
console.info({
  cdcOperations: s.cdcOperations,
  mainWalSize: s.mainWalSize,
  networkReceivedBytes: s.networkReceivedBytes,
  networkSentBytes: s.networkSentBytes,
  lastPullUnixTime: s.lastPullUnixTime,
  lastPushUnixTime: s.lastPushUnixTime,
  revision: s.revision,
});

@tursodatabase/serverless

The recommended package for any application that connects to a remote Turso Cloud database — Node.js servers, Docker containers, serverless functions (AWS Lambda, Vercel Functions), and edge runtimes (Cloudflare Workers, Deno Deploy). Uses only fetch — zero native dependencies, works everywhere fetch is available.

Installing

npm install @tursodatabase/serverless

Initializing

import { connect } from "@tursodatabase/serverless";

const conn = connect({
  url: process.env.TURSO_DATABASE_URL,
  authToken: process.env.TURSO_AUTH_TOKEN,
});

For compatibility with the @libsql/client API, use the compat module:

import { createClient } from "@tursodatabase/serverless/compat";

const client = createClient({
  url: process.env.TURSO_DATABASE_URL,
  authToken: process.env.TURSO_AUTH_TOKEN,
});

Querying

const stmt = await conn.prepare("SELECT * FROM users");
const rows = await stmt.all();

const stmt2 = await conn.prepare("SELECT * FROM users WHERE id = ?");
const row = await stmt2.get([1]);

@libsql/client

The @libsql/client package is built on libSQL, the open-source fork of SQLite that powers Turso Cloud today. It is production-ready, battle-tested, and the right choice when you need ORM integration beyond Drizzle (e.g., Prisma) or are working with an existing @libsql/client-based codebase.

With @libsql/client Embedded Replicas, reads are local and writes are sent to the cloud primary, then reflected back to the replica. Embedded Replicas are fully supported. For new projects that need sync, we recommend @tursodatabase/sync with Turso Sync.

Installing

Begin by installing the @libsql/client dependency in your project:

npm install @libsql/client

Initializing

import { createClient } from "@libsql/client";

export const turso = createClient({
  url: process.env.TURSO_DATABASE_URL,
  authToken: process.env.TURSO_AUTH_TOKEN,
});

If you’re using libsql locally or an sqlite file, you can ignore passing authToken.

In-Memory Databases

import { createClient } from "@libsql/client";

const client = createClient({
  url: ":memory:",
});

Local Development

You can work locally using an SQLite file and passing the path to createClient:

import { createClient } from "@libsql/client";

const client = createClient({
  url: "file:path/to/db-file.db",
  authToken: "...",
});

The @libsql/client/web does not support local file URLs.

Embedded Replicas

For workloads that need offline writes, bidirectional sync, or multi-writer convergence, we recommend @tursodatabase/sync — both reads and writes are local, and you sync explicitly with push() / pull().

You can work with embedded replicas by passing your Turso Database URL to syncUrl:

import { createClient } from "@libsql/client";

const client = createClient({
  url: "file:path/to/db-file.db",
  syncUrl: "libsql://[databaseName]-[organizationSlug].turso.io",
  authToken: "...",
});

Embedded Replicas only works where you have access to the file system.

Manual Sync

await client.sync();

Periodic Sync

import { createClient } from "@libsql/client";

const client = createClient({
  url: "file:path/to/db-file.db",
  syncUrl: "libsql://[databaseName]-[organizationSlug].turso.io",
  syncInterval: 60,
  authToken: "...",
});

Encryption

For new projects, we recommend @tursodatabase/database for local encryption — it is built on the Turso Database engine with better performance and concurrent write support.

TypeScript

import { createClient } from "@libsql/client";

const db = createClient({
  url: "file:encrypted.db",
  encryptionKey: process.env.ENCRYPTION_KEY,
});

Encrypted databases appear as raw data and cannot be read as standard SQLite databases. You must use the libSQL client for any operations — learn more.

Concurrency

By default, the client performs up to 20 concurrent requests:

import { createClient } from "@libsql/client";

const client = createClient({
  concurrency: 10,
});

Response

Each method listed below returns a Promise<ResultSet>:

Property	Type	Description
`rows`	`Array<Row>`	An array of Row objects containing the row values, empty for write operations
`columns`	`Array<string>`	An array of strings with the names of the columns in the order they appear in each Row, empty for write operations
`rowsAffected`	`number`	The number of rows affected by a write statement, `0` otherwise
`lastInsertRowid`	`bigint \| undefined`	The ID of a newly inserted row, or `undefined` if there is none for the statement

Simple query

You can pass a string or object to execute() to invoke a SQL statement:

const result = await client.execute("SELECT * FROM users");

Placeholders

libSQL supports the use of positional and named placeholders within SQL statements:

const result = await client.execute({
  sql: "SELECT * FROM users WHERE id = ?",
  args: [1],
});

const result = await client.batch(
  [
    {
      sql: "INSERT INTO users VALUES (?)",
      args: ["Iku"],
    },
  ],
  "write",
);

libSQL supports the same named placeholder characters as SQLite — :, @ and $.

Transaction Modes

Mode	SQLite command	Description
`write`	`BEGIN IMMEDIATE`	The transaction may execute statements that read and write data. Write transactions executed on a replica are forwarded to the primary instance, and can’t operate in parallel.
`read`	`BEGIN TRANSACTION READONLY`	The transaction may only execute statements that read data (select). Read transactions can occur on replicas, and can operate in parallel with other read transactions.
`deferred`	`BEGIN DEFERRED`	The transaction starts in read mode, then changes to write as soon as a write statement is executed. This mode change may fail if there is a write transaction currently executing on the primary.

Batch Transactions

Use batch() to send several SQL statements in a single call. Each item is either a SQL string or a { sql, args } object:

const result = await conn.batch([
  { sql: "INSERT INTO users (name) VALUES (?)", args: ["Iku"] },
  { sql: "INSERT INTO users (name) VALUES (?)", args: ["Iku 2"] },
]);

By default a batch is not transactional. Each statement runs in its own autocommit step, so a failure partway through leaves the statements that already succeeded committed. Pass a mode as the second argument to run the batch atomically. The statements are wrapped in BEGIN <mode> and COMMIT (rolling back on any failure) and dispatched as a single request, so the whole batch completes in one round trip:

const result = await conn.batch(
  [
    { sql: "INSERT INTO users (name) VALUES (?)", args: ["Iku"] },
    { sql: "INSERT INTO users (name) VALUES (?)", args: ["Iku 2"] },
  ],
  "immediate",
);

mode accepts the same values as transaction(): "deferred", "immediate", "exclusive", and "concurrent". Each maps to the matching BEGIN <mode> statement. When batch() runs inside a transaction() callback, the mode argument is ignored and the surrounding transaction is reused. batch() resolves to an object with rowsAffected (the total number of rows affected across every statement) and lastInsertRowid (the rowid of the last successful insert).

When you pass a mode, batch() manages the surrounding BEGIN, COMMIT, and ROLLBACK. Do not include your own transaction-control statements (BEGIN, COMMIT, ROLLBACK, SAVEPOINT, RELEASE) in the batch.

Interactive Transactions

Interactive transactions in SQLite ensure the consistency of a series of read and write operations within a transaction’s scope. These transactions give you control over when to commit or roll back changes, isolating them from other client activity.

Method	Description
`execute()`	Similar to `execute()` except within the context of the transaction
`commit()`	Commits all write statements in the transaction
`rollback()`	Rolls back the entire transaction
`close()`	Immediately stops the transaction

try {
  const userId = "user123";
  const withdrawalAmount = 500;

  const transaction = await client.transaction("write");

  const balanceResult = await transaction.execute({
    sql: "SELECT balance FROM accounts WHERE userId = ?",
    args: [userId],
  });

  const currentBalance = balanceResult.rows[0]["balance"] as number;

  if (currentBalance >= withdrawalAmount) {
    await transaction.execute({
      sql: "UPDATE accounts SET balance = balance - ? WHERE userId = ?",
      args: [withdrawalAmount, userId],
    });
  } else {
    console.log("Insufficient funds");
    await transaction.rollback();
    return;
  }

  await transaction.commit();
} catch (e) {
  console.error(e);
}

Interactive transactions in libSQL lock the database for writing until committed or rolled back, with a 5-second timeout. They can impact performance on high-latency or busy databases.

ATTACH

You can attach multiple databases to the current connection using the ATTACH attachment:

import { createClient } from "@libsql/client";

const client = createClient({
  url: process.env.TURSO_DATABASE_URL,
  authToken: process.env.TURSO_AUTH_TOKEN,
});

const txn = await db.transaction("read");

await txn.execute('ATTACH "<database-id>" AS attached');

const rs = await txn.execute("SELECT * FROM attached.users");

Make sure to allow ATTACH and create a token with the permission to attach a database — learn more

​@tursodatabase/database

​Installing

​Initializing

​Querying

​Encryption

​@tursodatabase/sync

​Installing

​Initializing

​Push and Pull

​Checkpoint

​Stats

​@tursodatabase/serverless

​Installing

​Initializing

​Querying

​@libsql/client

​Installing

​Initializing

​In-Memory Databases

​Local Development

​Embedded Replicas

​Manual Sync

​Periodic Sync

​Encryption

​Concurrency

​Response

​Simple query

​Placeholders

​Transaction Modes

​Batch Transactions

​Interactive Transactions

​ATTACH

@tursodatabase/database

Installing

Initializing

Querying

Encryption

@tursodatabase/sync

Installing

Initializing

Push and Pull

Checkpoint

Stats

@tursodatabase/serverless

Installing

Initializing

Querying

@libsql/client

Installing

Initializing

In-Memory Databases

Local Development

Embedded Replicas

Manual Sync

Periodic Sync

Encryption

Concurrency

Response

Simple query

Placeholders

Transaction Modes

Batch Transactions

Interactive Transactions

ATTACH