A Tauri plugin for libsql with built-in AES-256-CBC encryption, Drizzle ORM support, and a browser-compatible migration runner.
The fastest way to understand the full workflow (Drizzle ORM, browser-safe migrations, optional encryption, and Turso sync) is to run the demo app.
# build js plugin API package first
pnpm install
pnpm build
# run the example app
cd examples/todo-list
pnpm install
pnpm run tauri devFor a complete walkthrough, see examples/todo-list/README.md.
- Try the Example App
- Why this plugin?
- Features
- Installation
- Quick Start
- Database Location
- Drizzle ORM Integration
- Migrations
- Encryption
- API Reference
- Permissions
- Comparison with @tauri-apps/plugin-sql
- Turso / Remote Database
- Bundle Size
- Using AI to Integrate This Plugin
- Project Structure
Using raw SQL in Rust is verbose, and Rust ORMs (Diesel, SeaORM) require schema definitions in Rust, don't compose well with TypeScript frontends, and add significant build complexity. For Tauri apps where the real logic lives in TypeScript, you want to write database code in TypeScript too.
Drizzle ORM is excellent — type-safe queries, a clean migration system, great ergonomics. But it normally requires a Node.js or Bun runtime to open database files directly. Tauri's WebView has no such runtime.
This plugin solves that with Drizzle's sqlite-proxy pattern: Drizzle generates SQL, the proxy sends it via Tauri's invoke() to the Rust plugin, and the Rust plugin executes it with libsql. Your TypeScript code uses full Drizzle ORM with zero Node.js dependency.
Drizzle's built-in migrator reads .sql files from disk at runtime using Node's fs module — which doesn't exist in a browser/WebView context. Two workarounds exist:
- Tauri resource folder — bundle files as app resources and read them via Tauri's asset protocol. Works but requires extra Tauri config.
- Vite
import.meta.glob(this plugin's approach) — Vite bundles the SQL file contents directly into the JavaScript at build time. No runtime filesystem access needed, no extra config.
// Vite resolves these at build time — the SQL text is inlined into the JS bundle
const migrations = import.meta.glob<string>("./drizzle/*.sql", {
eager: true,
query: "?raw",
import: "default",
});
await migrate("sqlite:myapp.db", migrations);The migrate() function in this plugin receives the pre-loaded SQL strings, tracks applied migrations in a __drizzle_migrations table, and runs pending ones in order.
@tauri-apps/plugin-sql (which uses sqlx) has no encryption support. This plugin uses libsql's native AES-256-CBC encryption with no extra native libraries or FFI wrappers required.
- Full SQLite compatibility via libsql
- Native encryption — AES-256-CBC, configured once at the plugin level or per-database
- Drizzle ORM integration — sqlite-proxy pattern with
createDrizzleProxy - Migration runner — browser-safe
migrate()that bundles SQL files at build time via Vite - API compatible with
@tauri-apps/plugin-sqlwhere applicable - Cross-platform: macOS, Windows, Linux, iOS, Android
Tested on
- MacOS
- Windows
- Linux
- iOS
- Android
[dependencies]
tauri-plugin-libsql = "0.1.0"npm install tauri-plugin-libsql-api
# or
pnpm add tauri-plugin-libsql-api// src-tauri/src/lib.rs
// Default: databases resolve relative to current working directory
tauri::Builder::default()
.plugin(tauri_plugin_libsql::init())
.run(tauri::generate_context!())
.expect("error while running tauri application");To store databases in a fixed location:
use std::path::PathBuf;
let config = tauri_plugin_libsql::Config {
base_path: Some(PathBuf::from("/path/to/data")),
encryption: None,
};
tauri::Builder::default()
.plugin(tauri_plugin_libsql::init_with_config(config))
.run(tauri::generate_context!())
.expect("error while running tauri application");import { Database } from "tauri-plugin-libsql-api";
const db = await Database.load("sqlite:myapp.db");
await db.execute("INSERT INTO users (name) VALUES ($1)", ["Alice"]);
const users = await db.select<{ id: number; name: string }[]>(
"SELECT * FROM users",
);
await db.close();Relative paths (e.g. sqlite:myapp.db) resolve against base_path:
- Default:
std::env::current_dir()— the directory you launch the Tauri process from - Custom: set
base_pathin the plugin config (see above) - Absolute paths are used as-is
- In-memory:
sqlite::memory:
Relative paths are normalised (.. components are folded) and must remain within base_path. A path that would escape it (e.g. sqlite:../../secret) is rejected with an error.
import { drizzle } from "drizzle-orm/sqlite-proxy";
import { createDrizzleProxy } from "tauri-plugin-libsql-api";
import * as schema from "./schema";
const db = drizzle(createDrizzleProxy("sqlite:myapp.db"), { schema });
const users = await db.select().from(schema.users);createDrizzleProxy lazily loads the database connection on first use, so you don't need to call Database.load() separately when using it.
import { createDrizzleProxyWithEncryption } from "tauri-plugin-libsql-api";
const db = drizzle(
createDrizzleProxyWithEncryption({
path: "sqlite:encrypted.db",
encryption: {
cipher: "aes256cbc",
key: myKey32Bytes, // number[] | Uint8Array, 32 bytes
},
}),
{ schema },
);The standard drizzle-orm/sqlite-proxy/migrator reads from the filesystem at runtime, which doesn't work inside a Tauri WebView. This plugin ships a migrate() function that instead accepts SQL content pre-bundled by Vite's import.meta.glob.
1. Define your schema (src/lib/schema.ts):
import { integer, sqliteTable, text } from "drizzle-orm/sqlite-core";
export const users = sqliteTable("users", {
id: integer("id").primaryKey({ autoIncrement: true }),
name: text("name").notNull(),
});2. Configure drizzle-kit (drizzle.config.ts):
import { defineConfig } from "drizzle-kit";
export default defineConfig({
dialect: "sqlite",
schema: "./src/lib/schema.ts",
out: "./drizzle",
});3. Generate migration files:
npx drizzle-kit generate
# creates drizzle/0000_init.sql, drizzle/0001_add_column.sql, etc.4. Run migrations on startup:
import { Database, migrate } from "tauri-plugin-libsql-api";
// Vite bundles these SQL files into the app at build time
const migrations = import.meta.glob<string>("./drizzle/*.sql", {
eager: true,
query: "?raw",
import: "default",
});
// Startup sequence: load → migrate → query
await Database.load("sqlite:myapp.db");
await migrate("sqlite:myapp.db", migrations);
// Now safe to query
const db = drizzle(createDrizzleProxy("sqlite:myapp.db"), { schema });- Creates a
__drizzle_migrationstracking table if it doesn't exist - Parses migration filenames by their numeric prefix (
0000_,0001_, etc.) - Applies only pending migrations in order
- Records each applied migration by filename
# 1. Edit src/lib/schema.ts
# 2. Generate new migration
npx drizzle-kit generate
# 3. New migration runs automatically on next app launchawait migrate("sqlite:myapp.db", migrations, {
migrationsTable: "__my_migrations", // default: '__drizzle_migrations'
});Configure once in Rust — the frontend never handles the key:
let config = tauri_plugin_libsql::Config {
base_path: None,
encryption: Some(tauri_plugin_libsql::EncryptionConfig {
cipher: tauri_plugin_libsql::Cipher::Aes256Cbc,
key: my_32_byte_key, // Vec<u8>, exactly 32 bytes
}),
};const key = new Uint8Array(32);
crypto.getRandomValues(key);
const db = await Database.load({
path: "sqlite:secrets.db",
encryption: {
cipher: "aes256cbc",
key: Array.from(key), // number[] or Uint8Array
},
});Security notes:
- AES-256-CBC requires exactly 32 bytes
- Store keys in the OS keychain or secure storage — lost key = lost data
- Plugin-level encryption is preferred; it keeps keys out of JavaScript
// Simple
const db = await Database.load("sqlite:myapp.db");
// With encryption
const db = await Database.load({
path: "sqlite:myapp.db",
encryption: { cipher: "aes256cbc", key: myKey },
});const result = await db.execute("INSERT INTO todos (title) VALUES ($1)", [
"Buy milk",
]);
// result.rowsAffected, result.lastInsertIdconst rows = await db.select<{ id: number; title: string }[]>(
"SELECT * FROM todos WHERE completed = $1",
[0],
);Executes multiple SQL statements atomically in a single transaction. Use for DDL or bulk DML. Statements must not use bound parameters ($1 placeholders) — use execute() for parameterised queries.
await db.batch([
"CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT NOT NULL)",
"CREATE INDEX idx_users_name ON users(name)",
]);Pulls the latest changes from the Turso remote into the local replica. No-op for local-only databases (returns without error). Requires the replication feature.
await db.sync();await db.close();import { migrate } from "tauri-plugin-libsql-api";
const migrations = import.meta.glob<string>("./drizzle/*.sql", {
eager: true,
query: "?raw",
import: "default",
});
await migrate("sqlite:myapp.db", migrations);Returns a sqlite-proxy callback for use with drizzle(). Lazy-loads the connection.
Same as above but with encryption config.
import { getConfig } from "tauri-plugin-libsql-api";
const { encrypted } = await getConfig();Add to your tauri.conf.json:
{
"plugins": {
"libsql": {}
}
}Or configure granular capabilities:
{
"identifier": "libsql:default",
"permissions": [
"libsql:allow-load",
"libsql:allow-batch",
"libsql:allow-execute",
"libsql:allow-select",
"libsql:allow-close"
]
}| Feature | tauri-plugin-libsql | @tauri-apps/plugin-sql |
|---|---|---|
| SQLite | ✅ libsql | ✅ sqlx |
| Encryption | ✅ AES-256-CBC built-in | ❌ |
| Drizzle ORM | ✅ | ✅ |
| Migration runner | ✅ browser-safe | ❌ |
| MySQL / PostgreSQL | ❌ | ✅ |
| API compatibility | Partial | Full |
The plugin supports two remote connection modes powered by libsql.
A local SQLite file stays in sync with a Turso cloud database. Queries read from the local file (fast, offline-capable), writes sync to the remote.
⚠️ Limitation: Embedded replica encryption is currently broken onmainDue to an upstream libsql bug, local encryption is silently disabled when using embedded replicas (syncUrl). The V2 sync protocol (which Turso always uses) switches to a code path that drops theencryption_config, leaving the local replica file unencrypted even if you pass anencryptionconfig. See Issue #1 for details.Fix available on
fix/sync-encryptionbranch: That branch vendors a fork of libsql that threadsencryption_configthrough the V2 sync path and encrypts the bootstrapped replica viasqlite3_rekey. It works but cannot be published to crates.io (path dependencies are not allowed). Use it via git:tauri-plugin-libsql = { git = "https://github.com/HuakunShen/tauri-plugin-libsql", branch = "fix/sync-encryption", features = ["replication", "encryption"] }Workarounds (on
main):
- Use the
fix/sync-encryptionbranch (recommended if you need encrypted replicas)- Use pure remote mode (no local file) if you don't need offline access
- Use local-only databases with encryption for sensitive local data
- Accept the unencrypted replica (Turso access control still protects the remote data)
1. Enable the replication feature in your app's Cargo.toml:
tauri-plugin-libsql = { version = "0.1.0", features = ["replication"] }2. Load with syncUrl and authToken:
import { Database, migrate } from "tauri-plugin-libsql-api";
const db = await Database.load({
path: "sqlite:local.db", // local replica file
syncUrl: "libsql://mydb-org.turso.io",
authToken: "your-turso-auth-token",
});
// Sync on demand (e.g. on app resume / network reconnect)
await db.sync();On Database.load(), an initial sync pulls the latest data from Turso into the local file. Subsequent sync() calls pull incremental changes.
With Drizzle ORM:
const migrations = import.meta.glob<string>("./drizzle/*.sql", {
eager: true,
query: "?raw",
import: "default",
});
const db = await Database.load({
path: "sqlite:local.db",
syncUrl: "libsql://mydb-org.turso.io",
authToken: import.meta.env.VITE_TURSO_AUTH_TOKEN,
});
await migrate(db.path, migrations);
const drizzleDb = drizzle(createDrizzleProxy(db.path), { schema });All queries execute on Turso directly — no local file. Requires network for every query.
Enable the remote feature:
tauri-plugin-libsql = { version = "0.1.0", features = ["remote"] }const db = await Database.load({
path: "libsql://mydb-org.turso.io",
authToken: "your-turso-auth-token",
});For most Tauri apps, embedded replica is the better choice — it works offline and is significantly faster for reads.
Note on
batch()with embedded replicas: libsql'sexecute_batch()does not correctly route writes through the embedded replica layer in some versions. The plugin uses individualexecute()calls inside an explicitBEGIN/COMMITtransaction to avoid this.
Note on URL validation: libsql's builder calls
unwrap()internally on the sync URL and can panic on a malformed value (e.g. leading/trailing whitespace, wrong scheme). The plugin wraps this incatch_unwindso a bad URL surfaces as a proper error instead of hanging the IPC indefinitely.
Based on the included Todo List demo app (macOS, aarch64, release build):
| Format | With encryption | Without encryption |
|---|---|---|
.app bundle |
15 MB | 15 MB |
.dmg installer |
6.0 MB | 5.9 MB |
Disabling encryption saves essentially nothing — the AES cipher code is negligible compared to the SQLite native library that's always present. The encryption feature flag still exists to avoid compiling encryption-related code if you want to enforce at compile time that no database can be encrypted.
Encryption is a default feature. To opt out, disable default features and select only what you need:
Cargo.toml (in your Tauri app):
tauri-plugin-libsql = { version = "0.1.0", default-features = false, features = ["core"] }Available features:
| Feature | Default | Description |
|---|---|---|
core |
✅ | Local SQLite databases (always required) |
encryption |
✅ | AES-256-CBC encryption via libsql |
replication |
❌ | libsql replication support (adds TLS) |
remote |
❌ | Remote database support (planned, see below) |
When encryption is disabled, passing an EncryptionConfig to Database.load() returns an error at runtime. The TypeScript API surface is unchanged — no rebuild of your JS code needed.
A SKILL.md file is included at the root of this repository. It contains structured context about the plugin's architecture, startup sequence, migration workflow, encryption patterns, and common errors — written for AI coding assistants (Claude Code, Cursor, Copilot, etc.).
Copy SKILL.md into your project's .claude/skills/tauri-plugin-libsql/ directory:
mkdir -p .claude/skills/tauri-plugin-libsql
cp /path/to/tauri-plugin-libsql/SKILL.md .claude/skills/tauri-plugin-libsql/Claude Code discovers skills automatically. Once copied, you can prompt naturally:
"Add a
notestable to my Tauri app using tauri-plugin-libsql. Include the schema, migration, and startup sequence."
Claude will apply the correct startup order, use import.meta.glob for migrations, and handle the drizzle proxy pattern without extra guidance.
Paste the contents of SKILL.md directly into your system prompt or context window, then describe what you want to build. The skill covers enough context for the AI to generate correct, working code on the first attempt.
tauri-plugin-libsql/
├── src/ # Rust plugin
│ ├── lib.rs # Plugin init, command registration
│ ├── commands.rs # load, execute, select, close, ping
│ ├── wrapper.rs # DbConnection around libsql
│ ├── decode.rs # libsql::Value → serde_json::Value
│ ├── models.rs # Cipher, EncryptionConfig, QueryResult
│ ├── error.rs # Error types
│ ├── desktop.rs # Desktop config & base_path
│ └── mobile.rs # Mobile stub
├── guest-js/ # TypeScript source
│ ├── index.ts # Database class, getConfig, re-exports
│ ├── drizzle.ts # createDrizzleProxy, createDrizzleProxyWithEncryption
│ └── migrate.ts # migrate() — browser-safe migration runner
├── permissions/ # Tauri permission files
├── examples/todo-list/ # Demo: Todo app with Drizzle + migrations (15 MB .app / 6 MB .dmg)
├── SKILL.md # AI skill context for Claude Code and other assistants
├── build.rs
├── Cargo.toml
└── package.json
