@slothdb/wasm

Fast embedded SQL database in WebAssembly. Query Parquet, CSV, JSON, Arrow, SQLite, and Excel files directly from SQL — no server, no install, 1.3 MB wasm.

• Website: https://slothdb.org
• Try it live: https://slothdb.org/playground
• GitHub: https://github.com/SouravRoy-ETL/slothdb

SlothDB is a full analytical SQL engine compiled to WebAssembly. It reads seven file formats natively, runs in Node ≥18 and every modern browser, and runs 1.1×–8.6× faster than DuckDB on the benchmarked formats in its native build. The WASM build is single-threaded; expect browser-scale latencies, not native-scale.

Install

npm install @slothdb/wasm

No native dependencies. Works on Linux, macOS, Windows (and WSL).

Quick start

import { SlothDB } from '@slothdb/wasm';
import fs from 'node:fs/promises';

const db = await SlothDB.create();

// Load a file into the virtual filesystem:
const parquet = await fs.readFile('./sales.parquet');
db.loadFile('/data/sales.parquet', parquet);

// Query it:
const { columns, rows, ms } = db.query(`
    SELECT region, SUM(revenue) AS total
    FROM '/data/sales.parquet'
    GROUP BY region
    ORDER BY total DESC
`);

console.log(columns);  // ['region', 'total']
console.log(rows);     // [['EU', '14184859'], ['NA', '12295714'], ...]
console.log(`ran in ${ms.toFixed(1)} ms`);

File formats supported

All seven formats work directly from SQL by file path — no CREATE TABLE, no COPY FROM.

| Format | Example | |---|---| | Parquet | SELECT * FROM '/data.parquet' | | CSV / TSV | SELECT * FROM '/data.csv' | | JSON / NDJSON | SELECT * FROM '/data.json' | | Apache Arrow IPC | SELECT * FROM '/data.arrow' | | Avro | SELECT * FROM '/data.avro' | | SQLite | SELECT * FROM sqlite_scan('/app.db', 'users') | | Excel (.xlsx) | SELECT * FROM '/report.xlsx' |

SQL features

SELECT, WHERE, GROUP BY, HAVING, ORDER BY, LIMIT, OFFSET. Joins (INNER / LEFT / RIGHT / FULL / NATURAL / USING). CTEs including recursive. Window functions with QUALIFY. Set operations (UNION / INTERSECT / EXCEPT). MERGE. CAST. 70+ scalar functions — string, math, date/time, regex. DuckDB-compatible DATE_TRUNC with all intervals (MICROSECOND through MILLENNIUM), plus MONTHNAME, DAYNAME, LAST_DAY, MAKE_DATE.

Full SQL reference: https://slothdb.org/docs.html

Browser usage

SlothDB works the same in the browser — the only differences are that you load files via fetch/<input type="file"> instead of fs, and the package needs to be served over HTTP (not file://).

import { SlothDB } from '@slothdb/wasm';

const db = await SlothDB.create();

const response = await fetch('/data/sales.parquet');
const buffer = new Uint8Array(await response.arrayBuffer());
db.loadFile('/data/sales.parquet', buffer);

const result = db.query("SELECT COUNT(*) FROM '/data/sales.parquet'");

With Vite / webpack / esbuild, slothdb.wasm is loaded automatically alongside slothdb.js via the bundler's asset handling. No special config required for the common cases; see https://slothdb.org/docs.html for bundler-specific notes.

API

SlothDB.create(options?)

Boot a new in-memory database. Returns Promise<SlothDB>.

• options.moduleOptions — pass-through to the underlying Emscripten Module factory. Rarely needed.

db.query(sql)

Execute a SQL statement. Returns:

{
    columns: string[];
    rows: (string | null)[][];
    ms: number;         // engine-side execution time
    rowCount: number;
}

All non-NULL values come back as strings to preserve full precision. Cast at the JS layer (Number(v), BigInt(v)) as needed.

Throws Error on SQL error; the thrown error includes .ms and .sql properties for diagnostics.

db.loadFile(path, data)

Write a byte buffer (or string) to a virtual path. Accepts Uint8Array, ArrayBuffer, or string. Absolute paths required (starting with /).

db.readFile(path)

Read a file back from the virtual filesystem. Returns Uint8Array.

db.version()

Returns the engine version string, e.g. "0.1.5".

Limitations of the WASM build

• Single-threaded. Native SlothDB uses parallel readers; the WASM build does not (SharedArrayBuffer/COOP-COEP overhead isn't worth it for playground-scale workloads).
• No HTTP reads. FROM 'https://example.com/data.csv' returns an error. Fetch the file in JS and load it via loadFile instead.
• Memory cap. Starts at 64 MB, grows to a 2 GB hard cap. Plenty for most analytical workloads; too small for a 10 GB Parquet.

For the full feature set (threading, HTTP reads, extensions) use the native library: https://slothdb.org.

License

MIT. Copyright (c) 2026 Sourav Roy. See LICENSE.