Minnow ><(((('>

A columnar SQL database for the browser. Minnow runs PostgreSQL-style SQL over durable data in IndexedDB or OPFS. It needs no server and has no WebAssembly module to download or compile.

npm install @minnowdb/core

minnowdb.com has the full documentation, a live SQL and TypeScript console, and benchmarks that run in your browser.

Experimental API. Minnow is in 0.x, so minor releases can include API and SQL breaking changes. Stored data is stable: block format 2, snapshot formats 1 to 3, IndexedDB schema 8, and OPFS layout 13 are versioned separately and locked — an incompatible future writer must use a new format number. Pin exact package versions. See Versioning, the v1 support policy, and the Changelog.

Why Minnow

• PostgreSQL-style SQL. Use familiar joins, CTEs, window functions, grouping sets, upserts, RETURNING, triggers, stored generated columns, exact decimals, nested JSON/JSONB, zoneless DATE, arrays, enums, sequences, savepoints, and more. The PostgreSQL compatibility guide lists exact matches, differences, extensions, and exclusions.
• Fast analytical reads over application data. Compressed column blocks let a query read only the columns it needs and skip blocks that cannot match. Secondary indexes speed up selective lookups and ordered reads.
• Durable browser storage. IndexedDB (the default) and OPFS adapters publish writes atomically and default to strict durability: IndexedDB requests the final disk flush, while OPFS performs it. Every query reads one stable snapshot, even when another tab commits at the same time, and origin-persistence policy is explicit for applications that cannot accept automatic quota eviction.
• One writer at a time, across tabs. Write scopes, batch writes, SQL transactions, and schema changes from any number of tabs and workers take turns through Web Locks before they read anything, so concurrent writes land in a serial order without an application-side queue or retry loop. Callbacks run once; a tab that stops inside its turn is reported, never bypassed.
• Plain JavaScript. The engine with its larger durable adapter is about 441 KiB gzipped, with no Wasm download, compile step, special headers, or server process.
• Direct SQL or Kysely. Run PostgreSQL-style SQL through the engine API or use Kysely through @minnowdb/kysely. Kysely's DB type derives from the same schema used for migration, so tables are declared once, aggregate and built-in results are inferred without output generics, and typed JSON helpers build nested projections. Both paths use the same SQL engine.
• Built for responsive applications. A ready-made worker client keeps query work off the UI thread, and the page itself carries no SQL engine: the client, schema declarations, Kysely dialect, and React hooks come to about 31 KiB gzipped, not counting Kysely and React. Pull-driven cursors transfer one columnar page at a time; batch execution and spillable sorts and aggregates keep large working sets under a configurable budget.
• Search and reactive views. Full-text search uses MATCH and BM25 without index DDL. Live queries refresh after relevant commits, including commits from other tabs, with typed Kysely, keyed-change, bounded-window, and Suspense/SWR-friendly React adapters. A commit updates every affected query in the same turn, so a screen renders it once.
• Keys that can change. A record created offline under a temporary key moves to its server key with one UPDATE. References follow in the same commit, and the old key keeps reading and writing the row for 10 days, so nothing that held it loses the record. See Offline sync and rekeying.
• Portable and extensible. Snapshots move a committed database between stores. The public storage interface includes a shared test kit for custom adapters.

Minnow is designed for browser-local application data that needs both keyed reads and analytical queries. It does not provide a server, replication, or multi-device sync.

Packages

Package	Purpose
@minnowdb/core	SQL engine, schema management, workers, storage adapters, and snapshots.
@minnowdb/kysely	Kysely dialect with typed JSON, search, streaming, and live-query tools.
@minnowdb/react	React external-store hook for live queries.
@minnowdb/export	Streaming CSV and NDJSON over direct or worker cursors.
@minnowdb/devtools	Optional SQL console and database explorer for development and embedded demos.

Start with Installation and Your first query. The Comparison page explains when IndexedDB, Dexie, SQLite Wasm, PGlite, or DuckDB-Wasm is a better fit.

For coding agents, use llms.txt or add the concise Minnow agent rules to your project.

Development

npm install
npm test               # unit tests and the standard SQLLogicTest profile
npm run check          # format, types, lint, build, coverage, and SQL tests
npm run test:sql:full  # full supported SQLLogicTest profile
npm run test:browser   # library and site tests in real browsers, including the SQLLogicTest corpus and multi-tab interaction plans
npm run simulate:interactions # one replayable interaction plan against the shadow model
npm run soak           # generative suites on fresh seeds, the simulators included
npm run test:browser:conformance # full SQL corpus and long interaction plans in all three browsers
npm run test:browser:pos # isolated sustained POS correctness and latency checks
npm run test:consumer  # install packed libraries in a clean Vite app and run it
npm run site:dev       # docs, live console, and browser benchmarks

See Testing and benchmarks for the complete test map, Versioning for release commands, the v1 support policy for the 1.x contract, and the Changelog for release notes and migrations.

License

MIT — see LICENSE.