@minnowdb/core

Columnar SQL engine for the browser: real SQL over immutable snapshots, durable on IndexedDB or OPFS, with no WebAssembly to download. It runs in the thread that constructs it (MinnowDatabase), or in a worker through MinnowDatabaseClient, which mirrors the everyday query, write, migration, live-query, snapshot, and maintenance APIs.

minnowdb.com — documentation, a live console, and benchmarks you run yourself.

npm install @minnowdb/core

• SQL is the foundation — parser, planner, optimizer, and a batch-based executor written here, with no SQLite or DuckDB underneath. The supported surface ships as a checked-in feature matrix (@minnowdb/core/sql-feature-matrix.json) that the engine is tested against.
• One schema, in TypeScript — table(), column, and schema() declare tables once and migrate() evolves them without rewriting stored data.
• Full-text search — MATCH / BM25 on any column, with no index DDL to write.
• Live queries — liveQueries() hands back a set whose subscribers get a fresh result after any commit that could have changed it, across tabs.
• Durable browser storage — compressed columnar blocks on IndexedDB or OPFS, atomic multi-tab commits, snapshot reads, compaction and GC.
• Memory-aware queries — execution works in batches under a budget you set, and sorts and grouped results can spill to storage. The budget is not a hard heap limit yet.
• Subpath exports — /storage, /storage/toolkit, /client, /worker, /worker-protocol, /transactions, /plan, /block-format, /testing.

Guides, the API reference, and the SQL feature matrix live in the docs, which are the single source of truth. Building with an agent? minnowdb.com/agent-rules.md is a short rules file to drop into an AGENTS.md, and llms.txt indexes the documentation in markdown.

Every @minnowdb package shares a major version and moves independently inside it, so install them on the same major. See Versioning.

License

MIT