vastgrid

vastgrid opens huge CSV, TSV, Parquet, JSON Lines and Excel files like a spreadsheet, lets people fix them, and saves them safely. It is a canvas data grid for React with engines for the browser and Node.js, and the same grid comes as a command, a VS Code extension, an MCP App and a Jupyter widget.

Try the demo · Open a file in the browser · Documentation · 日本語

• Any size. Rows are read a block at a time as you scroll, and Parquet files are read in place, also over HTTP Range requests. A 100-million-row Parquet file sorts on DuckDB with its first rows ready in under a second. Cells are drawn on a canvas, and scrolling renders no React.
• A spreadsheet's way of working. Excel's keys and selections, sorting, filters, conditions, search and find, groups, pivots and totals, formulas, formats, copy, paste and fill.
• Edits you can trust. Typed input is checked against the column's type, writes are checked for conflicts and can be undone, and saving a CSV or TSV file rewrites only the edited fields and keeps every other byte. A file that changes on disk is followed, and is never saved over. Values that cannot survive a save, including empty strings in CSV and TSV, are refused with the edits kept.
• One state, typed commands. What the grid shows is a document (the view and its columns) and a session (where the person is), controlled or not. Every change is a command that a toolbar, a host or an AI agent can dispatch.
• Engines you choose. vastgrid-js in memory, in the page or a Web Worker; vastgrid-duckdb on Node.js; or your own, through a small set of messages.
• Nine languages (English, Japanese, Chinese in simplified and traditional characters, Korean, Spanish, Brazilian Portuguese, French and German), each its own import (vastgrid/locales/ja); light and dark themes through CSS variables, ARIA grid semantics and full keyboard operation.

Where to use it

Where	How	Engine
A React app	npm install vastgrid vastgrid-js (quick start)	in memory, a worker, or a server
The browser	vastgrid.yhay81.com/open/: drop a file; it stays in the browser	vastgrid-js in a Web Worker
The command line	npx vastgrid-cli sales.parquet	DuckDB on this computer
VS Code	the extension vastgrid (yhay81.vastgrid-vscode)	DuckDB in the extension
An AI chat	the MCP server vastgrid-mcp with its MCP App	DuckDB in the MCP server
A Jupyter notebook	pip install vastgrid, then vastgrid.show(df)	DuckDB in the kernel

Where vastgrid opens files compares them: formats, editing, saving and exporting.

Quick start

npm install vastgrid vastgrid-js

import { DataGrid, localTransport } from "vastgrid";
import "vastgrid/styles.css";
import { memoryEngine } from "vastgrid-js";

const transport = localTransport(
  memoryEngine({
    tables: {
      orders: {
        columns: [
          { name: "id", type: "integer" },
          { name: "customer", type: "string" },
          { name: "price", type: "decimal", precision: 10, scale: 2 },
          { name: "ordered", type: "date" },
        ],
        rows: [
          [1, "Acme", "12.50", "2026-01-31"],
          [2, "Globex", "8.00", "2026-02-01"],
        ],
        editable: true,
      },
    },
  }),
);

export function Orders() {
  return <DataGrid transport={transport} table="orders" style={{ height: 480 }} />;
}

For files larger than the page should hold, run the memory engine in a Web Worker, or the DuckDB engine on a server. The guides cover state, editing and saving, formulas, grouping, exporting, theming and your own engine.

The DuckDB engine reads CSV with comma separators and standard double quotes. For other separators or an explicit header, pass open={{ read: { delimiter: ";", headerRow: 1 } }} to the grid; headerRow: 0 means no header. See the engine's reading options.

Packages

Package	Contents
vastgrid	The React grid: DataGrid, GridProvider and the hooks; vastgrid/core (the store and the transports, without React), vastgrid/server, vastgrid/locales/<tag> (the texts of each language), vastgrid/styles.css.
vastgrid-js	The memory engine, for the browser and Node.js: rows of typed JSON, Arrow, CSV, TSV, JSON Lines, Excel and Parquet (also over HTTP).
vastgrid-duckdb	The DuckDB engine for Node.js: files of any size, writing cells, saving CSV and TSV, following files, exports; an HTTP handler.
vastgrid-spec	The specification as code: types, plan checking, commands, message schemas, the conformance corpus.
vastgrid-sql	Plans as DuckDB SQL.
vastgrid-cli	npx vastgrid-cli <file>: a file in a grid in the browser, read by DuckDB on this computer.
vastgrid-mcp	An MCP server with an MCP App.
vastgrid-vscode	The VS Code extension; install 0.3.0's VSIX from the GitHub release.

The npm packages are released together at one version. vastgrid needs React 19.3 or later (an optional peer dependency: servers use vastgrid/core without React) and a browser with CSS light-dark(): Chrome and Edge 123, Firefox 120, Safari 17.5 or later. The engines need Node.js 20 or later; DuckDB runs on Linux (x64, arm64; glibc and musl), macOS (x64, arm64) and Windows (x64, arm64).

Performance

On an Apple M2 Pro. The grid on a million rows in the memory engine shows its first rows 36–49 ms after mounting and drops no frame while scrolling at 60 Hz, with no React commit. The DuckDB engine sorts a 100-million-row Parquet file (1.38 GB) with its first rows ready in 953 ms, and reads a sorted 10-million-row CSV file 256 rows at a time in 4 ms. Performance has every measure and how to repeat them.

Documentation

• vastgrid.yhay81.com: guides, the API reference, the demo and the open page.
• docs/: the specifications (spec.md for the engines and their messages, react.md for the React package), how each app is built, the benchmarks, and migrating from 0.1.
• CHANGELOG.md, and plans in the GitHub issues.

Contributing

npm install
npm run build      # every package, the CLI's page, the MCP App and the extension
npm test           # the unit tests, on the sources
npm run dev        # examples/files: the grid on the DuckDB engine, at http://127.0.0.1:5182

CONTRIBUTING.md has the checks and the guidelines for a change, and SECURITY.md how to report a vulnerability.

License

MIT