FrameKit

DataFrames for TypeScript that don't make you wish you were using Python.

FrameKit is a TypeScript-first DataFrame engine for Node.js. Strong types, predictable null semantics, lazy and eager execution, and production-grade I/O — without the "just use pandas" compromise.

Try it in the browser

framekit-js Playground →

Write TypeScript directly in the browser — no install, no setup. Monaco editor with FrameKit types, live execution via esbuild-wasm, and a table view of results.

Table of Contents

• Why FrameKit
• Feature Overview
• Install
• 60-Second Example
• Eager and Lazy APIs
• I/O Matrix
• Core and Compat APIs
• Documentation
• Comparison
• Benchmarks and Quality
• Project Standards
• License

Why FrameKit

• Strong TypeScript ergonomics for both row-level and expression-level transformations.
• Shared semantics across eager and lazy execution.
• Practical data engineering surface: joins, reshape, aggregations, window ops, and streaming formats.
• Compatibility layer (framekit-js/compat) to ease migration from Arquero-style pipelines.

Feature Overview

Install

npm install framekit-js

60-Second Example

import { DataFrame, col, op } from 'framekit-js';

const sales = DataFrame.fromRows([
  { region: 'EU', product: 'A', qty: 2, price: 10 },
  { region: 'EU', product: 'B', qty: 1, price: 25 },
  { region: 'US', product: 'A', qty: 4, price: 10 },
  { region: 'US', product: 'B', qty: 3, price: 25 },
]);

const result = sales
  .withColumn('revenue', col<number>('qty').mul(col<number>('price')))
  .groupBy('region')
  .agg({
    total_revenue: col('revenue').sum(),
    avg_revenue: col('revenue').mean(),
  })
  .sortBy('total_revenue', 'desc');

const quality = sales.groupBy().agg({
  qty_price_corr: op.corr(col('qty'), col('price')),
});

console.log(result.toArray());
console.log(quality.toArray());

Expected output:

// result.toArray()
[
  { region: 'US', total_revenue: 115, avg_revenue: 57.5 },
  { region: 'EU', total_revenue: 45,  avg_revenue: 22.5 }
]

// quality.toArray()
[
  { qty_price_corr: 0.7071067811865476 }
]

Eager and Lazy APIs

import { DataFrame, col } from 'framekit-js';

const source = DataFrame.fromRows([
  { city: 'Rome', temp: 26 },
  { city: 'Milan', temp: 31 },
  { city: 'Turin', temp: 28 },
]);

const eager = source.filter(col<number>('temp').gt(27));

const lazy = source.lazy().filter(col<number>('temp').gt(27)).select('city', 'temp').collect();

I/O Matrix

Core and Compat APIs

• framekit-js: native API (DataFrame, expressions, lazy plans, I/O).
• framekit-js/compat: migration-friendly verbs (derive, rollup, fold, orderby) and helpers (all, not, range, desc).

Start here for migration examples: docs/guides/migration-arquero.md.

Documentation

• Getting started: docs/getting-started/quickstart.md
• Guides: docs/guides/
• Cookbook examples: docs/cookbook/
• API references: docs/reference/

Comparison

Benchmarks and Quality

• Benchmark compare runners and outputs live under tests/benchmarks/ and benchmarks/results/.
• CI includes smoke and nightly benchmark workflows.
• Regression checks are available in the benchmark harness.

Latest Benchmark Snapshot (FrameKit vs Arquero)

Current checked-in snapshot (BENCH_ROWS=50000, BENCH_ITERS=10, BENCH_WARMUP=3):

Run Benchmarks Locally

# Fast smoke check
npm run bench:smoke

# Full compare suite
npm run bench:full

For human-readable per-operation reports, see benchmarks/results/*.md.

Benchmark outputs are directional and environment-sensitive; validate on your own hardware and runtime before drawing hard conclusions.

Project Standards

• Changelog: CHANGELOG.md
• Contributing guide: CONTRIBUTING.md
• Security policy: SECURITY.md
• Code of conduct: CODE_OF_CONDUCT.md

License

MIT - see LICENSE.