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

| Area | Highlights | | ------------------ | --------------------------------------------------------------------------- | | DataFrame core | select, drop, filter, sortBy, groupBy, join, unique, sample | | Expressions | arithmetic, comparison, logical, null helpers, string/date accessors | | Aggregations | sum, mean, count, min, max, std, corr, list/mode, first/last | | Reshape | pivot, melt, explode, spread, unroll, transpose, concat | | Utility transforms | assign, relocate, lookup, reify, impute, derive, apply | | Lazy execution | query plans, optimizer passes, explain support, eager executor | | I/O | CSV, JSON, NDJSON, Arrow, Excel, Parquet, SQL export |

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

| Format | Read | Write | | ------------------ | ---- | ----- | | CSV | Yes | Yes | | JSON | Yes | Yes | | NDJSON (streaming) | Yes | Yes | | Arrow IPC | Yes | Yes | | Excel | Yes | Yes | | Parquet | Yes | Yes | | SQL (INSERT text) | No | Yes |

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

| | FrameKit | Arquero | Danfo.js | Polars (Node) | | ---------------- | -------- | -------- | -------- | ------------- | | TypeScript-first | Yes | No | Partial | No (Rust FFI) | | Lazy execution | Yes | No | No | Yes | | Expression API | Yes | No | No | Yes | | Null semantics | Explicit | Implicit | Implicit | Explicit | | I/O formats | 7 | 2 | 4 | 6 | | Browser support | Yes | Yes | Yes | No | | Zero native deps | Yes | Yes | No | No |

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):

| Operation | FrameKit Median (ms) | Arquero Median (ms) | Relative (FrameKit/Arquero) | Source | | --------- | -------------------: | ------------------: | --------------------------: | ----------------------------------------- | | Filter | 26.9228 | 36.6342 | 0.73x | benchmarks/results/compare-filter.json | | Sort | 25.5620 | 50.6992 | 0.50x | benchmarks/results/compare-sort.json | | GroupBy | 3.9674 | 5.0675 | 0.78x | benchmarks/results/compare-groupby.json | | Join | 20.4444 | 61.5447 | 0.33x | benchmarks/results/compare-join.json | | Reshape | 17.2577 | 65.6563 | 0.26x | benchmarks/results/compare-reshape.json | | Window | 22.2838 | 64.1068 | 0.35x | benchmarks/results/compare-window.json |

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.