NumPy-inspired array operations in TypeScript/WebAssembly
npm install numjs-wasmNumPy-inspired n-dimensional array operations in TypeScript, with performance-critical operations compiled to WebAssembly.
- NumPy-compatible API — familiar function names and semantics (zeros, ones, linspace, matmul, fft, mean, reshape, ...)
- WebAssembly acceleration — BLAS, LAPACK, FFT, sorting, statistics, and random number generation implemented in C and compiled to WASM
- Full TypeScript types — complete type definitions with documented parameters
- Dual module support — ESM and CommonJS, works in Node.js and browsers
- 600+ functions across linear algebra, FFT, statistics, random, polynomials, string operations, masked arrays, and more
``bash`
npm install numjs-wasm
`typescript
import {
loadWasmModule,
array,
zeros,
ones,
linspace,
reshape,
add,
matmul,
mean,
} from "numjs-wasm";
// Initialize the WASM module (required once before use)
await loadWasmModule();
// Create arrays
const a = array([1, 2, 3, 4, 5, 6]);
const b = zeros([3, 3]);
const c = ones([2, 3]);
const d = linspace(0, 1, 100);
// Reshape and compute
const matrix = reshape(a, [2, 3]);
const result = add(matrix, c);
const product = await matmul(reshape(a, [2, 3]), reshape(a, [3, 2]));
// Statistics
const avg = mean(d);
`
`typescript
import { linalg } from "numjs-wasm";
const result = await linalg.matmul(a, b);
const { values, vectors } = await linalg.eig(matrix);
const solution = await linalg.solve(coefficients, constants);
const determinant = await linalg.det(matrix);
`
matmul, dot, inv, det, solve, eig, eigh, svd, qr, cholesky, norm, cond, lstsq, matrix_rank, matrix_power, cross, kron, tensordot, ...
`typescript
import { fftModule } from "numjs-wasm";
const spectrum = fftModule.fft(signal);
const freqs = fftModule.fftfreq(n, dt);
`
fft, ifft, rfft, irfft, fft2, ifft2, fftn, ifftn, fftfreq, rfftfreq, fftshift, ifftshift, hfft, ihfft
`typescript
import { default_rng, Generator } from "numjs-wasm";
const rng = default_rng(42);
const samples = rng.normal(0, 1, [1000]);
const uniform = rng.uniform(0, 1, [100]);
`
Bit generators: PCG64, MT19937, Philox, SFC64. Distributions: normal, uniform, exponential, gamma, beta, binomial, poisson, and 20+ more.
`typescript
import { ma } from "numjs-wasm";
const masked = ma.array(data, { mask: [false, false, true, false] });
const avg = ma.average(masked);
`
`typescript
import { Polynomial, Chebyshev } from "numjs-wasm";
const p = new Polynomial([1, 2, 3]); // 1 + 2x + 3x^2
const roots = p.roots();
`
Polynomial, Chebyshev, Legendre, Hermite, HermiteE, Laguerre with full arithmetic, fitting, roots, and conversions.
- Strings (strings) — element-wise string operations on arraysrec
- Record Arrays () — structured/tabular data with named fieldstesting
- Testing () — assert_allclose, assert_array_equal, assert_raises, ...
| Category | Functions |
| ------------------ | ------------------------------------------------------------------------------------------------------------------------------------ |
| Array Creation | array, zeros, ones, empty, full, arange, linspace, logspace, geomspace, eye, identity, diag, meshgrid, ... |reshape
| Manipulation | , transpose, concatenate, stack, split, flip, roll, rot90, tile, repeat, pad, ... |add
| Math | , subtract, multiply, divide, power, sqrt, exp, log, sin, cos, tan, abs, clip, ... |mean
| Statistics | , median, std, var_, min, max, sum, prod, histogram, percentile, quantile, ... |sort
| Sorting | , argsort, argmax, argmin, searchsorted, partition, ... |all
| Logic | , any, where, logical_and, logical_or, logical_not, ... |equal
| Comparison | , greater, less, allclose, isclose, isnan, isinf, ... |unique
| Set Operations | , union1d, intersect1d, setdiff1d, isin, ... |save
| I/O | , load, loadtxt, savetxt, genfromtxt, frombuffer, ... |pi
| Constants | , e, inf, nan, euler_gamma, newaxis |
The documentation site serves machine-readable files following the llms.txt convention:
- llms.txt — project overview with module links
- llms-full.txt — complete API reference (all 600+ functions with signatures and descriptions)
The numwasm-mcp package provides a Model Context Protocol server that gives AI coding assistants searchable access to the full API docs. It ships with a bundled docs index — no network calls at runtime.
Tools exposed:
- search_numwasm_docs — search by function name, module, category, or keywordlist_numwasm_modules
- — list all modules and categories
Claude Desktop — add to claude_desktop_config.json:
`json`
{
"mcpServers": {
"numwasm-docs": {
"command": "npx",
"args": ["-y", "numwasm-mcp"]
}
}
}
Claude Code — add .mcp.json to your project root:
`json`
{
"mcpServers": {
"numwasm-docs": {
"command": "npx",
"args": ["-y", "numwasm-mcp"]
}
}
}
- Node.js >= 18
- pnpm
- Emscripten (for building WASM from C sources)
- Python 3 + NumPy (for generating test fixtures and running comparison benchmarks)
`bashInstall dependencies
pnpm install
$3
bash
Run all tests
npm testWatch mode
npm run test:watchRun comparison tests against NumPy reference vectors
npm run test:compareBrowser tests (Playwright)
npm run test:browser
`$3
bash
Full benchmark pipeline (build + NumPy + NumJS + report)
npm run benchmarkIndividual steps
npm run benchmark:numpy
npm run benchmark:numjs
npm run benchmark:combine
`$3
bash
Generate TypeDoc JSON
npm run docsDev server
npm run dev:docsFull static build (SSG + sitemap + llms.txt)
cd docs-site && pnpm run build:ssg
`Project Structure
src/
ts/ TypeScript implementation (NDArray, ufuncs, modules)
wasm/ C source files compiled to WebAssembly
scripts/
build-wasm.sh Emscripten build script
tests/
ts/ Vitest unit & integration tests
browser/ Playwright browser tests
python/ NumPy test fixture generators
benchmark/
ts/ TypeScript benchmark suites
python/ NumPy comparison benchmarks
docs-site/ React + Vite documentation website
dist/ Build output (library + WASM binary)
MIT