EMLib is a research-oriented TypeScript monorepo built around the operator

$$ \mathrm{eml}(x, y) = \exp(x) - \ln(y) $$

inspired by All elementary functions from a single binary operator by Andrzej Odrzywolek.

This repository turns the paper into a reusable TypeScript library, a Bun + React playground, and a small documentation set for contributors.

• Repository: TokenFlow-chat/emlib
• Demo: eml.tokenflow.chat

• packages/emlib: parsing, analysis, lowering, rewriting, evaluation, JSON graph serialization, and experimental search/training utilities for elementary expressions
• src/: a browser playground for exploring standard ASTs, pure EML forms, 3D force-directed expression graphs, and selected experiments
• docs/: implementation notes and a local copy of the reference paper

The repository is optimized for local Bun workflows, and README claims are intentionally limited to behavior implemented in code and covered by tests.

• Bun

bun install
bun run dev

To build production assets:

bun run build

Static output is written to dist/.

bun run dev        # local dev server with HMR
bun run start      # run the app in production mode
bun run build      # build the static site to dist/
bun run test       # run Bun tests across the workspace
bun run fmt        # format with oxfmt
bun run fmt:check  # verify formatting
bun run lint       # run oxlint
bun run typecheck  # tsgo typecheck + emlib build
bun run check      # lint + typecheck + test + build

import {
  analyzeExpr,
  evaluate,
  evaluateLossless,
  parse,
  reduceTokens,
  serializeExpr,
  toPureEml,
  toString,
  valueToExpr,
} from "emlib";

const expr = parse("exp(x) - ln(y)");

console.log(analyzeExpr(expr));
console.log(toString(toPureEml(expr)));
console.log(toString(reduceTokens(expr)));
console.log(evaluate(expr, { x: 0.5, y: 2 }));
console.log(toString(valueToExpr(evaluateLossless(parse("(1 + 2*i) / (3 - 4*i)")))));
console.log(serializeExpr(expr));

For syntax support, API details, and behavior notes, see packages/emlib/README.md.

Issues and pull requests are welcome.

1. Install dependencies with bun install.
2. Run bun run check before opening a pull request.
3. Keep README and docs changes aligned with current code and tests.
4. Be explicit about whether a change affects mathematical semantics, library APIs, playground behavior, or documentation.
5. Add or update tests when changing lowering, rewriting, evaluation, synthesis, or training behavior.

• Paper title: All elementary functions from a single binary operator
• arXiv entry: 2603.21852

This repository is licensed under the Apache License 2.0.