Language-agnostic lossless binary serializer. Encode values in JavaScript, decode them in other languages (Python / Rust / Go / …) and use them as-is.

• Lossless for everything the JS runtime can represent — cycles, shared identity, Map/Set with arbitrary keys, BigInt, typed arrays, Date, RegExp, Error, symbols, and more.
• Zero runtime dependencies. No Buffer / structuredClone / vm — only standard DataView / TextEncoder / BigInt.
• One format, many readers. The byte format is a single source of truth (spec/FORMAT.md), verified byte-for-byte across JS / Python / Rust / Go conformance ports.

A primary use case: take JS mock data (e.g. from zod-v4-mocks) and reuse it as test fixtures in other languages.

npm add greft-codec      # or: pnpm add greft-codec / yarn add greft-codec

import { encode, decode } from "greft-codec";

const bytes = encode(value); // value -> Uint8Array (lossless, cycles + shared identity)
const back = decode(bytes); // Uint8Array -> value

Losslessly serialize values the core rejects (class instances, domain types, …) by registering a TypeExtension:

import { encode, decode } from "greft-codec";

const types = [/* TypeExtension[] */];
const bytes = encode(value, { types });
const back = decode(bytes, { types });

WeakMap / WeakSet contents are supplied explicitly via a provider:

encode(value, { provider });

Inspect or hand-edit an encoded graph as JSON (identity-lossy; cycles rejected):

import { toJSON, fromJSON } from "greft-codec";

The package installs a graft binary:

npm i -g greft-codec
graft inspect <file.bin>   # tree view + tag histogram
graft diff <a.bin> <b.bin> # structural diff of two value graphs

One-off, without installing:

npx -p greft-codec graft inspect <file.bin>

decode(bytes, { maxNodes }) bounds work; malformed input (out-of-range refs, counts exceeding the buffer) is rejected rather than trusted.

• Format spec: spec/FORMAT.md
• Conformance ports (JS / Python / Rust / Go): conformance/README.md
• Changelog: CHANGELOG.md

ISC