toon-asijs

TOON (Token-Oriented Object Notation) adapter for AsiJS.

TOON is a compact, token-efficient encoding of the JSON data model designed for LLM/AI clients — indentation instead of braces, tabular forms for uniform arrays, ~30–60% fewer tokens than JSON in typical cases, lossless for JSON-serializable values. This package wires the official @toon-format/toon SDK into the AsiJS formats layer, so the framework understands TOON natively: request bodies parse by Content-Type, responses (including errors and 404s) serialize to TOON, and Accept negotiation works out of the box.

bun add toon-asijs

Usage

import { Asi } from "asijs";
import { registerToonFormat } from "toon-asijs";

registerToonFormat(); // enable TOON: negotiation + parsing + setFormat("toon")

const app = new Asi({ format: "toon" }); // default response format = TOON
// or: app.setFormat("toon");

app.post("/config", async (ctx) => {
  const body = await ctx.parseBody(); // Content-Type: application/toon → parsed
  return body;                        // serialized to TOON
});

// Accept negotiation just works:
//   Accept: application/toon      → TOON response
//   Accept: application/json      → JSON response (even with TOON default)
//   Content-Type: application/toon → request body decoded from TOON

API

createToonFormat(options?) → DataFormat

Creates a TOON format object (structural match for AsiJS DataFormat). Pure — only needs @toon-format/toon, so it works without AsiJS in plain fetch servers or CLI tools.

import { createToonFormat } from "toon-asijs";
import { registerFormat } from "asijs";

registerFormat(createToonFormat()); // manual registration

| Option | Type | Default | Description | |---|---|---|---| | indentSize | number | 2 | Spaces per indentation level | | delimiter | "," \| "\t" \| "\|" | "," | Separator for array values / tabular rows | | strict | boolean | true | Strict decoding (array counts, indentation, duplicate keys) |

getToonFormat(options?) → DataFormat

Cached TOON format singleton — repeated calls return the same instance.

registerToonFormat(options?) → DataFormat

Creates the format and registers it with AsiJS (registerFormat), enabling setFormat("toon"), Content-Type: application/toon parsing and Accept negotiation. Idempotent.

MIME types

| Field | Value | |---|---| | Canonical content type | application/toon | | Aliases | text/toon, application/x-toon | | Extension | .toon |

Why TOON?

For LLM-facing endpoints (MCP tools, agent APIs, chat backends), token count is a real cost. TOON keeps the JSON data model — decode() returns plain JS objects/arrays/primitives, so TypeBox validation and AsiJS serializers work unchanged — while cutting tokens via:

• indentation instead of { } and ,
• tabular headers for uniform arrays: items[2]{sku,qty}: + rows
• unquoted keys and numbers where unambiguous

License

MIT