Convert JSON to official TOON, compare tokens and decode back locally.
The JSON to TOON converter turns your JSON data into the TOON format (Token-Oriented Object Notation) and back again, using the format's official implementation. TOON is designed to take fewer tokens in LLM prompts: it writes the field names of an array of objects once, like a table header, and lists the values in rows, instead of repeating the keys in every element, the way JSON does. Nested objects use YAML-style indentation, arrays of simple values become a compact line with a length marker, and much of the quoting and punctuation disappears. You can choose the delimiter (comma, tab or pipe), enable "safe key folding" (which merges single-key objects into a dotted path) and decode TOON back into JSON. On every conversion, the tool measures input and output characters and tokens, with the savings as a percentage, using a tokenizer of the kind used by language models. Everything runs locally: JSON, TOON and values are never sent to servers or analytics.
The biggest savings come from arrays of uniform objects: instead of repeating the keys in every item (like JSON), TOON writes a header in the form key[N]{field1,field2}: and lists the values in indented rows, similar to a CSV with a header and a row count. Nested objects use YAML-style indentation (key: value), arrays of simple values become a line like tags[3]: a,b,c, and unnecessary quotes are omitted.
Two options fine-tune the JSON → TOON output: the value delimiter (comma, tab or pipe) and "safe key folding", which collapses chains of single-key objects into a dotted path, like a.b.c: value. In the reverse direction (TOON → JSON), "expand paths" rebuilds those paths into nested objects. The conversion uses the format's official implementation.
Example 1, tabular array: the input {"users":[{"id":1,"name":"Ada"},{"id":2,"name":"Grace"}]} becomes a block that starts with the header users[2]{id,name}: and has two indented rows, "1,Ada" and "2,Grace". The id and name fields appear once in the header, and each user is a row, that is where tokens are saved.
Example 2, nesting and key folding: the input {"user":{"profile":{"name":"Ana"}},"tags":["a","b","c"]}, with safe key folding, becomes two lines: "user.profile.name: Ana" and "tags[3]: a,b,c". The single-key objects collapse into a dotted path, and the array of simple values sits on one compact line with the length marker [3].
TOON makes the most sense in LLM prompts, where every token counts and you send structured data in the context, especially large lists of objects that share the same fields, the scenario where the header-plus-rows table saves the most. The tool shows characters and tokens before and after, so you measure the gain instead of assuming it.
About the limits: the savings depend heavily on the shape of the data, very irregular or deeply nested data saves little, and fewer characters do not always mean fewer tokens, because that depends on the model's tokenizer. JSON remains the standard for APIs, storage and interoperability; use TOON at the specific point where token cost matters, and validate the decode back to check fidelity.
Paste the code into your HTML and the tool shows up on your page, without J-Kit's navigation and ads. It still runs in the browser of whoever visits your site.
<iframe
src="https://jkit.tools/embed/en-US/json-to-toon"
width="100%"
height="600"
style="border:0"
loading="lazy"
title="JSON to TOON"
></iframe>These references help contextualize formulas, standards, APIs and limitations used on this page. They do not replace professional validation when a result has legal, financial, medical or operational impact.
Yes. It converts JSON to TOON using the format's official implementation and also decodes TOON back into JSON. You can switch the mode manually or use automatic detection, and on every conversion the tool shows the character and token counts on both sides, with the savings as a percentage.
Conversion, decode and token counting happen in your browser. JSON, TOON and values are never sent to analytics.