Fix broken JSON from ChatGPT, Claude & other LLMs
Paste the JSON your AI assistant returned — fences, smart quotes, truncation and all — and braceful turns it into valid JSON you can use. It runs entirely in your browser, so the response never leaves this page.
Free forever · No account · 2-minute setup
Try it right here — your JSON never leaves this page
This is the actual viewer, running entirely in your browser. Open DevTools, watch the Network tab: zero requests. Load a sample or paste your own.
Why LLM JSON breaks — and how braceful fixes it
Language models generate JSON as text, not as a real data structure, so the output picks up artifacts a strict parser rejects. These are predictable, and braceful handles each one:
- Markdown code fences. Models love to wrap output in
```json … ```and add a line of explanation before or after. braceful strips the fences and surrounding prose and keeps the JSON. - Smart / curly quotes. Chat UIs and the models
themselves often emit
“curly”quotes instead of straight"quotes, which JSON won't accept. braceful converts them back. - Mid-response truncation. The single most common failure: the model hits a token limit and stops mid-string, leaving an unterminated value and unclosed braces. braceful closes the open structures so the rest is usable.
- Trailing commas, single quotes, unquoted keys. The same ordinary JSON mistakes humans make, which models reproduce. braceful normalizes all of them.
This matters most when you're piping LLM output into code. A response
that fails JSON.parse halfway through a pipeline is a
frustrating, intermittent bug; pasting it here to see exactly what was
malformed — and getting valid JSON back — turns a guessing game into a
two-second fix. And because it's all on-device, the prompt and the data
in that response are never uploaded.
{
name: "order-92",
"items": [42, 17,]
"paid": true,
"note": "truncated at sav{
"name": "order-92",
"items": [42, 17],
"paid": true,
"note": "truncated at sav"
}- quoted unquoted key
- removed trailing comma
- closed truncated string + object
This is the AI-specific case of braceful's general repair broken JSON capability. If the model dumped a very large document, see opening large JSON files without freezing, and once it parses you can format & beautify it offline.
Fixing LLM JSON — questions
Why is JSON from ChatGPT or Claude so often invalid?
LLMs generate text, not data structures, so their JSON picks up text artifacts: they wrap it in markdown ``` code fences, add a sentence of explanation before or after it, use curly "smart" quotes instead of straight ones, leave trailing commas, and — most commonly — get cut off mid-response when they hit a length limit, leaving an unterminated string and unclosed braces. Any one of these makes JSON.parse throw.
What LLM-specific breakage does braceful fix?
The artifacts LLMs add on top of ordinary JSON mistakes: it strips the surrounding markdown fences and prose, converts smart/curly quotes back to straight quotes, removes trailing commas, quotes unquoted keys, and closes structures that were truncated when the model ran out of room — so a half-finished response becomes valid JSON you can actually use.
Is my AI conversation or prompt uploaded to fix it?
No. braceful repairs the JSON entirely in your browser on your device — there is no upload and no server. The free viewer makes zero network requests, so pasting an LLM response (which may contain proprietary prompts or data) is safe. You can verify it in DevTools → Network: the request count never moves.
Does it work with any LLM, not just ChatGPT?
Yes — the breakage is the same shape across ChatGPT, Claude, Gemini, Llama and others, because they all emit JSON as generated text. braceful fixes the artifacts, not a vendor-specific format, so it works regardless of which assistant produced the output.
Still have a question? sherazp995@gmail.com
Turn half-finished AI JSON into valid JSON.
Add braceful and fix malformed LLM output the moment you get it — fences, smart quotes, truncation and all — offline, in the browser.
Free forever · No account · 2-minute setup