
Recast Pro
Messy AI output → clean, valid, explained JSON
Most of the time when I ask an LLM for JSON, what comes back is almost right — but not quite.
Single quotes instead of double quotes.
Trailing commas.
Python-style True / False / None.
Markdown fences with some extra prose.
Truncated responses.
Or the whole thing wrapped the way n8n sometimes exports it.
Any one of those makes the output invalid. Downstream tools reject it. I ended up fixing the same issues by hand over and over, especially when the data was going into n8n, Make, or a script.
I got tired of black-box “JSON fixers” that either silently changed things or gave no explanation at all. So I built Recast.
It runs entirely in the browser. Nothing is uploaded.
You paste the messy output and get:
• Valid JSON
• A plain-English list of every change it made
• An explicit check that nothing was dropped
That last part mattered a lot to me. I wanted to know the tool wasn’t quietly losing data.
Free tier is 10 cleans per day. Pro ($8/month) removes the limit and unlocks Batch mode.
Live here: https://recast-indol.vercel.app
Curious what failure modes other people are still hitting with LLM JSON. What’s the most annoying one you keep running into?
About
I got tired of fixing broken AI JSON by hand and distrusting black-box tools with my data. Recast exists so people can clean LLM output, see every change, and know nothing was dropped — without uploading anything.

Comment