JSONSheet：作者以 Next.js 16 建立 22 款免費瀏覽器資料轉換工具

Every time I needed to turn an API response into a CSV, or paste a JSON blob out of ChatGPT into something usable, I ended up on a tool site that either wanted me to upload my data, buried the tool under 15 ads, or choked on quoted fields.

So I built JSONSheet — 22 converters that run 100% in your browser. No upload. No signup. No server-side processing at all.

What it does

JSON ↔ CSV with proper RFC 4180 parsing (quoted fields, embedded commas/newlines, escaped quotes)
JSON ↔ YAML — clean block-style output, correct quoting of ambiguous values
JSON → XML and JSON → JSONL (NDJSON) for pipelines and fine-tuning datasets
JSON ↔ Markdown tables — for READMEs, Notion, GitHub issues
CSV/JSON → SQL INSERTs (Postgres & MySQL, with CREATE TABLE + type inference)
JSON → TypeScript interfaces — nested objects become named interfaces
Flatten / unflatten JSON to dot-notation paths
JSON escape/unescape, Base64 encode/decode (UTF-8 safe), TSV → JSON (paste straight from Excel/Sheets), CSV dedupe & transpose, CSV → HTML table, JSON ↔ URL query string

The "AI-paste" feature

The part I use daily: paste JSON straight out of ChatGPT/Claude — markdown fences, "Here's your data:" prose and all — and the tool strips the noise before converting. One step instead of four. I wrote a full guide about this workflow: Convert ChatGPT / Claude output to CSV the clean way.

The privacy angle (and why it's also cheap)

Everything is a pure TypeScript function running in your tab. There is no API to call, so:

Your data never leaves your device — genuinely, not as a marketing line
The whole site is a static export — nginx serves ~2MB of files, the VPS idles at 0% CPU
No DB, no auth, no PII to protect — the security surface is tiny

Stack

Next.js 16.4.3, static export (output: 'export') — no Node runtime needed in production
React 19 + TypeScript, zero UI libraries — the converters are plain functions
76 unit tests on the converter core (npm test in the repo)
Deployed on a small VPS behind Cloudflare with Let's Encrypt; deploys via a 30-line bash script (no CI/CD needed at this scale)

Try it

👉 jsonsheet.com · Source: github.com/kami2k1/jsonsheet

If you hit a file it chokes on, open an issue with a sample — the CSV parser is RFC 4180-tested but real-world data always finds new edge cases.