What field types are supported?

uuid, int, float, name, firstName, lastName, email, url, bool, date, datetime, lorem, choice(a,b,c), regex.

Can I use this for AI fine-tuning?

Yes — pair it with our fine-tune validators to generate placeholder chat-format datasets and check the shape.

No — everything is generated locally. No external API calls, no real PII.

JSONL Generator

updated 12 May 2026

JSONL generator. Generate synthetic JSONL (JSON Lines) data from a tiny schema. Useful for testing pipelines, mocking APIs and stress-testing fine-tuning code without real data. Runs in your browser.

Schema

Records: Seed:

Schema syntax: field: type per line. Types: uuid, int(min,max), float(min,max), name, firstName, lastName, email, url, bool, date(2020,2026), datetime, lorem(words), choice(a,b,c), regex(pattern).

Schema

Output (JSONL)

Before you start

The generator is for synthetic test data — not real data. It's reproducible: the same Seed always produces the same output, so you can pin your fixtures.

How to use it

Edit the Schema: one line per field as name: type. Types accept arguments in parens.
Set how many records you want.
Optionally pin a Seed to make output reproducible.
Click Generate, then Copy or Download.

Field types

uuid — RFC-4122-ish v4 UUID string
int(min, max) — integer in [min, max] inclusive. Default 0..100.
float(min, max) — float to 4 decimal places.
bool — random true/false.
name, firstName, lastName — built-in name pool.
email — [email protected] style.
url — random URL on a built-in domain pool.
date(year_from, year_to) — ISO date YYYY-MM-DD.
datetime — ISO 8601 timestamp.
lorem(word_count) — placeholder text.
choice(a, b, c, …) — pick one literal value.
regex(pattern) — tiny generator covering \d, \w and character classes like [A-Z].

Example

Schema:

id: uuid
user: name
email: email
age: int(18, 80)
plan: choice(free, pro, team)
signed_up: date(2023, 2026)

Output (seed 42, 2 records):

{"id":"d36f...","user":"Maya Jones","email":"[email protected]","age":31,"plan":"pro","signed_up":"2024-08-14"}
{"id":"4b21...","user":"Hugo Lee","email":"[email protected]","age":67,"plan":"free","signed_up":"2025-12-02"}

Tips & common pitfalls

Nested fields use dot notation. profile.bio: lorem(10) produces {"profile": {"bio": "lorem ipsum…"}}.
Always pin the seed when committing fixtures, so two devs get the same data.
Pair with the validators. Generate a chat-shaped dataset, then run it through the matching OpenAI / Anthropic validator to confirm the shape before wiring it into your training script.
500k cap. Above that we run out of browser memory. For huge synthetic sets, generate in chunks or use a CLI tool.

Frequently asked questions

Is the data real?

No — purely synthetic. No external API calls, no real PII. Safe to use in public demos.

Can I generate fine-tune chat samples?

Indirectly. Define a flat schema for prompts and answers, then wrap with messages arrays in a tiny post-processing step. A dedicated visual editor exists for hand-authoring chat rows.

Why does the same seed not match other tools?

The seed only matches inside this generator. We use a small mulberry32 PRNG; other tools use different algorithms.