Realistic data and types

The panel closes the loop around your data model in both directions: paste a model and get realistic data, or take a live response and get its types.

Model → data

Paste a TypeScript interface — a dirty real-world one is fine: extends, Pick<...>, imports from libraries that are not present all resolve through the real TypeScript compiler, not a hand-rolled parser. laqi infers a shape from it and generates seeded values that look real, not "string" and 0:

interface User {
  id: number
  email: string
  createdAt: string
  priceCents: number
}

{ "id": 1, "email": "dell.roberts@example.net", "createdAt": "2026-08-24T09:12:00.000Z", "priceCents": 741.32 }

The same seed always produces the same output — pass seed to get byte-identical data across runs, useful for snapshot tests.

How field names pick a generator

Every string and number/integer field is matched against a table of naming rules, in order — the first one whose pattern matches the field name wins. This is what makes email fields look like emails and createdAt look like a date, instead of both falling back to a random word.

String fields:

Field name pattern	Produces
ends in _at/At, or contains date/time	a recent ISO date
contains email	an email address
exactly username	a username
exactly filename	a filename with a .txt extension
name, firstName, lastName, fullName, displayName	a person’s full name
contains phone	a phone number
contains avatar, image, photo	an image URL
contains url, link	a URL
contains city	a city name
contains street, address	a street address
contains country	a country name
contains zip, postal	a zip code
contains uuid, guid	a UUID
contains description, bio, summary	a sentence
contains title	three lorem-ipsum words
anything else	two lorem-ipsum words

Number and integer fields:

Field name pattern	Produces
exactly id or _id	a sequential integer — 1, 2, 3, … per field, per call
ends in _id or Id (e.g. userId, order_id)	a foreign key, 1-1000
contains price, total, amount, cost	a decimal price
contains age	an integer 18-80
contains count, quantity, qty	an integer 0-100
anything else	a random number in range, integer or decimal

Field-name matching is word-aware, not a raw substring test — candidate does not match date (the letters are buried mid-word), and userName matches the person-name rule before the generic fallback because name is a whole word inside it.

Every other JSON type generates directly from its shape: boolean, null, arrays (each item generated independently), objects (each field generated independently), and tuples (exact length, exact type per position — unlike the JSON Schema bridge used for type printing below, data generation never loses positional precision).

Data → types

Every endpoint’s response has a Copy types button in the panel, with a language picker. Types are derived from the live response body on demand, so they can never go stale against a body you changed by hand.

The same operation works from an existing endpoint or from a pasted model, and covers 25 languages:

C (cJSON), C++, C#, Crystal, Dart, Elixir, Elm, Flow, Go, Haskell, Java, JavaScript, JavaScript PropTypes, JSON Schema, Kotlin, Objective-C, PHP, Pike, Python, Ruby, Rust, Scala 3, Smithy, Swift, TypeScript — plus TypeScript with Zod or Effect Schema validators built in.

// TypeScript
interface User {
  id: number
  email: string
}

// Go
type User struct {
  ID    int64  `json:"id"`
  Email string `json:"email"`
}

Nothing about this is persisted beyond ordinary mock JSON — a pasted model is never saved, and generated data lands in your laqi/ files through the same write path the panel’s endpoint editor uses.

From an AI agent

Both directions are MCP tools — generate_data and get_types — so an agent can paste a model from a backend discussion, generate a realistic body, and hand back typed code in one pass. See Using laqi with AI agents.