🧱 langschema

one-line LLM output parsers for JS/TS. no code bloat. one file.

how to use

1. install the package

npm i langschema

2. use any of our one-line parsers

import { bool, list, categorize, asZodType } from 'langschema'

const isGood: boolean = await bool('Is this review positive? Review: Best bang for your buck.')

const foodsAte: string[] = await list(
  'What foods did this review user like? Review: i loved pizza and milkshakes', 
  ['pizza', 'burger', 'fries']
)

const rating: string = await categorize(
  `What rating would this review user give?
  Review: could NOT recommend it more, best ive ever eaten`, 
  ['1 star', '2 stars', '3 stars', '4 stars', '5 stars']
)

const userSchema = z.object({
  name: z.string(),
  age: z.number(),
})

const user = await asZodType('hi my name is John and im 21 years old', userSchema)
// => { name: "John", age: 21 }

features

boolean parser

useful for parsing outputs that are in a binary format, i.e., true or false.

categorize parser

it assists in parsing outputs that can be categorized into a specific set of strings (enums).

list parser

this is intended to parse outputs that are lists of specific set of strings (enums).

zod parser

parse ANY zod type with this parser.