llm-json-extractor

Extract clean code blocks or structured JSON from conversational, messy, or truncated LLM responses.

Features

• 📦 Zero dependencies — Under 800 bytes minified.
• ⚡ Built for Edge — Run in Node.js, Bun, Cloudflare Workers, or Vercel Edge.
• 🛡️ Fully type-safe — Native TypeScript support with generic parsing.
• 🧩 Resilient parsing — Isolates blocks even when cut off or missing closing fences.

Installation

# npm
npm install llm-json-extractor

# pnpm
pnpm add llm-json-extractor

# yarn
yarn add llm-json-extractor

# bun
bun add llm-json-extractor

Usage

import { unfenceJson } from 'llm-json-extractor'

// Messy conversational LLM completion
const response = `
Here is the user profile you requested:

\`\`\`json
{
  "id": "usr_9982",
  "name": "Sarah Jenkins",
  "roles": ["admin"]
}
\`\`\`

Let me know if you need anything else!
`

interface User {
  id: string
  name: string
  roles: string[]
}

// Parse messy response into a typed object
const user = unfenceJson<User>(response)

console.log(user.name) // => "Sarah Jenkins"

API

| Function | Signature | Returns | Description | | :--- | :--- | :--- | :--- | | unfence | (text: string, language?: string) | string | Extracts contents of the first matching code block. | | unfenceJson | <T>(text: string) | T | Extracts and parses the first JSON block. | | UnfenceError | class extends Error | Error | Thrown when parsing fails; exposes raw and extracted text. |