ai-output-extract-json Edge Function — Ai

ai-output-extract-json

Extracts the first valid JSON object or array embedded in LLM free-text output. Handles markdown code fences (\``json ... ```) and falls back to scanning for raw { }or[ ]` blocks.

Install

aerostack function add ai-output-extract-json

Usage

import { extractJson } from './functions/ai-output-extract-json';

const llmOutput = `Here is the data:
\`\`\`json
{"name": "Alice", "age": 30}
\`\`\`
Hope that helps!`;

const result = extractJson(llmOutput);
// { success: true, data: { json: { name: "Alice", age: 30 }, raw: '{"name": "Alice", "age": 30}', found: true } }

// Strict mode: only accept objects and arrays
const strictResult = extractJson('```json\n42\n```', true);
// { success: true, data: { json: null, raw: '', found: false } }

API

POST /api/ai-output-extract-json

Request

{
  "text": "The result is: {\"status\": \"ok\", \"count\": 42}",
  "strict": false
}

Response

{
  "success": true,
  "data": {
    "json": { "status": "ok", "count": 42 },
    "raw": "{\"status\": \"ok\", \"count\": 42}",
    "found": true
  }
}

Extraction Priority

1. Markdown code fences -- \``json ... ```or``` ... ```` are tried first
2. Raw JSON blocks -- scans for { ... } or [ ... ] with matching brace/bracket depth
3. Not found -- returns { json: null, raw: "", found: false }

Notes

• Handles nested objects, escaped quotes, and braces inside string values.
• In strict mode, only objects and arrays are accepted; primitives like 42, "hello", or true are rejected.
• Always returns { success: true, data: ... } -- never throws. Parse failures simply result in found: false.