Pull the JSON out of an LLM reply
You ask for JSON and the model answers with a sentence, a ```json block and sometimes stray braces. Extract the first valid object or return null.
Write extractJson(reply) that returns the first valid JSON object in reply, or null when there is none.
If there is a code block fenced with three backticks (with or without the json tag) and its content is a valid object, use it. Otherwise look at each { and its closing }, taking nested braces and braces inside strings such as "a}" into account, and try JSON.parse. If it fails, move on to the next {.
Only objects count: an array or a number is not a valid answer.
One more check uses LIVE: real output from your own Ollama if you switch on "My Ollama", or a different sample if not. That check looks at structure, not exact words.
Challenges 0/5
- Reads the ```json block in REPLY
- Finds the object inside plain prose
- Respects braces inside strings
- Returns null when there is no valid object
- Reads the real reply from your model (My Ollama) or another sample
function extractJson(reply) {
// 1. try a fenced code block first
// 2. otherwise scan every '{', find its matching '}' (strings may contain braces)
// 3. JSON.parse the slice; return the first plain object, or null
try {
return JSON.parse(reply);
} catch (e) {
return null;
}
}
console.log(extractJson(REPLY));Go deeper: the Ollama reference →
This in production, with your data? Let's talk for 15 minutes →