I’m currently running frequently into the issue that the output of any AI-Agent- or Langchain-Node doesn’t match my requirements and lead to an error (sometimes within the Agent-Node, sometimes in one of the next nodes).
What is the error message (if any)?
{
“errorMessage”: “Model output doesn’t fit required format”,
“errorDescription”: “To continue the execution when this happens, change the ‘On Error’ parameter in the root node’s settings”,
“errorDetails”: {},
“n8nDetails”: {
“time”: “27.9.2025, 14:21:27”,
“n8nVersion”: “1.112.5 (Self Hosted)”,
“binaryDataMode”: “default”
}
}
AI models’ output isn’t always robust – or isn’t close to what you needed in the first place. AI Models usually respond with plain text rather than in a formatted or structured format. Although the response you want to receive from an agent can be customized in a system message, the knowledge base differs from one model to another. So with a smarter model you may achieve your goal – and have a robust response.
But I suggest formatting the response of the AI by attaching a code node and formatting the output after it leaves the agent node altogether. Because I don’t trust the AI having a similar output to what I need all the time.
Thanks for the explanation! That makes sense, formatting the output separately after the agent node. Could you maybe share a concrete example of how you ensure the output from an AI agent node is actually transformed into the desired format? I’m especially interested in how you practically implement this with a code node and what steps you take to reliably structure the result.
That would really help me understand it better. Thanks in advance!
Well there are many ways to structure the output of the agent afterwards. N8N has a built-in AI node specifically for this type of cases. '‘Information Extractor’’ lets you extract certain parts of a text message and form a JSON response and it’s AI based but very robust – fits your case (refer to image for reference).
However, you can still simplify the system message given to the agent and attach a code node afterwards with this code (example of a code piece – haven’t tested it yet);
def parse_agent_response(response):
lines = response.split('\n')
segments = []
current_segment = {}
for line in lines:
if line.startswith('Segment:'):
if current_segment:
segments.append(current_segment)
current_segment = {'segment': line[8:].strip()}
elif line.startswith('Begründung:'):
current_segment['justification'] = line[11:].strip()
if current_segment:
segments.append(current_segment)
return {"codedSegments": segments}
Here is the node by itself (change the descriptions to fit your case):
