Is anyone else having issues with AI Agents failing to parse their own JSON and unable to use GPTo1?

Questions
1

I’ve been running into two major problems with using n8n’s AI Agents.

First of all, more often than not they run into the following error message after 5-10 calls: “Unable to parse JSON response from chat agent.”

Second, whenever I try to use GPTo1 in an AI Agent set up, I get this error: “400 Unsupported value: ‘messages[0].role’ does not support ‘system’ with this model”. This is the case with whichever type of agent I use.

My workflow is pretty standard, I’m just using an agent with an OpenAI chat model, SerpAPI tool and a structured output.

My n8n version is 1.60.0 running on n8n cloud.

2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
3

Could just be down to the model? Here’s a post on the OpenAI forum with a similar complaint about o1 JSON responses.

Personally would recommend either waiting until o1 comes out of preview or a few months whilst all the bugs are ironned out.

4

It’s pretty simple to fix this. Just deactivate “Output Content as JSON”.

5

Simple solution: include your desired structure within the prompt.

I don’t use the output formatter. I started instructing LLMs to give me strict format output long before OpenAI built out the feature. The AI Agent node’s Structured Output Parser leaves a lot to be desired.

6

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.