Not sure why, but n8n-ai is stuck at the thinking stage. I have tried reloading the browser, and then it works. But for the second request, it gets stuck again. Any idea what might be the issue?
Also how can I contact n8n support for a query like this?
Hi @hefty , welcome!
Are you referring to the built-in AI assistant in the n8n editor (the AI helper that builds workflows), or an AI Agent node in one of your workflows?
Could you share your n8n version and whether you’re on Cloud or self-hosted? That’ll help narrow it down.
@JohnHalex I have been noticing the error from yesterday.
@houda_ben yes, I am talking about the built-in ai assisstant by n8n that helps with creating workflows. I am using 2.15.0 version. And I am using cloud version of n8n.
Adding to the great suggestions above — if this is the n8n AI assistant (the built-in AI that generates workflows), a common cause is a WebSocket connection timeout. The AI builder uses a persistent WebSocket connection, and if your network drops briefly or there’s a proxy/firewall in between, it can hang silently.
Fix to try:
If it’s an AI Agent node stuck in a loop: that’s usually the LLM going into infinite tool calls. Adding maxIterations in the node settings can help.
Hi @hefty Welcome!
For support you can read this:
Also for your question that it works but it does not makes me think that it is mainly a problem related to the browser you use, for now lets switch a browser from maybe Chrome to Zen, also just in case try using a different AI provider maybe is causing some problem. (Also try to access n8n from a different device that will confirm the issue)
welcome to the n8n community @hefty
I couldn’t find an exact GitHub issue for this, so I think it would be worth opening one.
Since you’re on n8n Cloud 2.15.0 and already tested the usual browser-side fixes, this looks specific enough to report, especially with the Maximum call stack size exceeded console error.
If possible, I’d also include a minimal reproduction: the shortest prompt sequence that works once and then consistently gets the assistant stuck. That usually makes it much easier for the team to verify and debug.
@hefty Hi
The loop is happening inside a file called useDataSchema.ts.
When you ask the AI a question, it tries to read the structure of the data flowing through your workflow, so it can give you an accurate answer.
If your workflow has a very deeply nested JSON object, a massive amount of data, or a circular reference in one of the node outputs, the schema reader gets trapped in a loop trying to map it all out. Once it crashes, the AI is left “thinking” forever because the background process died.
The 403 (Forbidden) Sentry errors at the bottom just mean n8n’s system is trying to send an automated bug report about the crash to their developers, but it’s getting blocked.
Before asking the AI a question, try clearing the execution data in your nodes, or test the AI on a blank workflow. If it works on a blank workflow, it confirms the AI is choking on the specific data structure of your current workflow.
