Connection Rejected when using public Chat Trigger (AI Agent workflow)

Questions
1

Hello everyone,

I’m building a public chatbot using n8n (LangChain AI Agent + Chat Trigger), and I’m running into an issue when trying to make the chatbot publicly accessible.

:gear: My setup

  • Chat Trigger node (public enabled)

  • AI Agent node connected to:

    • OpenAI Chat Model (GPT-5-mini)

    • Pinecone Vector Store (RAG setup)

    • OpenAI Embeddings (1536 dimensions)

  • Chat Response node enabled

  • Response mode configured for response nodes

:cross_mark: Problem

When I send a message to the public chatbot URL, I get:

  • “Connection rejected”

  • Sometimes: 500 Internal Server Error

  • In executions, the workflow appears stuck at “waiting” or never completes the response node

:magnifying_glass_tilted_left: What I observed

  • The workflow does trigger correctly

  • The AI Agent starts execution

  • But the final response is never returned to the chat node

  • Execution sometimes stays “running” without finishing

:test_tube: What I already checked

  • Credentials are correctly set (OpenAI + Pinecone)

  • Vector dimension is correct (1536)

  • No local nodes are used (fully cloud setup)

  • Response mode is set to Using Response Nodes

  • AI Agent is connected properly to Chat Trigger and Chat Response node

:red_question_mark: Question

Has anyone experienced this “Connection rejected” issue with public chat triggers + AI Agent?

Is there a known issue with:

  • responseNodes mode

  • or Chat Response node not returning output properly

  • or Pinecone tool blocking execution?

Any guidance would be really appreciated :folded_hands:

4

welcome to the n8n community @Mariem_Jammeli
One thing I don’t see mentioned yet, is your workflow actually active , or are you testing it from the editor? if you copied the chat URL while the workflow was in test mode, or before activating it, you’re hitting a webhook that no longer exists — which is exactly what “connection rejected” looks like from the outside.

Activate the workflow first, then grab the Chat URL from the node settings again and test with that. It should be a different URL from what you used in test mode.

5

Hi @Mariem_Jammeli Welcome! switch your Chat Trigger response mode from “Using Response Nodes” to “When Last Node Finishes” and remove the Chat Response node entirely — that mode is fragile with AI Agent chains and is almost certainly why your execution hangs at “waiting” without returning output. also fix the model to gpt-4o-mini like Benjamin said. here’s a minimal working setup:

plug in your OpenAI creds, wire your Pinecone vector store as a tool on the agent, activate the workflow and grab the production chat URL.