n8n - 💡(How to fix) Fix Streaming intermediate steps from AI Agent are lost in webhook responses [1 comments, 2 participants]

Bug Description

Problem Description

When using an AI Agent node in a workflow with webhook streaming enabled, only the agent's final output is returned to the webhook client. The intermediate reasoning steps (tool calls, observations, agent thinking process) that the agent generates during execution are lost and not included in the streamed response.

What I'm Trying to Do

I want to build an LLM agent workflow that:

1. Receives a user request via webhook with streaming enabled
2. The AI Agent processes the request and calls multiple tools to gather information
3. During execution, I want the webhook client to receive:
   • ✅ The agent's reasoning process
   • ✅ Each tool call the agent makes (e.g., "Searching for...", "Calling API...")
   • ✅ Results from each tool (observations)
4. At last the final answer. But at the moment the step 3 doesn't happen.

Why This Matters

This is essential for:

• User experience: Users see progress as the agent thinks through the problem
• Transparency: Understanding what the agent is doing step-by-step
• Real-time feedback: Long-running operations don't appear frozen
• Server-Sent Events (SSE): Streaming responses pattern that's standard for LLM applications

To Reproduce

Steps to Reproduce

1. Create a new workflow
2. Add Webhook trigger with:
   • HTTP Method: POST
   • Response Mode: Streaming ← Key!
3. Add AI Agent node connected to webhook
   • Enable agent streaming
   • Enable return intermediate steps
4. Add RespondToWebhook node to return agent response
5. Deploy and make webhook request with a complex prompt (one that requires multiple tool calls)
6. Observe webhook response: Only final answer arrives, no intermediate steps in the response stream

Expected behavior

Current Behavior (Broken)

User Request via Webhook
    ↓
Webhook Node (streaming enabled)
    ↓
AI Agent Node executes and generates:
  - Intermediate Step 1: "Calling search tool..."  
  - Intermediate Step 2: "Search result: [data]"
  - Intermediate Step 3: "Calling database query..."
  - Intermediate Step 4: "Query result: [data]"
  - Final Output: "Based on the search and database query, here's your answer..."
    ↓
RespondToWebhook Node
    ↓
Webhook Response  (❌ "Calling search tool...") (❌ "Search result: [data]"S) (❌"Calling database query...") (❌ "Query result: [data]") (✅ FINAL)

Expected Behavior (What Should Happen)

User Request via Webhook
    ↓
Webhook Node (streaming enabled)
    ↓
AI Agent Node executes and streams intermediate steps:
  - Intermediate Step 1: "Calling search tool..."  
  - Intermediate Step 2: "Search result: [data]"
  - Intermediate Step 3: "Calling database query..."
  - Intermediate Step 4: "Query result: [data]"
  - Final Output: "Based on the search and database query, here's your answer..."
    ↓
RespondToWebhook Node (forwards all streamed data)
    ↓
Webhook Response (✅ "Calling search tool...") (✅ "Search result: [data]"S) (✅ "Calling database query...") (✅ "Query result: [data]") (✅ FINAL)

Debug Info

Following is a sample streamed response from the AI agent node.

{"type":"begin","metadata":{"nodeId":"b8","nodeName":"AI Agent","itemIndex":0,"runIndex":0,"timestamp":1775757598665}} {"type":"item", "content": "", "metadata":{"nodeId":"b8","nodeName":"AI Agent","itemIndex":0,"runIndex":0,"timestamp":1775757598665}} {"type":"item", "content": "First response chunk", "metadata":{"nodeId":"b8","nodeName":"AI Agent","itemIndex":0,"runIndex":0,"timestamp":1775757598665}} {"type":"item", "content": "Last response chunk", "metadata":{"nodeId":"b8","nodeName":"AI Agent","itemIndex":0,"runIndex":0,"timestamp":1775757598665}} {"type":"end","metadata":{"nodeId":"b8","nodeName":"AI Agent","itemIndex":0,"runIndex":0,"timestamp":1775757598665}}

Operating System

Ubuntu

n8n Version

2.14.2

Node.js Version

24

Database

SQLite (default)

Execution mode

main (default)

Hosting

self hosted