Building AI Agent Workflows

Overview

AI Agent workflows require special attention to configuration and validation. This guide shows you how to build production-ready AI Agent workflows using n8n’s LangChain nodes.

AI Agent Architecture

A typical AI Agent workflow consists of:

Trigger

Entry point for the workflow:

Manual Trigger
Chat Trigger
Webhook
Form Trigger

Language Model

The AI model that powers the agent:

OpenAI Chat Model
Anthropic Chat Model
Google PaLM
HuggingFace Models

Agent Node

The orchestrator that uses tools and the language model:

AI Agent node
Conversation Chain
ReAct Agent

Tools

Capabilities the agent can use:

HTTP Request nodes (as tools)
Database operations
Custom functions
Built-in LangChain tools

Memory (Optional)

Conversation history storage:

Buffer Memory
Window Buffer Memory
Redis Memory

Output Parser (Optional)

Structure the agent’s response:

Structured Output Parser
JSON Parser
Custom parsers

Discovering AI Nodes

Search for AI Nodes

search_nodes({
  query: 'AI agent langchain',
  includeExamples: true
})

Key AI Nodes

AI Agent

@n8n/n8n-nodes-langchain.agentThe main agent node that orchestrates tools and language models.

OpenAI Chat Model

@n8n/n8n-nodes-langchain.lmChatOpenAiGPT-3.5 and GPT-4 chat models.

Memory

@n8n/n8n-nodes-langchain.memoryBufferWindowStores conversation history.

Tool

Any n8n node can be an AI tool when connected to an agent.

Building an AI Agent Workflow

Basic Agent
Agent with Tools
Agent with Memory

{
  "name": "Simple AI Agent",
  "nodes": [
    {
      "name": "Manual Trigger",
      "type": "n8n-nodes-base.manualTrigger",
      "position": [250, 300]
    },
    {
      "name": "OpenAI Chat Model",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "position": [450, 200],
      "parameters": {
        "model": "gpt-4",
        "temperature": 0.7
      }
    },
    {
      "name": "AI Agent",
      "type": "@n8n/n8n-nodes-langchain.agent",
      "position": [650, 300],
      "parameters": {
        "promptType": "define",
        "text": "You are a helpful assistant. Answer questions clearly and concisely."
      }
    }
  ],
  "connections": {
    "Manual Trigger": {
      "main": [[{ "node": "AI Agent", "type": "main", "index": 0 }]]
    },
    "OpenAI Chat Model": {
      "ai_languageModel": [[{ "node": "AI Agent", "type": "ai_languageModel", "index": 0 }]]
    }
  }
}

{
  "name": "AI Agent with HTTP Tool",
  "nodes": [
    {
      "name": "Chat Trigger",
      "type": "@n8n/n8n-nodes-langchain.chatTrigger",
      "position": [250, 300]
    },
    {
      "name": "OpenAI Chat Model",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "position": [450, 200],
      "parameters": {
        "model": "gpt-4",
        "temperature": 0.7
      }
    },
    {
      "name": "Fetch Weather Tool",
      "type": "n8n-nodes-base.httpRequest",
      "position": [450, 400],
      "parameters": {
        "url": "https://api.weather.com/v1/current",
        "method": "GET",
        "responseFormat": "json"
      }
    },
    {
      "name": "AI Agent",
      "type": "@n8n/n8n-nodes-langchain.agent",
      "position": [650, 300],
      "parameters": {
        "promptType": "define",
        "text": "You are a weather assistant. Use the weather tool to get current conditions."
      }
    }
  ],
  "connections": {
    "Chat Trigger": {
      "main": [[{ "node": "AI Agent", "type": "main", "index": 0 }]]
    },
    "OpenAI Chat Model": {
      "ai_languageModel": [[{ "node": "AI Agent", "type": "ai_languageModel", "index": 0 }]]
    },
    "Fetch Weather Tool": {
      "ai_tool": [[{ "node": "AI Agent", "type": "ai_tool", "index": 0 }]]
    }
  }
}

{
  "name": "AI Agent with Memory",
  "nodes": [
    {
      "name": "Chat Trigger",
      "type": "@n8n/n8n-nodes-langchain.chatTrigger",
      "position": [250, 300]
    },
    {
      "name": "OpenAI Chat Model",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "position": [450, 200],
      "parameters": {
        "model": "gpt-4"
      }
    },
    {
      "name": "Window Buffer Memory",
      "type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
      "position": [450, 400],
      "parameters": {
        "windowSize": 10
      }
    },
    {
      "name": "AI Agent",
      "type": "@n8n/n8n-nodes-langchain.agent",
      "position": [650, 300],
      "parameters": {
        "promptType": "define",
        "text": "You are a helpful assistant with memory of our conversation."
      }
    }
  ],
  "connections": {
    "Chat Trigger": {
      "main": [[{ "node": "AI Agent", "type": "main", "index": 0 }]]
    },
    "OpenAI Chat Model": {
      "ai_languageModel": [[{ "node": "AI Agent", "type": "ai_languageModel", "index": 0 }]]
    },
    "Window Buffer Memory": {
      "ai_memory": [[{ "node": "AI Agent", "type": "ai_memory", "index": 0 }]]
    }
  }
}

AI Workflow Validation

AI Agent workflows require comprehensive validation to prevent runtime failures.

Use validate_workflow for complete AI workflow validation:

validate_workflow(workflow)

What’s Checked

Language Model Detection

Ensures an AI Agent has a connected language model:

Error Example

{
  "type": "ai_configuration",
  "message": "AI Agent 'AI Agent' is missing a language model connection",
  "fix": "Connect a language model node (OpenAI, Anthropic, etc.) via ai_languageModel connection type"
}

Fix:

Add a language model node
Connect it with ai_languageModel connection type

Tool Connection Validation

Verifies tools are properly connected:

Error Example

{
  "type": "ai_configuration",
  "message": "Tool 'HTTP Request' is not connected to any AI Agent",
  "fix": "Connect this tool to an AI Agent node using ai_tool connection type"
}

Fix:

Connect tool nodes to agent with ai_tool connection type
Any n8n node can be a tool when connected this way

Streaming Mode Constraints

Checks for incompatible streaming settings:

Warning Example

{
  "type": "ai_configuration",
  "message": "AI Agent cannot use streaming with output parser",
  "suggestion": "Disable streaming or remove the output parser"
}

Streaming Limitations:

Cannot be used with output parsers
Not compatible with certain memory types
May not work with all language models

Memory and Output Parser Checks

Validates memory and parser configurations:

Example

{
  "type": "ai_configuration",
  "message": "Memory node 'Buffer Memory' not connected to AI Agent",
  "fix": "Connect via ai_memory connection type"
}

Requirements:

Memory nodes use ai_memory connection type
Output parsers use ai_outputParser connection type
Some combinations are incompatible (e.g., streaming + output parser)

Connection Types

AI workflows use special connection types:

Connection Type	Purpose	Example
`ai_languageModel`	Connect LLM to agent	OpenAI → Agent
`ai_tool`	Provide tool to agent	HTTP Request → Agent
`ai_memory`	Add conversation memory	Memory → Agent
`ai_outputParser`	Structure output	Parser → Agent
`main`	Regular data flow	Trigger → Agent

Example Connections Object

{
  "connections": {
    "Chat Trigger": {
      "main": [[{ "node": "AI Agent", "type": "main", "index": 0 }]]
    },
    "OpenAI Chat Model": {
      "ai_languageModel": [[{ "node": "AI Agent", "type": "ai_languageModel", "index": 0 }]]
    },
    "Weather Tool": {
      "ai_tool": [[{ "node": "AI Agent", "type": "ai_tool", "index": 0 }]]
    },
    "Memory": {
      "ai_memory": [[{ "node": "AI Agent", "type": "ai_memory", "index": 0 }]]
    }
  }
}

Node-Specific Validation

AI Agent Node

validate_node({
  nodeType: '@n8n/n8n-nodes-langchain.agent',
  config: {
    promptType: 'define',
    text: 'You are a helpful assistant'
  },
  mode: 'full',
  profile: 'runtime'
})

Checks:

Prompt configuration
Agent type compatibility
Required connections

Language Model Nodes

validate_node({
  nodeType: '@n8n/n8n-nodes-langchain.lmChatOpenAi',
  config: {
    model: 'gpt-4',
    temperature: 0.7,
    maxTokens: 2000
  },
  mode: 'full'
})

Checks:

Model availability
Parameter ranges
Credential configuration

Common Patterns

RAG (Retrieval Augmented Generation)

Chat Trigger → Vector Store → AI Agent → Response
                    ↓
              Document Loader

Multi-Tool Agent

Trigger → AI Agent → Response
            ↑
        Tool 1 (HTTP)
        Tool 2 (Database)
        Tool 3 (Custom)

Conversational Agent

Chat Trigger → AI Agent → Response
                 ↑
            Memory Node
            Language Model

Best Practices

Always Validate

Use validate_workflow to catch AI-specific issues

Test with Real Data

Test AI agents with representative inputs

Use Appropriate Models

Choose models based on complexity and cost

Implement Safeguards

Add validation and error handling

Monitor Token Usage

Track token consumption for cost control

Version Your Prompts

Document prompt changes for debugging

Troubleshooting AI Workflows

Missing Language Model

Error: AI Agent has no language model connectionSolution:

{
  "OpenAI Chat Model": {
    "ai_languageModel": [[
      {
        "node": "AI Agent",
        "type": "ai_languageModel",
        "index": 0
      }
    ]]
  }
}

Tool Not Connected

Error: Tool node not accessible to agentSolution:

{
  "HTTP Request Tool": {
    "ai_tool": [[
      {
        "node": "AI Agent",
        "type": "ai_tool",
        "index": 0
      }
    ]]
  }
}

Streaming Incompatibility

Warning: Streaming mode conflicts with output parserSolution:

Disable streaming: Set streaming: false in language model
OR remove output parser if streaming is required

Memory Not Persisting

Issue: Agent doesn’t remember conversationSolution:

Verify memory node is connected with ai_memory type
Check session ID is passed correctly for chat triggers
Ensure memory node configuration (e.g., windowSize) is appropriate

Testing AI Workflows

If n8n API is configured:

Test Execution

n8n_test_workflow({
  workflowId: 'workflow-id',
  // For chat triggers
  trigger: {
    message: 'Hello, can you help me?',
    sessionId: 'test-session-123'
  }
})

Monitor executions:

n8n_executions({
  action: 'list',
  workflowId: 'workflow-id',
  status: 'error'
})

Next Steps

Building Workflows

General workflow building process

Validation Strategies

Detailed validation modes and profiles

Using Templates

Find AI agent workflow templates

Troubleshooting

Common AI workflow issues

Get Started

Deployment

Configuration

IDE Integration

Core Concepts

Guides

​Overview

​AI Agent Architecture

​Discovering AI Nodes

​Key AI Nodes

AI Agent

OpenAI Chat Model

Memory

Tool

​Building an AI Agent Workflow

​AI Workflow Validation

​What’s Checked

​Connection Types

​Example Connections Object

​Node-Specific Validation

​AI Agent Node

​Language Model Nodes

​Common Patterns

​RAG (Retrieval Augmented Generation)

​Multi-Tool Agent

​Conversational Agent

​Best Practices

Always Validate

Test with Real Data

Use Appropriate Models

Implement Safeguards

Monitor Token Usage

Version Your Prompts

​Troubleshooting AI Workflows

​Testing AI Workflows

​Next Steps

Building Workflows

Validation Strategies

Using Templates

Troubleshooting

Build docs developers (and LLMs) love

Overview

AI Agent Architecture

Discovering AI Nodes

Key AI Nodes

Building an AI Agent Workflow

AI Workflow Validation

What’s Checked

Connection Types

Example Connections Object

Node-Specific Validation

AI Agent Node

Language Model Nodes

Common Patterns

RAG (Retrieval Augmented Generation)

Multi-Tool Agent

Conversational Agent

Best Practices

Troubleshooting AI Workflows

Testing AI Workflows

Next Steps