How yvain13/servicenow-n8n-rag-chatbot works
Project Type & Purpose
• This is an n8n-based orchestration system providing two ServiceNow automation solutions:
a) L1 Support Triage Agent (LLM-powered ticket routing)
b) Knowledge RAG Chatbot (ingest Knowledge Articles into Qdrant + retrieval-augmented chat)
Main Components
• Frontend Chat Widget (ServiceNow Virtual Agent XML)
• n8n Orchestrator instance hosting two workflows
• ServiceNow REST API (Incident table, Knowledge Article table, Service Catalog API)
• AI Services:
• OpenAI Chat (LLM for classification & responses)
• OpenAI Embeddings (vector generation)
• SerpAPI (web search)
• Memory Store (Simple Memory node in n8n)
• Qdrant Vector Database
Workflows & Subcomponents
A. L1 Support Triage Agent Workflow
1) Chat Trigger Node (webhook)
2) Text Classifier Node (OpenAI chat model) → three branches:
• INCIDENT → ServiceNow Incident Creation Node
• REQUEST → HTTP Request Node (Service Catalog endpoint)
• GENERAL → AI Agent sub-flow:
– OpenAI Chat Response
– SerpAPI Search
– Simple Memory for context
– Summarization Chain to create ticket description
3) Return response to Chat UI
B. Knowledge RAG Chatbot Workflow
Pink Panel – Data Ingestion (manual trigger):
1) Execute Workflow Trigger
2) Get Many Table Records (ServiceNow Knowledge Articles)
3) Default Data Loader
4) Recursive Character Text Splitter
5) OpenAI Embeddings → high-dim vectors
6) Qdrant Vector Store → persist embeddings & metadata
Green Panel – RAG Chat Interaction:
1) Chat Trigger Node (webhook)
2) Query Embedding Node (OpenAI Embeddings)
3) Qdrant Search Node → retrieve top N chunks
4) AI Agent Node (OpenAI Chat Model + Simple Memory)
5) Return response to Chat UI
Relationships & Data Flows
• Frontend UI → n8n Chat Trigger → appropriate workflow
• In L1 Agent: classification branch → external API call → back to UI
• In Knowledge Chat: query embedding → Qdrant search → LLM response → UI
• Ingestion flow separate from chat flow, both within same n8n instance
Architectural Patterns & Principles
• Event-driven orchestration (n8n triggers)
• Pipeline pattern for data ingestion
• Decision router pattern (LLM-powered branching)
• Retrieval-Augmented Generation (RAG) for knowledge chat
• Separation of concerns: triage vs. knowledge retrieval