How to using n8n Automation guide

Build automated workflows that use your self-hosted vLLM endpoint. Examples: document summarization, chatbots, scheduled AI tasks, webhook-triggered processing. Prerequisite: A running vLLM endpoint (see the vLLM Docker Deployment Guide).

1. What is n8n?

n8n is a workflow automation tool (like Zapier/Make, but self-hosted). It connects apps, APIs, and AI models into automated pipelines via a visual editor.

Common LLM use cases:

• Summarize incoming emails/documents automatically

• Build a chatbot triggered by webhook

• Scheduled content generation

• Extract structured data from text

• Multi-step AI pipelines (classify → route → respond)

2. Deployment

Standalone docker-compose.yml

services:
  n8n:
    image: n8nio/n8n:latest
    container_name: n8n
    restart: unless-stopped
    ports:
      - "5678:5678"
    environment:
      - N8N_HOST=0.0.0.0
      - N8N_PORT=5678
      - N8N_PROTOCOL=http
      - NODE_ENV=production
      - GENERIC_TIMEZONE=Asia/Bangkok
      # Basic auth (recommended)
      - N8N_BASIC_AUTH_ACTIVE=true
      - N8N_BASIC_AUTH_USER=admin
      - N8N_BASIC_AUTH_PASSWORD=change-me-strong-password
    volumes:
      - ./n8n-data:/home/node/.n8n

volumes:
  n8n-data:

Add to Existing vLLM Stack

If you want n8n in the same compose file as vLLM (so they share a Docker network), add the n8n service block above into your existing docker-compose.yml. This lets n8n reach vLLM via the service name http://vllm:8000.

Launch

docker compose up -d
docker compose logs -f n8n

Access: http://<your-server-ip>:5678

On first launch, create your owner account.

3. Connecting n8n to vLLM

n8n has a built-in OpenAI node that works with any OpenAI-compatible API — including vLLM.

Create OpenAI Credentials in n8n

1. In n8n: Credentials → New → OpenAI API

2. Set:

   • API Key: your vLLM --api-key value (e.g. sk-llm-key)

   • Base URL: your vLLM endpoint

     • Same compose stack: http://vllm:8000/v1

     • Separate host: http://<server-ip>:8000/v1

3. Save

Network Note

n8n location	Base URL
Same compose file as vLLM	http://vllm:8000/v1
Different host/server	http://<vllm-server-ip>:8000/v1
Behind reverse proxy	https://api.yourdomain.com/v1

4. Example Workflows

Example A: Webhook Chatbot

Receives a message via webhook, sends to vLLM, returns the answer.

Nodes: Webhook → OpenAI → Respond to Webhook

1. Webhook node

   • Method: POST

   • Path: /chat

2. OpenAI node

   • Resource: Chat

   • Model: Qwen/Qwen2.5-7B-Instruct-AWQ

   • Messages: ={{ $json.body.message }}

3. Respond to Webhook node

   • Body: ={{ $json.choices[0].message.content }}

Test:

curl -X POST http://<server-ip>:5678/webhook/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "What is Kubernetes?"}'

Example B: Scheduled Document Summarizer

Runs daily, fetches documents, summarizes with the LLM.

Nodes: Schedule Trigger → HTTP Request (fetch docs) → OpenAI (summarize) → Send Email / Save

1. Schedule Trigger — daily at 09:00

2. HTTP Request — fetch your data source

3. OpenAI node

   • System prompt: "Summarize the following in 3 bullet points"

   • User: ={{ $json.content }}

4. Output — email, Slack, database, etc.

Example C: Text Classification & Routing

Classify incoming text, then route to different actions.

Nodes: Webhook → OpenAI (classify) → Switch → (multiple branches)

1. OpenAI node

   • Prompt: "Classify this message as: support / sales / spam. Reply with one word only."

2. Switch node — route based on classification

3. Branch actions per category

Tips for LLM Nodes

• Set temperature low (0.1-0.3) for consistent/structured output

• Use system prompts to control format

• For JSON output, instruct the model explicitly and parse with a Code node

• Set reasonable max tokens to control response length

5. Troubleshooting

n8n can't reach vLLM

• Same compose file → use http://vllm:8000/v1 (service name, not localhost)

• Different host → use the server IP and ensure port 8000 is reachable

• Test from n8n container:

  docker exec -it n8n wget -qO- http://vllm:8000/v1/models

"Unauthorized" / 401

• API key in n8n credentials must match vLLM --api-key

• If vLLM has no --api-key, you can put any placeholder in n8n

Model not found

• Use the exact model name from curl http://<vllm>:8000/v1/models

• Must match the --model value in your vLLM config

Workflow times out on long generations

• Increase n8n node timeout in workflow settings

• Reduce max tokens, or use a smaller/faster model

Data not persisting after restart

• Ensure the ./n8n-data:/home/node/.n8n volume is mounted

• Without it, workflows are lost on container recreation