Local AI Workflow Automation (n8n + Ollama)

A self-hosted automation platform with AI built in. n8n is a visual workflow builder with 500+ integrations and native AI agent nodes. Connect it to Ollama for local LLM inference, and you get a private alternative to Zapier + ChatGPT - no monthly fees, no data leaks.

What you get

• Visual workflow builder - drag-and-drop for complex automations
• AI agent nodes - built-in agent loop, code interpreter, tool calling
• 500+ integrations - Slack, email, databases, webhooks, CRMs, and more
• Local LLM inference - Ollama serves models on your GPU, zero API costs
• Cron + webhook triggers - run on schedule or in response to events
• Self-hosted - no data leaves your network, air-gap compatible
• $0/mo - compared to $20-100+/mo for Zapier + OpenAI

Architecture

Recommended GPU: RTX 3060 12GB for 14B models. For CPU-only workflows, a Beelink S9 mini PC is enough.

Prerequisites

• A GPU with ≥12 GB VRAM for local LLM inference (optional - n8n works CPU-only without local AI)
• Docker + Docker Compose
• ~2 GB free disk for n8n + model storage

Setup

Docker Compose (Ollama + n8n)

Save this as docker-compose.yml:

services:
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    volumes:
      - ollama:/root/.ollama
    ports:
      - "11434:11434"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    restart: unless-stopped
 
  n8n:
    image: docker.n8n.io/n8nio/n8n
    container_name: n8n
    depends_on:
      - ollama
    environment:
      - N8N_RUNNERS_ENABLED=true
      - GENERIC_TIMEZONE=America/New_York
      - TZ=America/New_York
      - N8N_ENFORCE_SETTINGS_FILE_PERMISSIONS=true
    volumes:
      - n8n_data:/home/node/.n8n
    ports:
      - "5678:5678"
    restart: unless-stopped
 
volumes:
  ollama:
  n8n_data:

Start it

docker compose up -d

Pull a model for AI-powered workflows:

docker exec ollama ollama pull qwen3:14b

Open http://localhost:5678 to access n8n.

Connect n8n to Ollama

In n8n, the Ollama node connects automatically when both containers are on the same Docker network. In the node config:

1. Add an Ollama Chat Model node
2. Set Base URL: http://ollama:11434
3. Select model: qwen3:14b
4. Use it with n8n's AI Agent node for agentic workflows

Use it

Email Summarizer

1. Trigger: Gmail / IMAP - new email received
2. AI step: Ollama Chat Model - "Summarize this email in 2 sentences"
3. Output: Slack - send summary to channel

Content Generator

1. Trigger: Cron - every Monday at 9 AM
2. AI step: Ollama Chat Model - "Write a weekly newsletter about [topic]"
3. Output: Email - send to subscribers

Data Enricher

1. Trigger: Webhook - receives CSV rows
2. AI step: Ollama Chat Model - "Extract company name, industry, and location from this text"
3. Output: PostgreSQL / Google Sheets - save enriched data

Smart Classifier

1. Trigger: Webhook - incoming support tickets
2. AI step: Ollama Chat Model - "Classify as billing/technical/feature-request and assign priority"
3. Output: n8n Split - route to different teams

Cost vs cloud

After 3-6 months the hardware has paid for itself versus a Zapier subscription - and you get unlimited AI calls, unlimited workflows, and total privacy.

Troubleshooting

• n8n can't reach Ollama → Both services share the same Docker network in the compose file above. Use http://ollama:11434 as the base URL.
• Slow AI responses → Check GPU usage. Smaller models like Mistral 7B are faster for simple tasks.
• No GPU in Ollama → Install the NVIDIA Container Toolkit and verify with nvidia-smi in a test container.
• n8n won't start → Check docker compose logs n8n. Port 5678 might be in use; change the left-side port mapping.

Swap components

• CPU-only setup → Drop the GPU section from the compose file, use Mistral 7B GGUF on CPU (slower but works).
• Use Qdrant for memory → Add Qdrant as a vector store for RAG-enabled workflows.
• Prefer another LLM → Swap the Ollama node for n8n's built-in OpenAI or Anthropic nodes if you want hybrid.
• SQLite vs PostgreSQL → n8n defaults to SQLite. For production, add a PostgreSQL container and set DB_TYPE=postgresdb.