Deploying n8n AI Starter Kit on Raspberry Pi

This guide simplifies the official n8n Self-Hosted AI Starter Kit for Raspberry Pi. Unlike the original, this setup:

• Works on Raspberry Pi with Ollama (CPU-only).
• Uses Portainer for easy stack management.
• Requires no CPU profile configuration.
• Strips away unnecessary complexity.

---

Step 1: Create Docker Volumes

Run the following commands to create persistent storage for your containers:

docker volume create n8n_storage && \
docker volume create postgres_storage && \
docker volume create qdrant_storage && \
docker volume create ollama_storage

If something goes wrong and you need a fresh start, delete the volumes:

docker volume rm n8n_storage postgres_storage qdrant_storage ollama_storage

Also create a shared folder:

mkdir -p ~/n8n-shared

🔹 Why n8n-shared? This folder allows you to pass files between n8n workflows and Ollama/Qdrant. For example, you can store prompt files, workflow outputs, or logs in one place accessible by both services.

---

Step 2: Prepare the .env File

Create a .env file with the following content:

POSTGRES_USER=root
POSTGRES_PASSWORD=strongpass
POSTGRES_DB=n8n

N8N_ENCRYPTION_KEY=<secret key>  # Use `openssl rand -hex 32`
N8N_USER_MANAGEMENT_JWT_SECRET=<secret key> # Use `openssl rand -hex 32`

SHARED_FOLDER=/home/<user>/n8n-shared

# Hostname n8n uses to generate webhook & callback URLs.
# Use "localhost" for local dev, replace with your real domain in production.
N8N_DOMAIN=localhost

📌 Notes:

• Use strong, randomly generated secrets for encryption and JWT.
• Adjust <user> in SHARED_FOLDER to your actual Linux username.

---

Step 3: Deploy the Stack in Portainer

1. Go to Portainer → Stacks → Add Stack.
2. Paste the following docker-compose.yaml into the editor.

networks:
  n8nnet:

services:
  # n8n (main app)
  n8n:
    image: n8nio/n8n:latest
    networks: ['n8nnet']
    hostname: n8n
    container_name: n8n
    restart: unless-stopped
    ports:
      - "5678:5678"
    environment:
      # Database connection (Postgres)
      DB_TYPE: postgresdb
      DB_POSTGRESDB_HOST: postgres
      DB_POSTGRESDB_USER: ${POSTGRES_USER}
      DB_POSTGRESDB_PASSWORD: ${POSTGRES_PASSWORD}
      DB_POSTGRESDB_DATABASE: ${POSTGRES_DB}

      # n8n hardening
      N8N_DIAGNOSTICS_ENABLED: "false"
      N8N_PERSONALIZATION_ENABLED: "false"
      N8N_ENCRYPTION_KEY: ${N8N_ENCRYPTION_KEY}
      N8N_USER_MANAGEMENT_JWT_SECRET: ${N8N_USER_MANAGEMENT_JWT_SECRET}

      # Ollama endpoint
      OLLAMA_HOST: http://ollama:11434

      # Store binary data on filesystem not in Postgres
      N8N_DEFAULT_BINARY_DATA_MODE: filesystem

      # Custom url
      N8N_HOST: ${N8N_DOMAIN}
      N8N_PROTOCOL: https
      WEBHOOK_URL: https://${N8N_DOMAIN}/
      N8N_EDITOR_BASE_URL: https://${N8N_DOMAIN}
      N8N_SECURE_COOKIE: "true"
    volumes:
      - n8n_storage:/home/node/.n8n
      - ${SHARED_FOLDER}:/data/shared
    depends_on:
      postgres:
        condition: service_healthy

  # PostgreSQL (Database for n8n)
  postgres:
    image: postgres:16-alpine
    networks: ['n8nnet']
    hostname: postgres
    container_name: postgres
    restart: unless-stopped
    ports:
      - "5432:5432"
    environment:
      POSTGRES_USER: ${POSTGRES_USER}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
      POSTGRES_DB: ${POSTGRES_DB}
    volumes:
      - postgres_storage:/var/lib/postgresql/data
    healthcheck:
      test: ['CMD-SHELL', 'pg_isready -h localhost -U ${POSTGRES_USER} -d ${POSTGRES_DB}']
      interval: 5s
      timeout: 5s
      retries: 10

  # Qdrant (vector Database)
  qdrant:
    image: qdrant/qdrant:latest
    networks: ['n8nnet']
    hostname: qdrant
    container_name: qdrant
    restart: unless-stopped
    ports:
      - "6333:6333"
    volumes:
      - qdrant_storage:/qdrant/storage

  # Ollama (CPU only for Raspberry Pi)
  ollama:
    image: ollama/ollama:latest
    networks: ['n8nnet']
    hostname: ollama
    container_name: ollama
    restart: unless-stopped
    ports:
      - "11434:11434"
    volumes:
      - ollama_storage:/root/.ollama

# Tell Compose to use pre-existing Docker volumes with these
# exact names. Compose will NOT create/delete them.
volumes:
  n8n_storage:
    external: true
  postgres_storage:
    external: true
  ollama_storage:
    external: true
  qdrant_storage:
    external: true

1. Upload your .env file in Portainer.

When creating the stack in Portainer → Stacks → Add Stack:

• Scroll down to the Environment variables section.
• Toggle “Upload from file”.
• Click Browse and select your prepared .env file from your computer (the one you created in Step 2).
• Once uploaded, Portainer will automatically inject those variables into the stack.

1. Deploy the stack.

2. Check all containers are running correctly.

🚨 Troubleshooting Qdrant on Raspberry Pi

If the Qdrant container fails to start and the logs show:

<jemalloc>: Unsupported system page size
terminate called without an active exception

This means your Raspberry Pi is running with 16 KB memory pages (the Pi 5 default). Qdrant’s allocator (jemalloc) only supports 4 KB pages, so it aborts immediately.

Fix: switch your Pi to the 4 KB-page kernel:

sudo nano /boot/firmware/config.txt

Add near the top (outside any [section]):

kernel=kernel8.img

Save, reboot, and verify:

getconf PAGESIZE   # should show 4096

Now Qdrant will run correctly.

---

Step 4: Open Required Ports

Allow n8n’s UI port (5678) on your server firewall:

iptables -A INPUT -p tcp --dport 5678 -j ACCEPT

⚠️ Important: After modifying iptables, you may need to restart Docker. This is because Docker manages iptables rules for container networking, and changes outside may not take effect until Docker reloads its own network configuration.

---

Step 5: Access n8n

• Without a domain:

  http://192.168.1.50:5678
  

• With Nginx Proxy Manager (NPM):

  https://n8n.mydomain.com
  

👉 If using a domain, ensure you configure reverse proxy with SSL in Nginx Proxy Manager.

---

Step 6: Ollama Endpoint Fix

Inside n8n, set the Ollama endpoint to:

http://ollama:11434

❌ Do not use http://localhost:11434, as localhost resolves inside the n8n container, not on your Raspberry Pi host.

---

Step 7: Download LLM Models

Perfect — since you want to refine Step 7 for Raspberry Pi 8GB or 16GB RAM, I’ll lock it down to only small, efficient models that are realistic to run on the Pi. Here’s the polished section:

---

Step 7: Download LLM Models

Your Ollama container is running, but it has no models yet. Since we’re on Raspberry Pi (8GB or 16GB RAM), stick to lightweight models that balance speed and quality. Avoid large models — they’ll swap and crawl.

7.1 Enter the Ollama container

Open a shell in Portainer (Containers → ollama → Console → /bin/sh) or from your Pi:

docker exec -it ollama sh

7.2 Pull small chat models (choose 1–2 to start)

# DeepSeek
ollama pull deepseek-r1:1.5b # OR ollama pull deepseek-r1:1.5b-qwen-distill-q4_K_M

# Gemma
ollama pull gemma3:1b  # OR ollama pull gemma3:1b-it-q4_K_M
ollama pull gemma3:1b-it-q8_0

# Qwen
ollama pull qwen3:0.6b
ollama pull qwen3:1.7b # OR ollama pull qwen3:1.7b-q4_K_M

7.3 Pull embedding models (for Qdrant)

For vector search in Qdrant, install at least one embedding model:

ollama pull nomic-embed-text:latest
ollama pull bge-m3:latest
ollama pull mxbai-embed-large:latest

7.4 Verify installed models

ollama list

7.5 Quick test (optional)

ollama run deepseek-r1:1.5b "Write a 2-sentence summary of what n8n does."

7.6 Manage space (optional)

If you need to remove unused models later:

ollama rm qwen3:1.7b

✅ You now have a lightweight, Raspberry Pi-optimized AI automation stack running n8n, Qdrant, Postgres, and Ollama — all managed via Portainer.