Let’s get started with setting up an AI workflow using n8n in Docker on a Windows machine, featuring local service integration with Ollama. Whether you are familiar with n8n or starting from scratch, this guide walks you through the steps needed.
Understanding the Basics
n8n is a versatile, self-hosted automation tool designed to connect and automate the use of over 400 services, now including AI components. When integrated with various large language models (LLMs) such as OpenAI’s chat models, Google’s Gemini Chat Model, or Ollama, it extends its capabilities significantly.
Prerequisites: Ollama and Other Services
If Ollama is already running on your local machine, you’re off to a great start. However, for a comprehensive AI workflow, integrating additional services like a vector database (Qdrant) and a relational database (PostgreSQL) is crucial. These databases help manage large datasets effectively, enabling advanced features such as semantic search and AI recommendations.
If You’re Starting from Scratch
Before diving deep, check out this Git repository, not affiliated directly, but it includes:
- n8n - A low-code platform packed with integrations and AI components.
- Ollama - A cross-platform LLM for running local models.
- Qdrant - A high-performance vector database.
- PostgreSQL - A robust relational database.
For Mac or Linux Users
The installation instructions in the mentioned repository are straightforward for macOS or Linux users. However, for Windows users wishing to run n8n with Docker but keep Ollama separate, here are some tailored instructions.
Setup on Windows: Step-by-Step
Install Ollama Locally
-
Download and Install Ollama: Visit Ollama’s website and download the installer. Choose models based on your GPU memory (e.g., 1b for 1GB).
-
Settings Configuration: Open Ollama settings post-installation and enable “Expose Ollama to the network” to allow API access by n8n.
Modify Docker Compose File
For running n8n, Qdrant, and PostgreSQL in a Docker container, you’ll make some modifications to the docker-compose.yml file in the self-hosted AI starter kit:
1volumes:
2 n8n_storage:
3 postgres_storage:
4 ollama_storage:
5 qdrant_storage:
6
7networks:
8 demo:
9
10x-n8n: &service-n8n
11 image: n8nio/n8n:latest
12 networks: ['demo']
13 environment:
14 - DB_TYPE=postgresdb
15 - DB_POSTGRESDB_HOST=postgres
16 - DB_POSTGRESDB_USER=${POSTGRES_USER}
17 - DB_POSTGRESDB_PASSWORD=${POSTGRES_PASSWORD}
18 - N8N_DIAGNOSTICS_ENABLED=false
19 - N8N_PERSONALIZATION_ENABLED=false
20 - N8N_ENCRYPTION_KEY
21 - N8N_USER_MANAGEMENT_JWT_SECRET
22 - OLLAMA_HOST=${OLLAMA_HOST:-ollama:11434}
23 env_file:
24 - path: stack.env
25 required: true
26
27x-ollama: &service-ollama
28 image: ollama/ollama:latest
29 container_name: ollama
30 networks: ['demo']
31 restart: unless-stopped
32 ports:
33 - 11434:11434
34 volumes:
35 - ollama_storage:/root/.ollama
36
37x-init-ollama: &init-ollama
38 image: ollama/ollama:latest
39 networks: ['demo']
40 container_name: ollama-pull-llama
41 volumes:
42 - ollama_storage:/root/.ollama
43 entrypoint: /bin/sh
44 environment:
45 - OLLAMA_HOST=ollama:11434
46 command:
47 - "-c"
48 - "sleep 3; ollama pull llama3.2"
49
50services:
51 postgres:
52 image: postgres:16-alpine
53 hostname: postgres
54 networks: ['demo']
55 restart: unless-stopped
56 environment:
57 - POSTGRES_USER
58 - POSTGRES_PASSWORD
59 - POSTGRES_DB
60 volumes:
61 - postgres_storage:/var/lib/postgresql/data
62 healthcheck:
63 test: ['CMD-SHELL', 'pg_isready -h localhost -U ${POSTGRES_USER} -d ${POSTGRES_DB}']
64 interval: 5s
65 timeout: 5s
66 retries: 10
67
68 n8n-import:
69 <<: *service-n8n
70 hostname: n8n-import
71 container_name: n8n-import
72 entrypoint: /bin/sh
73 command:
74 - "-c"
75 - "n8n import:credentials --separate --input=/demo-data/credentials && n8n import:workflow --separate --input=/demo-data/workflows"
76 volumes:
77 - ./n8n/demo-data:/demo-data
78 depends_on:
79 postgres:
80 condition: service_healthy
81
82 n8n:
83 <<: *service-n8n
84 hostname: n8n
85 container_name: n8n
86 restart: unless-stopped
87 ports:
88 - 5679:5678
89 volumes:
90 - n8n_storage:/home/node/.n8n
91 - ./n8n/demo-data:/demo-data
92 - ./shared:/data/sha
93 depends_on:
94 postgres:
95 condition: service_healthy
96 n8n-import:
97 condition: service_completed_successfully
98
99 qdrant:
100 image: qdrant/qdrant
101 hostname: qdrant
102 container_name: qdrant
103 networks: ['demo']
104 restart: unless-stopped
105 ports:
106 - 6333:6333
107 volumes:
108 - qdrant_storage:/qdrant/storage
109
110 ollama-cpu:
111 profiles: ["cpu"]
112 <<: *service-ollama
113
114 ollama-pull-llama-cpu:
115 profiles: ["cpu"]
116 <<: *init-ollama
117 depends_on:
118 - ollama-cpu
119 python:
120 image: python:3.11-slim
121 container_name: n8n-python
122 networks: ['demo']
123 restart: unless-stopped
124 working_dir: /scripts
125 volumes:
126 - ./n8n-scripts:/scripts
127 command: bash -c "pip install pillow flask && python /scripts/design_api.py"
128 ports:
129 - "5001:5000"
The Ollama service and corresponding GPU configurations have been removed from the Docker Compose setup
Running the Docker Compose
Execute the docker-compose file by running:
docker-compose up -d
Ensure the environment variable OLLAMA_HOST in line 22 is set to your local IP address; DO NOT use 127.0.0.1; Otherwise, you will need to manually update the n8n node each time the service restarts.
Simplify Management with Portainer
For easier service management, consider using Portainer:
-
Installation: Get the Community Edition for free or Business Edition for 3 free nodes from Portainer.io.
-
Create a Stack: Use the modified YAML configuration as a stack with your environment settings in Portainer for a hassle-free setup.
Building Your AI Workflow in n8n
Once everything is set:
- Open the n8n editor.
- Add a Basic LLM Chain node.
- Connect it to the Ollama model, enhanced by Qdrant’s performance, to create a robust, scalable AI workflow capable of semantic search, literature generation, or recommendations.
With these steps, you can efficiently harness the power of local LLM services and databases for a truly mature AI workflow right from the comfort of your Windows setup.
Enjoyed this article? Support my work with a coffee ☕ on Ko-fi.
