Build a Local AI Stack on Your Mac Using n8n and Ollama
A step-by-step guide to running AI agents and local language models directly on your Mac without sending data to the cloud.
Stock photo for illustration only, not from the actual event
- Deploy the official Self-hosted AI Starter Kit via Docker Compose for a fast setup.
- Run Ollama natively on macOS to bypass Apple Silicon GPU limitations inside Docker containers.
- Connect n8n to Ollama using host.docker.internal to bridge container and host networks.
If you want AI agents running entirely on your own machine with complete data privacy, combining n8n and Ollama is an ideal solution. Serving as part three of the series, this guide moves past initial database setups and LLM fundamentals into practical workflow automation using local models.
Your primary prerequisite is Docker. If it is not installed yet, set up Docker Desktop for Mac following the official documentation. The fastest path is utilizing n8n's official Self-hosted AI Starter Kit, a Docker Compose template bundling n8n, Ollama, Qdrant vector store, and PostgreSQL preconfigured for immediate integration.
Stock photo for illustration only, not from the actual event
You can initialize the setup by cloning the repository and copying the environment template file. Update your credentials inside the .env file with specific parameters such as:
- POSTGRES_USER = admin
- POSTGRES_PASSWORD = root
- POSTGRES_DB = n8n
Make sure to also replace the N8N_ENCRYPTION_KEY and N8N_USER_MANAGEMENT_JWT_SECRET values with your own unique random strings.
Running a fully self-hosted AI infrastructure gives developers complete ownership over sensitive data and intellectual property. By orchestrating tasks with n8n and serving models locally via Ollama, teams can eliminate recurring cloud API expenses while maintaining absolute operational security on local hardware.
A crucial detail for Mac users is that Docker on Apple Silicon cannot directly access the host machine's GPU. Consequently, the starter kit documentation recommends running Ollama natively on macOS for speed and letting containers connect remotely. Configure this in your environment file:
- OLLAMA_HOST = host.docker.internal:11434
Launch the entire stack using the docker compose command, then open http://localhost:5678 to create your initial n8n account where your workflows and autonomous agents will reside.
For the Ollama side of the stack, you can install it via Homebrew, pull your preferred language model, interact with it directly in the terminal, and list your downloaded models effortlessly.
Inside n8n, Ollama appears as a dedicated model node ready for integration. Simply create an Ollama credential with the base URL pointing to host.docker.internal, build a workflow featuring a Chat Trigger and an AI Agent node, and select your local model. You can further empower your agent by attaching sub-tools like HTTP requests or Qdrant vector databases.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment