running Open AI OSS N8N and Ollama locally

---

---

https://www.youtube.com/watch?v=mnV-lXxaFhk The video’s creator, Lucas, explains that OpenAI released its first open-source model, gpt-oss, since GPT-2. This model runs locally, costs nothing, and performs comparably to GPT-4. Lucas plans to demonstrate how to set it up on a local machine, integrate it with n8n, and use it in automated AI agents without API keys or cloud dependencies. The Gameplan:

1. Setup n8n locally: Lucas recommends using Docker for easy setup and teardown. He directs viewers to the n8n Docker documentation page. Prerequisite: Docker Desktop must be installed. Lucas shows how to download it for different operating systems and verify its installation via the terminal. He then executes two Docker commands in the terminal: docker volume create n8n_data: To create a persistent data volume for n8n workflows, executions, and debug data. He verifies this by listing Docker volumes. docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8n: To spin up and run the n8n Docker container, binding the data volume and exposing it on port 5678. After running the command, n8n starts, and Lucas navigates to http://localhost:5678 in his browser to access the n8n editor. He completes the initial owner account setup and skips the onboarding session.

2. Install Ollama: Lucas goes to ollama.com and downloads the appropriate installer for his macOS system. He demonstrates dragging Ollama to the Applications folder (though he already has it installed).

3. Install **gpt-oss** (using Ollama): Lucas visits ollama.com/library/gpt-oss to find the gpt-oss model. He copies the command ollama run gpt-oss:latest and executes it in his terminal. This downloads the model weights for gpt-oss:latest. Once the download is complete, he tests it by asking gpt-oss to write a poem about n8n. The model responds quickly, demonstrating its local operation.

4. Hook up Ollama Chat Model inside n8n and start prompting: Lucas returns to n8n and first stops the current Ollama server in his terminal, as n8n needs to connect to the Ollama server itself, not the interactive chat. He runs ollama serve in a new terminal window to start the Ollama server. He creates a new workflow in n8n, adding a “Manual Trigger” and an “Ollama Chat Model” node. To connect n8n to the Ollama server, he needs to change the Base URL in the “Ollama Chat Model” credential settings. He finds the correct URL (http://host.docker.internal:11434) from the n8n-io/self-hosted-ai-starter-kit GitHub repository. After updating the Base URL and saving, the connection tests successfully. He then connects a “Basic LLM Chain” node to the “Ollama Chat Model” and adds a “User Message” prompt: “write a Linkedin post comparing n8n vs make.com”. Upon executing the workflow, he encounters an error because the model llama3.2 was not found. He corrects this by selecting gpt-oss:latest from the “Model” dropdown in the “Ollama Chat Model” node. He re-executes the workflow. It takes slightly longer but successfully generates a LinkedIn post comparing n8n and make.com in markdown format. He then extends the workflow by adding a “Chat Trigger” node, an “AI Agent” node, a “Simple Memory” node, a “Think Tool” node, and another “Edit Fields” node. He configures the “AI Agent” to use the “Ollama Chat Model” and “Simple Memory.” He adds the “Think Tool.” He tests this more complex agent by providing a prompt: “Think deeply about the differences between n8n and make.com and write me a LinkedIn post breaking down the pros and cons of each.” The agent successfully executes, utilizes the “Think” tool (as seen in the logs), and generates a comprehensive LinkedIn post.

Lucas concludes by highlighting the incredible potential of running powerful AI models locally for free, opening up new opportunities for AI developers in industries with strict privacy concerns like defense, healthcare, and legal. He encourages viewers to like, subscribe, and join their free School community to access this automation template and other AI resources.