Local AI Lab 002: First n8n to Ollama Workflow

Question

Send content from n8n to a locally hosted Ollama model and receive an AI-generated summary.

Hypothesis

If n8n sends properly formatted JSON to Ollama, then a local model should be able to return a structured response that can be reused inside an automation workflow.

Setup

• n8n running on the home server
• Ollama serving qwen2.5:3b
• HTTP Request node configured to call the Ollama API
• Test content passed from n8n into the request body

Steps

Manual Trigger
    ↓
Set / Input Content
    ↓
HTTP Request
    ↓
Ollama
    ↓
qwen2.5:3b
    ↓
Structured Response

Result

n8n successfully passed content to Ollama running qwen2.5:3b on the home server.

The model returned a structured response containing:

• Summary
• Key takeaways
• Action items
• Follow-up questions

What Worked

• Separating instructions from the content body made the prompt easier to manage.
• Sending JSON directly to Ollama produced reusable output for workflow steps.
• The local model was sufficient for summarization and simple structure.

What Failed

• Invalid JSON caused by multiline content.
• Incorrect body format when using Form-Data instead of JSON.
• Boolean type mismatch for the stream parameter.

Lessons Learned

• APIs often require strict JSON formatting.
• Data types matter. A string like "false" is not the same as a boolean false.
• Building prompts from separate instruction and content fields creates more maintainable workflows.
• Streaming responses can be useful for real-time output, but they are harder to process in automation workflows.

Next Experiment

Configure the workflow to return a single JSON response and automatically save the generated output into a Markdown note.