Build an n8n Workflow with Local AI (Ollama)

Run an AI-powered workflow entirely on your machine. No data leaves your server.

Prerequisites

• Ollama installed (ollama run llama3.2)
• n8n running (Docker or npm)

Step 1: Create a workflow

In n8n, add a trigger (Webhook, Schedule, or Manual). Add an Ollama node. Configure: base URL http://host.docker.internal:11434 (or localhost:11434), model llama3.2.

Step 2: Design the prompt

In the Ollama node, set the prompt. Use expressions to inject data: {{ $json.input }} or {{ $node["Webhook"].json.body }}. Add system instructions if the node supports it.

Step 3: Add logic

Use an IF node to branch: e.g., if sentiment is positive, route to one action; if negative, route to another. Use a Switch node for multiple outcomes.

Step 4: Handle output

Add nodes to process the AI response: save to DB, send email, post to Slack, or trigger another workflow. Map $json.response or the relevant output field.

Step 5: Error handling

Add an Error Trigger. On failure, log to a sheet, send an alert, or retry. Prevents silent failures.

Example: Support ticket classifier

Webhook receives ticket → Ollama classifies (urgent/normal/low) → IF urgent → Slack alert. Else → add to sheet. All local, no API keys.