AI Workflows with Ollama

AI Automation Without the API Bill

Every major automation platform is pushing AI features right now. Zapier has AI actions. Make has AI modules. They all charge extra for them, and they all route your data through third-party LLM APIs — typically OpenAI — at $0.01 to $0.06 per 1K tokens.

Run that across thousands of automated tasks per month and you're looking at real money on top of your already-expensive automation subscription.

Here's the alternative: Ollama running locally on your RamNode VPS, connected to n8n. You get LLM-powered workflows with zero API costs, zero token metering, and zero data leaving your server.

What You Need

Ollama runs language models in memory. The model size determines your RAM requirements:

For most workflows, an 8B parameter model on a 4 GB+ VPS hits the sweet spot between quality and performance.

Step 1: Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

Pull a model — we'll start with Llama 3.1 8B:

ollama pull llama3.1:8b

Verify it's working:

ollama run llama3.1:8b "Respond with just 'OK' if you're working."

By default, Ollama listens on localhost:11434. Since n8n runs in Docker, we need Ollama accessible from inside the Docker network.

Step 2: Configure Ollama for Docker Access

Edit the Ollama service to bind to all interfaces:

sudo systemctl edit ollama

Add the following override:

[Service]
Environment="OLLAMA_HOST=0.0.0.0:11434"

Restart Ollama:

sudo systemctl restart ollama

Now update your docker-compose.yml to give n8n access to the host network:

  n8n:
    image: docker.n8n.io/n8nio/n8n
    restart: unless-stopped
    extra_hosts:
      - "host.docker.internal:host-gateway"
    environment:
      - N8N_HOST=n8n.yourdomain.com
      # ... rest of your environment variables

cd ~/n8n-docker
docker compose up -d

n8n can now reach Ollama at http://host.docker.internal:11434.

Step 3: Configure Ollama Credentials in n8n

1. Go to Credentials in the left sidebar
2. Click "Add Credential"
3. Search for "Ollama"
4. Set the Base URL to http://host.docker.internal:11434
5. Save the credential

Workflow 1: Intelligent Email Classifier

Receives emails via IMAP, classifies them using your local LLM, and routes them to the appropriate handler.

Trigger: Email Trigger (IMAP)

1. Add an Email Trigger (IMAP) node
2. Configure your mailbox credentials
3. Set it to check every 2 minutes

Classify with Ollama

Add an Ollama node with this prompt:

Classify the following email into exactly one category:
- SUPPORT: Technical issues, bug reports, help requests
- SALES: Pricing inquiries, demo requests, purchase interest
- BILLING: Invoice questions, payment issues, refund requests
- SPAM: Marketing, unsolicited offers, irrelevant content
- OTHER: Anything that doesn't fit the above categories

Respond with ONLY the category name, nothing else.

Subject: {{ $json.subject }}
Body: {{ $json.text.substring(0, 1000) }}

Route with Switch

• SUPPORT → Create ticket in your support system via HTTP Request
• SALES → Forward to sales team, log in CRM
• BILLING → Forward to billing, flag in accounting system
• SPAM → Move to spam folder, no notification

This entire pipeline runs on your VPS at zero marginal cost. On Zapier with AI actions, you'd pay per classification and per downstream action.

Workflow 2: Document Summarizer

Watches for new documents, summarizes them with your local LLM, and posts summaries to Slack.

Summarize with Ollama

Summarize the following document in 3-5 bullet points.
Focus on key decisions, action items, and important facts.
Keep each bullet point to one sentence.

Document:
{{ $json.content.substring(0, 4000) }}

Post to Slack

📄 *New Document Summary*
Source: {{ $('Webhook').first().json.filename }}
Processed: {{ $now.format('yyyy-MM-dd HH:mm') }}

{{ $json.text }}

Workflow 3: Support Ticket Auto-Responder

Instead of paying per-ticket for AI triage, your VPS handles it all.

Step 1 — Analyze the Ticket

Analyze this support ticket and provide:
1. SEVERITY: low, medium, high, critical
2. CATEGORY: billing, technical, account, feature-request
3. SUGGESTED_RESPONSE: A brief, professional response addressing the customer's issue.

Ticket:
Subject: {{ $json.subject }}
Message: {{ $json.body.substring(0, 2000) }}

Step 2 — Parse the Response

const response = $input.first().json.text;
const severity = response.match(/SEVERITY:\s*(low|medium|high|critical)/i)?.[1] || 'medium';
const category = response.match(/CATEGORY:\s*(\w[\w-]*)/i)?.[1] || 'general';
const suggestedResponse = response.match(/SUGGESTED_RESPONSE:\s*([\s\S]+)/i)?.[1]?.trim() || '';

return [{
  json: {
    severity,
    category,
    suggestedResponse,
    originalSubject: $input.first().json.subject,
    originalBody: $input.first().json.body
  }
}];

Step 3 — Route by Severity

• Critical: Immediately page on-call engineer + auto-respond
• High: Post to team Slack channel + auto-respond
• Medium/Low: Auto-respond with suggested response, queue for human review

Workflow 4: Log Analyzer

Feed server logs through your local LLM to detect patterns and anomalies. Triggered every hour on a schedule.

Collect Logs

journalctl --since "1 hour ago" --no-pager -q | tail -100

Analyze with Ollama

Analyze these server logs from the last hour. Report:
1. Any errors or warnings (with counts)
2. Unusual patterns or anomalies
3. Suggested actions if any issues are detected

If everything looks normal, respond with "STATUS: NORMAL" and nothing else.

Logs:
{{ $json.stdout }}

Conditional Alert

• IF response does NOT contain "STATUS: NORMAL" → Send Slack alert with the analysis
• IF response contains "STATUS: NORMAL" → No action (save on notification noise)

Performance Tips for Ollama on a VPS

A Note on Data Privacy