Stanley J

Posted on Jul 30 • Originally published at istealersn.substack.com

Autonomous Email Summarization

#ai #automation #tooling #javascript

Building an Autonomous Email Summarization Workflow with n8n, AI, and Obsidian

How I built a daily automation that fetches financial newsletters, summarizes them with AI, and saves clean markdown notes to Obsidian - completely hands-free.

The Problem

Like many investors, I receive daily market newsletters packed with valuable insights. But who has time to read through lengthy emails every morning? I wanted an automation that could:

Fetch emails automatically from specific senders
Extract and clean the content
Generate intelligent summaries using AI
Save formatted notes to my knowledge management system
Run completely autonomous - no manual intervention needed

The Solution Architecture

After exploring various approaches, I settled on this tech stack:

🕐 n8n: Workflow automation and orchestration
📧 Gmail API: Email fetching with smart filtering
🤖 Groq: Free AI summarization (Llama models)
📝 Obsidian: Knowledge management with Local REST API
⚙️ PM2: Process management for 24/7 operation

Step 1: Setting Up the Daily Trigger

The foundation is a schedule trigger that runs every day at 9:15 AM. This timing ensures I get summaries of yesterday’s emails right when I start my day.

// Cron expression for 9:15 AM IST daily
Schedule Trigger: 15 9 * * *
Timezone: Asia/Kolkata

Key insight: Always set the timezone explicitly to avoid confusion when testing vs production.

Step 2: Smart Email Filtering

Rather than processing all emails repeatedly, I implemented dynamic date filtering:

// Only fetch emails from last 24 hours
Received After: {{ new Date(Date.now() - 24 * 60 * 60 * 1000).toISOString() }}
Sender: specific-newsletter@domain.com

This ensures:

Only new emails are processed daily
No duplicate processing
Efficient API usage
Relevant, timely content

Step 3: Email Content Extraction

Gmail API returns emails in a complex nested structure. The key was handling both plain text and HTML emails while cleaning up the content:

// Extract email body from Gmail's complex structure
function extractBody(emailData) {
  let from = 'Unknown Sender';
  if (emailData.from && emailData.from.value && emailData.from.value[0]) {
    const sender = emailData.from.value[0];
    from = sender.name ? `${sender.name} <${sender.address}>` : sender.address;
  }

  // Extract subject and date from emailData
  const subject = emailData.subject || 'No Subject';
  const date = emailData.date || new Date().toISOString();

  // Prioritize plain text, fallback to cleaned HTML
  let emailBody = '';
  if (emailData.text && emailData.text.trim().length > 0) {
    emailBody = emailData.text;
  } else if (emailData.textAsHtml) {
    emailBody = emailData.textAsHtml
      .replace(/<p>/g, '\n')
      .replace(/<\/p>/g, '')
      .replace(/<br\s*\/?>/g, '\n')
      .replace(/&nbsp;/g, ' ')
      .replace(/<[^>]*>/g, '')
      .trim();
  }

  return { subject, from, date, emailBody };
}

Lesson learned: Always handle multiple email formats and include robust text cleaning for better AI processing.

Step 4: AI-Powered Summarization

For AI summarization, I chose Groq for its generous free tier (14,400 requests/day) and fast inference. The key was crafting prompts specifically for financial content:

Individual Summary Prompt:
Analyze and summarize this financial newsletter content focusing on:

"{text}"

Provide a CONCISE SUMMARY that includes:
- Key stock recommendations and target prices
- Important market insights and trends  
- Sector analysis and opportunities
- Risk factors or warnings mentioned
- Actionable investment advice
- Any time-sensitive information

Format the summary in clear bullet points for easy reading.

Pro tip: Customize your AI prompts for your specific domain. Generic summarization prompts don’t capture domain-specific insights effectively.

Step 5: Obsidian Integration Challenge

This is where things got interesting. I initially tried Obsidian’s Local REST API:

// HTTP Request to Obsidian Local REST API
Method: PUT
URL: https://127.0.0.1:27124/vault/VaultName/filename.md
Headers: Authorization: Bearer <API_KEY>
Body: Markdown content

The Problem: Obsidian API returns HTTP 204 “No Content” on successful file creation, which n8n’s HTTP Request node can’t parse properly, causing server log errors.

The n8n HTTP 204 Bug

After extensive debugging, I discovered this is a known limitation in n8n:

File gets created successfully in Obsidian
Workflow completes and functions correctly
Server logs show parsing errors due to empty response body

The research revealed:

Multiple GitHub issues (#11283)
Community forum requests for better 204 handling
This affects many APIs that return 204 responses

Current workarounds:

Enable “Never Error” option
Enable “Include Response Headers and Status”
Accept the log noise (functionality works fine)

Step 6: Production Deployment with PM2

Running automation workflows requires 24/7 operation. PM2 was the perfect solution:

# Install PM2
npm install -g pm2

# Start n8n as a service
pm2 start n8n --name "n8n-automation"

# Set up auto-start on boot
pm2 startup
pm2 save

Benefits of PM2:

Auto-restart on crashes
Survives system reboots
Runs independently of user sessions
Built-in logging and monitoring
No need for VSCode or browser to stay open

The Final Workflow

Here’s what runs every morning at 9:15 AM:

Schedule Trigger → Wakes up the workflow
Gmail Node → Fetches new emails from specific sender
Code Node → Extracts and cleans email content
Summarization Chain → AI-powered content analysis
HTTP Request → Creates formatted markdown in Obsidian

Output: Clean, formatted daily notes like:

---
date: 2025-07-29T14:30:00.000Z
source: Financial Newsletter
type: Email Summary
---

# Daily Email Summary - July 29, 2025

## Original Email Details
- **From**: Newsletter Name <sender@domain.com>
- **Subject**: Daily Market Analysis
- **Date**: 2025-07-29T09:15:00.000Z

## AI Summary

**Top Stock Recommendations:**
- Company A (BUY) - Strong Q1 results
- Company B (HOLD) - Mixed earnings outlook

**Market Insights:**
- Tech sector showing resilience
- Energy stocks under pressure due to...

---
*Generated automatically at 14:30 IST*

Key Learnings

What Worked Well:

Modular approach: Each step handles one responsibility
Smart filtering: Date-based filtering eliminates duplicate processing
Free AI tier: Groq’s generous limits make this cost-effective
Local integration: Direct file creation in knowledge management system

Challenges Overcome:

Gmail API complexity: Nested email structures require careful parsing
n8n HTTP limitations: 204 response handling is currently buggy
Production deployment: Moving from development to 24/7 operation
Content cleaning: Raw email HTML needs significant processing

Future Improvements:

Multiple email sources: Expand to other newsletter subscriptions
Content categorization: Use AI to tag and organize summaries
Trend analysis: Weekly/monthly rollups of key themes
Alert system: Notify on high-priority market events

Cost Breakdown

Monthly Operating Costs: ~$0

n8n: Free (self-hosted)
Gmail API: Free tier sufficient
Groq AI: Free tier (14,400 requests/day)
Obsidian: Local storage
Mac Mini: Already owned

Time Investment:

Initial setup: ~4 hours
Maintenance: ~5 minutes/month
Daily time saved: ~15 minutes

ROI: After 2 weeks, the automation pays for itself in time savings.

Technical Considerations

Security:

API keys stored in n8n’s encrypted credential system
Local-only processing (no cloud data exposure)
HTTPS with certificate validation for all external calls

Reliability:

PM2 ensures 99%+ uptime
Error handling for network issues
Graceful degradation when services are unavailable

Scalability:

Easy to add new email sources
AI processing handles variable content lengths
File system can grow indefinitely

Conclusion

Building this automation transformed my morning routine. Instead of manually reading through lengthy financial newsletters, I now get concise, AI-generated summaries delivered directly to my knowledge management system.

The technical journey revealed interesting challenges around email parsing, AI prompt engineering, and n8n’s HTTP handling limitations. But the end result is a robust, autonomous system that saves significant time daily.

Total setup time: 4 hours

Daily time saved: 15 minutes

Monthly maintenance: 5 minutes

Technical satisfaction: Immeasurable 😊

Next Steps

If you’re inspired to build something similar:

Start simple: Begin with one email source and basic summarization
Test thoroughly: Run manually several times before automating
Handle edge cases: Empty emails, API failures, network issues
Monitor initially: Watch logs for the first week to catch issues
Iterate: Add features gradually based on actual usage

The combination of n8n’s visual workflow builder, modern AI APIs, and local knowledge management creates powerful possibilities for personal automation.

Have you built similar automation workflows? What challenges did you face and how did you solve them? Share your experiences in the comments below!

About Me

Engineering Leader with a decade of experience building scalable frontend architectures for high-traffic applications. Specialized in headless CMS implementations, performance optimization, and developer experience tooling. Currently architecting content delivery systems that serve millions of users while maintaining exceptional developer productivity.

Passionate about sharing real-world engineering lessons learned from production deployments, code reviews, and the inevitable 2 AM debugging sessions that teach us the most.

Connect with me:

LinkedIn: linkedin.com/in/stanley-j
Twitter: @istealersn_dev
Substack: @iStealersn.SubStack

Follow for more insights on frontend architecture, headless CMS patterns, and battle-tested engineering practices that actually work in production.

Have an optimization story or trick that worked wonders? Drop it in the comments or tag me on your platform of choice. Let’s keep learning from each other.

DEV Community