I Used Codex to Clean Up 77,000 Emails

Khanh Nguyen — Wed, 22 Jul 2026 05:30:03 +0000

A few days ago, I needed to find an old email in a Gmail account I had barely touched for years.

Opening the inbox was overwhelming.

It contained 77,502 emails, including 72,356 unread messages. Important conversations were buried beneath newsletters, promotions, job alerts, receipts, and automated notifications.

Manually reviewing tens of thousands of emails was not realistic, and deleting everything was not an option.

Instead, I used Codex with a connected Gmail workflow to analyze the mailbox, design a safer organization system, and clean up the inbox in controlled stages.

Understanding the problem

Before changing anything, I asked Codex to analyze the mailbox and identify where most of the volume was coming from.

The analysis revealed that more than 92% of the mailbox consisted of updates and promotions.

That changed how I approached the cleanup.

The real problem was not simply the number of emails. It was that high-priority conversations and low-priority notifications were competing for the same space.

The inbox needed structure before it needed deletion.

The system behind it

The prompts I used

The workflow was driven by a small set of prompts:

How many emails are in my mailbox?

Analyze my mailbox and suggest how to organize it using labels, rules, and a cleanup schedule.

Start with labels and archive first.

The most important instruction was:

Start with labels and archive first.

That constraint kept the workflow controlled and reversible.

Archiving removed messages from the inbox without deleting them, while labels made it easier to review what had been classified before taking stronger action.

Rules and label structure

I used a few simple rules based on Codex’s recommendations and my own preferences to reduce the risk of removing something important:

Keep Finance and Security emails visible
Archive newsletters, job alerts, and routine notifications
Exclude starred and important messages
Preserve the original read and unread state
Keep cleanup actions reversible

Instead of organizing messages only by sender, I grouped them according to what I might need to do next:

Action/Finance
Action/Security
Career/Job Alerts
Reading/Tech
Cleanup/30 Days

This made the labels more useful.

A sender-based label tells me where an email came from. An action-based label tells me why the email still matters.

The results

After applying the workflow, the inbox became significantly smaller and easier to navigate.

The inbox was reduced from 77,502 to 28,745 emails, a 62.7% reduction.

During the cleanup:

48,296 emails were archived
461 emails were moved to Trash
Read and unread states were preserved
Finance and security emails remained visible
No permanent bulk deletion was performed

The goal was not to erase years of email history.

It was to remove low-value messages from the inbox while keeping useful information searchable and recoverable.

Final takeaway

The biggest improvement was not simply archiving 48,296 emails.

It was turning an abandoned inbox into a system where:

Important messages remain visible
Useful emails remain searchable
Routine notifications leave the inbox
Low-value messages become easier to review
Future cleanup becomes more manageable

Codex helped handle the scale, but the workflow succeeded because the prompts were clear, the actions were staged, and the important decisions remained reversible.

This article was created with AI assistance

I Moved My n8n Instance from Render to Azure

Khanh Nguyen — Wed, 10 Jun 2026 04:01:55 +0000

My n8n instance on Render worked perfectly.

It was fast, simple, fully managed, and cost-effectiveness. And I barely had to think about infrastructure.

So why move it?

Because I wanted:

More control over my environment
Centralized cloud billing in Azure
Hands-on experience with Infrastructure as Code (IaC), GitHub Actions, and cloud security
A real-world project instead of another tutorial

This post shares what I built and what I learned along the way.

Repository

git clone https://github.com/TQKNG/lab-n8n-azure.git

Check out README file for full guide

What I Wanted to Build

I wanted a fully reproducible deployment where I could:

Deploy everything from GitHub Actions
Authenticate to Azure without long-lived secrets (OIDC)
Run n8n securely behind HTTPS
Add identity + edge protection layers
Tear everything down when not needed

Think: personal SaaS-grade infrastructure, but lightweight

Architecture

This setup adds two important layers compared to a basic VM deployment:

Cloudflare as the edge layer
Azure AD (Entra ID) as the identity gate

What Gets Deployed

Compute

Ubuntu 22.04 LTS VM (Azure)
Trusted Launch enabled
SSH key-based access

Networking

Virtual Network (isolated subnet)
Network Security Group (restricted ports)
Public IP behind controlled ingress

Application

n8n (Docker container)
Caddy reverse proxy (HTTPS + routing)

Edge & Identity

Cloudflare (DNS + protection layer)
Azure AD (Entra ID authentication gate)

GitHub Actions Pipeline

To trace logs, I use below manually triggered workflows. These allows full lifecycle control from GitHub. Feel free to customize it based on your own workflow setup.

Setup → initialize repo variables
OIDC bootstrap → configure Azure trust
Verify → test authentication
Deploy → run Bicep infrastructure
Start → validate VM + n8n endpoint
Teardown → destroy resources safely

Key Security Layers

This architecture is intentionally layered:

Cloudflare → edge protection + TLS
Azure AD → identity verification before access
NSG → network-level filtering
Caddy → HTTPS termination + reverse proxy
Docker → container isolation

This creates a simple zero-trust-inspired flow.

What I Learned

Azure OIDC is a game changer. No more storing long-lived secrets in GitHub.
Bicep is actually pleasant. Much easier than ARM templates and still powerful.
Cloudflare simplifies edge security. It reduces direct exposure of Azure resources.
Infrastructure as Code changes everything. Even a small project becomes reproducible and disposable.

Next Step

Add Azure Key Vault for secrets
Add automated backups for n8n workflows
Move monitoring to Azure Monitor dashboards
Restrict SSH via IP allowlist or JIT access
Evaluate Azure Container Apps instead of VM
Add cost automation (auto shutdown dev environment)

Resources

Project

GitHub Repository: https://github.com/TQKNG/lab-n8n-azure
n8n Documentation: https://docs.n8n.io
Cloudfare Developer Docs: https://developers.cloudflare.com/

Azure

Bicep Docs: https://learn.microsoft.com/azure/azure-resource-manager/bicep
Azure OIDC with GitHub Actions: https://learn.microsoft.com/azure/developer/github/connect-from-azure-openid-connect
Azure Well-Architected Framework: https://learn.microsoft.com/azure/architecture/framework

CI/CD & Infra

GitHub Actions Docs: https://docs.github.com/actions
Docker Docs: https://docs.docker.com
Caddy Server Docs: https://caddyserver.com/docs

DEV Community: Khanh Nguyen