Forum: Azure VM to StaticWebApp
Context & Motivation
Since my last update, I’ve significantly advanced my toolchain. Full integration of OpenAI into my workflow has dramatically boosted my efficiency:
- ChatGPT 4.1 for code, text, and logic design
- DALL·E for image generation
- Codex for end-to-end coding
The following outlines how I leveraged these tools—especially Codex—to re-architect my Discourse-based forum infrastructure.
Problem with the Discourse Setup
As described in this earlier post, I deployed a Discourse forum via a Bitnami image on Azure. While functionally solid, it introduced disproportionate cost for a lightweight use case:
- VM idle cost: ~30 €/month, even with optimizations
- PostgreSQL performance overhead = non-reducible baseline
- Difficult to scale horizontally without deep-diving into AKS/CosmosDB

The visual above illustrates the transition from a bloated resource group to a single Static Web App—eliminating all compute and networking overhead.
New Approach: Static WebApp + Hugo
Given my positive experience with Hugo (which also powers this blog), I transitioned the entire forum to a Hugo-based Static Web App on Azure’s Hobby Tier:
- 0 €/month running cost
- Disqus for commentary
- Room for growth via Azure Functions, if needed later
This simplified architecture now fully replaces the prior VM stack.
Codex-Powered Development Workflow
All development was executed after-hours, leveraging OpenAI Codex for GitHub integration:
- Repo linked to Codex.
- Prompted Codex with functionality requirements.
- Codex auto-generated code + pull requests.
- I reviewed & merged ~40 PRs.

The process exceeded expectations—Codex even migrated a complex C# code-generation module to Python without intervention.
Early Result Preview
After a few dozen iterations and Codex contributions, the new forum frontend was live. The layout was minimal, but fully functional and lightweight:

Feature Breakdown
The final system integrates several modern frontend features, all implemented with Codex-driven PRs:

- Dark mode toggle
- Client-side search via
lunr.js
- Web Share API
- Fully responsive layout
- Collapsible sidebar navigation
- GitHub Actions pipeline
- Kit.com newsletter signup (also now live on this blog)
- Disqus as a lightweight pseudo-forum solution
- Two live entrypoints:
reference.wtf-model.com
forum.wtf-model.com
Conclusion
This migration not only eliminated a recurring infrastructure cost but also redefined my delivery workflow through deep AI integration.
OpenAI Codex demonstrated high stability even under complex build cycles. With this experience, I’m confident in treating it as a dependable contributor to future mid-sized projects.