The story
This page tells the story behind the iO Smart Search plugin and how it was vibe‑coded by Per Englund at iO Campus Stockholm with the help of Cursor. Per started with a clear goal: replace fragile keyword search with something that actually understands what people are looking for, even when they do not phrase it exactly like the content on the site.
With Bedrock, the Pilago starter and a clean plugin architecture, he designed a backend‑agnostic system that can run on classic WordPress, REST APIs and WPGraphQL without locking editors into a single stack. Instead of building throwaway demos, the focus from day one was production readiness: proper hooks for indexing, MySQL and Qdrant vector backends, OpenAI and Bonzai embeddings, and a hybrid ranking model that balances keyword relevance, semantic similarity and editorial authority.
Per’s background is in digital experiences where content, design and technology have to work together. That’s why the plugin is built to be editor‑friendly first: one central settings screen, clear authority levels for content, per‑post controls and transparent debug information so you always understand why a certain result appears. It should be possible to hand this to a content team and let them tune it without touching code.
On the technical side, the plugin talks to both OpenAI directly and iO’s own Bonzai platform, with proper fallback so the site keeps working even when providers have a bad day. It can start with a simple MySQL vector table in a local MAMP environment and later grow into a Qdrant‑powered setup without changing themes or frontends, which makes it suitable for both prototypes and long‑lived projects.
The development process for iO Smart Search has been a collaboration between human judgement and AI‑assisted tooling. Cursor has been used both as a coding assistant and as a way to keep a living technical README, capturing decisions, trade‑offs and discarded ideas along the way. The result is a plugin that is not just a demo, but something iO would be comfortable running in production for real clients.