Connecting WordPress to cloud-based AI APIs involves recurring costs, latency, and data privacy concerns. But what if a fully autonomous AI agent could manage your website while operating entirely from your users’ browsers?

Using our open-source plugin, wp-agentic-admin, as a case study, this highly technical talk explores how to build client-side, AI-powered website administration. We will take a detailed look at how to combine the capabilities of modern browsers with intelligent execution patterns:

The infrastructure: How to leverage WebGPU and WebLLM to run open-source models such as Qwen entirely locally, using Service Workers to maintain the agent’s state across WordPress admin screen updates.

In-browser RAG: How we implemented a fully client-side Retrieval Augmented Generation (RAG) workflow, using in-browser vector storage to inject local WordPress records and documentation directly into the model’s context window.

The brain: How to implement a client-side ReAct (Reasoning + Action) loop, enabling the model to autonomously choose tools, read error logs, check site status, and apply fixes through the WordPress Abilities API.

By the end, you will have a clear action plan for how client-side processing can eliminate API bills and privacy risks while bringing truly professional-grade workflows directly into the WordPress ecosystem.

Presentation Slides »