WP Agentic Admin

The Local-First AI Site Reliability Engineer (SRE) for WordPress

We’re building a fully local, privacy-first AI agent that lives directly in your browser and acts as your personal WordPress Site Reliability Engineer, allowing you to execute tasks within the protected realm of your WordPress instance. We’re pushing the limits of WebGPU and WebLLM to run a Small Language Model entirely on the client side. No server costs. No API calls. Complete privacy.

How it works  

You simply tell it: “My site is throwing a 500 error”

The Agent uses its thinking process to:

1. Read your debug log
2. Identify the rogue plugin
3. Propose a fix
4. If there is a solution, execute it at the user’s command

The Vision

Make agentic code work within smaller models so that workflows and tool calling can be possible via smart reasoning techniques to help users figure out issues inside their WordPress install and fix them without sharing sensitive data.

What’s Working

• Local AI runs in-browser via WebLLM
• Adaptive tool selection + workflows
• Integration of WordPress abilities
• Service Worker mode keeps model loaded across page navigation
• Integration with WordPress Abilities API
• Automated test coverage (partially)

What Needs Your Help

Small language models face specific challenges that larger models handle easily. We’ve identified 5+ critical improvement areas we could work on, such as reasoning, expanding on the abilities end as well as on the workflows and tooling creating. Also, feature upgrades that we might be missing that are super useful for the whole ecosystem around WordPress.

Target Audience

Frontend Wizards (WebGPU/WebLLM)

• Wrangling the WebLLM library
• Managing model state in-browser
• Building the chat interface that talks to the GPU
• Optimizing Service Worker persistence

WordPress Integration Engineers

• Building Abilities API integrations
• Writing PHP functions for safe cache flushing, SQL queries, plugin toggling
• Ensuring WordPress security best practices

Prompt Engineers & AI Whisperers

• Design System Prompts for small models
• Build ReAct loop logic that teaches tool usage
• Explain a debug.log to a small language model
• Test behavioral edge cases

QA & Testing Engineers

• Write test cases for edge cases
• Create behavioral test suites
• Stress-test with real WordPress environments
• Measure success rates

Hackathon Goals

1. Get the local LLM to successfully “reason” and choose the right tool
2. Improve small model reliability  
3. Better error handling  
4. Smarter error log analysis  
5. Add more abilities
6. Add more workflows
7. Google AI API introduction via browser extension
8. Adoption of core proposal feature WP AI Client