MW
Deep dive 15 min read

Building an AI WordPress assistant

Stitch tools, memory, and planning into an assistant that can run multi-step WordPress workflows.

Assistant Memory Planning Agentic
Auto-cycle
1. The assistant observes context and plans the next tool call.

Click a node to step through

1. Overview

A WordPress-aware assistant is a small loop: read context, plan, call tools, observe, repeat. The tools are MCP. The context is your site.

2. Memory systems

Short-term

Conversation buffer (last N turns).

Episodic

Per-task journal stored in D1.

Semantic

Embeddings of posts/comments in Vectorize.

3. Planning loop

  1. 1

    Observe

  2. 2

    Plan

  3. 3

    Call

  4. 4

    Reflect

4. Tool catalogue

Surface only the tools relevant to the task. Too many tools confuses the planner; too few cripples it.

5. Evaluation

Build a small eval set with expected tool sequences. Run nightly. Fail the deploy on regression.

An assistant that knows your site

Three memory tiers. One planning loop.

Tools picked per task, evaluated nightly.

  • Memory
  • Tools
  • Eval-gated
Advanced workflows