Every course so far in this track that touched the PHP AI Client SDK or MCP assumed someone was logged in. Connect AI Assistants & Agents to WordPress is the clearest example: an admin authenticates an external client like Claude Desktop with an Application Password, and everything downstream, permissions, audit logs, rate limits, is built on top of that one authenticated WordPress user. That’s the right model for an internal assistant. It is the wrong model for a chat widget sitting on your public homepage, because the person typing into it isn’t logged in, doesn’t have a WordPress account, and never will.
This course builds the opposite surface: an anonymous site visitor talking to a widget
embedded in a page, with no Application Password, no MCP client, and no current_user_can()
check to lean on. You’ll build a real chat widget, wire it to a custom, lightly-protected
REST endpoint whose PHP callback is the only thing holding your AI provider’s API key,
handle sessions for visitors who have no server-side user object to attach state to, be
honest about what streaming actually means with today’s SDK, and add the guardrails and
human handoff a public-facing AI surface needs that an internal admin tool doesn’t.
If you haven’t yet, start with WordPress AI Foundations
for the underlying wp_ai_client_prompt() setup this course builds on. And because a
public chat widget that hallucinates about your own products or policies is a real
liability, Grounding AI Answers in a WordPress Knowledge Base
is the recommended companion course for making sure the widget’s answers come from your
actual site content instead of a model’s best guess.