Context Assistant

Context Assistant adds an AI chat widget to your site — but unlike a
generic chatbot, it is built around three ideas:

• Page context. The assistant knows where the visitor is: which
  post or product is open, its title and categories. Never the page
  body, never form input, never cookies — only small facts you allow.
• Answers from your content. Published posts, pages and products
  are synced into a knowledge base; the assistant answers questions
  about your site, grounded in retrieval, not model memory.
• Safe actions. The assistant can act — search content, draft
  posts, check a customer’s own order, update stock — always with the
  signed-in user’s own WordPress capabilities, and any change only
  after an explicit preview card and an Apply click. Deletions need a
  second confirmation and go to the trash, never permanent removal.

The heavy lifting (models, retrieval, orchestration) runs on a Context
Assistant server; this plugin connects your site to it. Your secret API
key is encrypted at rest and never enters browser code. Public chat uses a
public identifier and temporary guest tokens. Private integrations keep
using WordPress-issued sessions and permissions.

WooCommerce: with WooCommerce active, the assistant additionally
searches the catalog with live prices and stock, shows a signed-in
customer the status of their own orders, and lets shop managers update
stock through the preview/Apply flow.

For developers: add your own assistant tools with the
context_assistant_tools filter, extend the page context with
context_assistant_page_context, and control widget visibility with
context_assistant_should_render. WP-CLI: wp context-assistant sync.

External services

This plugin does not work on its own: the models, retrieval and
orchestration run on a Context Assistant server that you configure
in Settings → Context Assistant (the API URL). That server is
the hosted Context Assistant service.

What the plugin sends to that server, and when:

• On pages where the configured widget is enabled — the browser loads
  https://contextassistant.io/assets/embed.js over HTTPS. This service request
  exposes normal connection information such as IP address and user agent.
  The script request uses a no-referrer policy; no API key or chat token is
  included in its URL. The widget is not loaded on unconfigured/disabled pages.
• When the widget loads — a short-lived embed token is minted for
  the browser to fetch widget settings and restore conversation history.
  The page context you allowed (page URL,
  titles, public taxonomy terms; each behind its own privacy toggle,
  never the page body or form input) is attached to the conversation.
• On each message — the visitor’s message text and that page
  context, so the assistant can answer. Public chat uses a random guest
  identity, not a caller-supplied user ID. A signed session credential is
  stored in browser local storage for up to 30 days between renewals.
  Existing private integrations keep their WordPress identity handling.
• When knowledge sync is enabled — the text of your published
  posts, pages and products (never drafts or private content), so the
  assistant can answer from your own content.
• When the Remote Tools bridge is enabled — the server calls back
  into your site to run tools under the signed-in user’s own WordPress
  capabilities; your secret API key stays encrypted on your server and
  is never sent to the browser.

Conversation transcripts are stored on that server and can be exported
or deleted (see the FAQ). The service processes messages according to
its privacy policy. The widget uses browser local storage to remember
whether a visitor has dismissed or engaged with its automatic greeting.

If you use the hosted Context Assistant service, it is operated
under its own terms of service and privacy policy:
Terms: https://contextassistant.io/terms/
Privacy: https://contextassistant.io/privacy/