Try the live assistant
AI assistant for WordPress

AI answers your site can stand behind.

Grounded in your own content, cited every time, honest when the answer isn’t there — on your own AI key, with the consent and retention machinery already built in.

A WordPress plugin. Install from your dashboard; runs on any site you control.

Example exchangeauthored · not a captured log
Visitor10:42
Can AnswerRails book appointments for my clinic?
Assistant · declined10:42
I don’t know — there’s nothing about appointment booking in this site’s published docs, so I won’t guess.
logged → refusal on record · offered handoff
Assistant · grounded10:43
It does answer product and policy questions from your own pages — with sources your visitor can click.
grounded in → /docs/how-it-works/what-it-does/ · link verified
Nothing invented

Those two exchanges are illustrations. Try the same two questions on the real assistant — it answers from this site’s published pages and cites what it used.

Five checks between a question and an answer

Every visitor message passes the same pipeline, on every tier. There is no configuration in which a check is skipped.

1 · Consent nothing sent until agreed 2 · Moderation first abuse never reaches your key 3 · Grounding your published content only 4 · Citation check bad links fail closed Answer, with sources citations your visitor can click “I don’t know” + handoff an honest no, on the record

The refusal is a designed outcome, not a failure state: when your published content can’t support an answer, the assistant says so, offers a human route, and the exchange goes on the record. When it says “I don’t know” →

Trust is free. Leverage is Pro.

Free is a complete assistant: any provider, any model, on your own key, grounding your site’s own pages honestly and provably, with the whole privacy layer inside it. Pro adds business leverage — your catalogue and the web as sources, and what you do with what visitors ask. Every privacy and compliance control is free — and we don’t move them behind payment.

Free

trust, on every tier
  • Grounded answers from your own pages, posts, documents and JSON
  • Citations on every answer — visitors see the sources
  • Honest refusals when the content isn’t there
  • Any provider, any model, on your own key — OpenAI, Anthropic or Mistral
  • Moderation on every message, and the whole privacy layer: consent receipts, storage modes, retention and DSAR — controls and all

Pro

business leverage
  • WooCommerce catalogue & external-URL sources
  • Business analytics on what your visitors actually ask
  • Idea & lead capture from real conversations
  • Human-handoff escalations with transcripts

Your key. Your provider. No middleman.

AnswerRails is bring-your-own-key: your site talks to your own account at OpenAI, Anthropic or Mistral. Conversations don’t route through our servers — because there aren’t any in the path.

Predictable economics come with it: usage is billed by your provider at their rates, to your account. New installs default to Mistral, whose API key is free to create — so a working assistant costs nothing to run — and the setup walkthrough covers turning off free-tier training before the first question is answered.

Your key, your bill →   Where your data goes →

Compliance machinery, not a compliance badge.

Plenty of plugins name-check GDPR. AnswerRails ships the working parts — the things a data-protection question actually asks for.

Consent, recorded

Server-side consent receipts, including the disclaimer text agreed to. Visitors can withdraw from inside the chat as easily as they gave it.

→ docs/guide/consent-privacy-retention/

Retention, enforced

Logging modes and retention periods are policy the plugin executes — anonymise or purge on schedule, not a promise in a privacy page.

→ docs/guide/consent-privacy-retention/

DSAR, built in

WordPress’s own Export and Erase Personal Data flows cover chat data, with subject search, export and anonymise-in-place for direct requests.

→ docs/how-it-works/where-your-data-goes/

Moderation, always

Every visitor message is safety-checked before it reaches the model — abuse never engages your brand and never reaches your provider account. No off-switch, on any tier.

→ docs/how-it-works/what-it-cant-do/

“There are questions this product is built to not answer.”

Off-topic, ungrounded, or unsafe requests end in a designed refusal — visible to you in the logs, honest to the visitor on the page. That is the feature.

Documentation that answers for itself

These docs are also the assistant’s knowledge base on this site — every answer it gives here is grounded in, and cites, pages published here. Written to be read on their own: no account, no sales call, no demo gate.

Or just ask: open the assistant and put a question to this site’s published pages. If the answer should have been in the docs, that tells us which page to write next.