A plugin for your CMS. It reads your pages, your products and your policies, then answers shipping, returns, sizing and stock in your own words, with the page every answer came from attached.
Live in about a minute. Not shipped yet: no card, no charge, and the founding discount when it opens.
And you find out what they actually ask.
No page covers these yet. That is a content plan your customers wrote for you.
Most support AI asks you to document your policies first, then fails because nobody ever does. DFYe reads the documentation you already have: every published page, product and post, split into passages and kept with its source. That is what makes the answers yours rather than generic. Nothing to curate, no embeddings to wire up. DFYe is short for Done For You Employee, and this is the part where it learns the job.
Pages, posts, products, the blog, each language on its own. What it reads and, just as importantly, what it never touches: nothing noindex, nothing behind a login, nothing in draft.
Every published page read and split into passages, so an answer can point at the exact page it came from rather than a vague summary of your site.
How it worksYour blog is answerable too. A visitor asking about something you wrote three years ago gets the answer and the link, without searching for it.
What the bot readsTitles, descriptions, attributes and prices out of your own catalogue, so a sizing or materials question is answered from that product rather than from a policy page that does not mention it. This is where generic support AI gives up and yours does not.
What it indexesAnything noindex, anything password-protected, anything in draft or trash. A per-post switch takes out the rest, and nothing private is ever a passage.
ExclusionsEdit a page and the passage behind it updates. There is no retraining step to remember and no stale answer waiting to embarrass you.
Re-indexingEach answer carries the page it came from as a real link. The visitor can check it in one click, and so can you.
Accuracy and refusalA Dutch visitor is answered from your Dutch pages, not from a machine translation of the English ones. Each language is indexed on its own.
MultilingualConversations go from your site to your own model provider. Nothing routes through us, so there is no markup, no queue and nothing of yours on our servers.
PrivacyConversations go from your site straight to your own model provider. Nothing routes through us, so there is no per-message markup, no queue behind other customers, and no transcript of yours on our servers.
Answers are assembled from your own sentences, so your tone and your policy wording come through without anybody writing a persona document. The widget matches too: not a rounded purple bubble with a cartoon robot, but a default that belongs on a site somebody cared about, and one panel of custom properties when you want it to look like you instead.
Every answer carries a real link to the page it came from. That is the whole trust mechanism, it is on by default, and it is never optional.
You bring your own model key. There is no per-ticket pricing, no queue behind other customers, and no conversation of yours on our infrastructure.
Answers out of your own pages, a source link on every one, no conversation on our servers and a widget that cannot break your theme. The things that make a support plugin a liability are simply not there.
Answers come out of your pages or they do not come at all. Below a retrieval confidence floor it says it does not know, which is worth more to you than a confident invention about your own returns policy.
One file, loaded after paint, rendered in a shadow root. It blocks nothing, shifts nothing, and the size is measured on every build against a budget the build enforces.
No cookies, no third-party analytics, no conversation data on our servers. Your visitors talk to your model provider through your site, and that is the whole data path.
Every colour, radius, font and shadow is a custom property. The widget cannot leak into your theme and your theme cannot reach into it, so it looks the same everywhere.
It answers from this site only, and it will tell you when it does not know. That refusal is the whole point.
Every question, the page each answer came from, and the ones it declined. That last list is a support backlog your customers wrote for you, and it lands in your own admin.
Do you ship to Belgium?·14xWhat is your return window?·11xIs the oak desk in stock?·9xCan I pay with iDEAL?·7x/shipping/·62/returns/·41/faq/·28/products/oak-desk/·17Do you do trade discounts?·6xIs there a showroom in Ghent?·4xCan I collect on a Sunday?·3xA visitor who would have left with an unanswered question gets the answer and the page it came from. And because your pages stay clean HTML with real headings, ChatGPT, Claude and Perplexity can read and cite them too.
A visitor who would have left with an unanswered question gets your answer and the page it came from, at three in the morning if that is when they asked. Nobody was rostered for it.
Clean HTML, real headings and structured data, so ChatGPT, Claude and Perplexity can read and cite your pages as well as your visitors can.
Every declined question is a gap in your own content, listed for you. That log is a content plan written by your customers.
This is what done feels like.
Yes, for the model. DFYe does the setup once the key is in, and your conversations never pass through our servers, so there is no per-message markup and none of your traffic on our infrastructure.
It answers out of your pages and nothing else, which is the whole reason it is personalised to your store rather than generic. Below a retrieval confidence floor it says it does not have that on the site rather than guessing, and every answer above the floor carries a link to the page it came from. The demo runs the same engine, so you can test that claim before you buy.
It is one file, loaded after paint, rendered in a shadow root. It blocks nothing and shifts nothing. The size is measured on every build against a budget the build enforces, and the figure published on this site comes straight out of that build rather than from a slide.
By default every published page, post and product. Never anything marked noindex, nothing password-protected or private, and no drafts, revisions or trashed content. A per-post switch takes anything else out, and editing a page updates the passage behind it without a retraining step.
Because AI Engine is a framework and DFYe is a finished thing. AI Engine has around 100,000+ installs and a 4.9 rating on the WordPress.org listing, checked 12 September 2026, and it is genuinely excellent; if you enjoy assembling your own AI stack it is the better foundation, and our comparison page says exactly that.
On WordPress the PHP is GPLv2 or later, because a plugin there has to be, and every platform gets whatever its own licence requires. What you buy is updates, support and the licence key, never permission to run the code. A lapsed licence keeps working; it just stops receiving updates.
None of them yet, because it is not on sale. It is being built for WordPress and WooCommerce first, and is planned for Shopify, Magento, Drupal, Joomla, Shopware, Craft CMS, Ghost and TYPO3. The state of each one is listed at /works-with/, and no platform is called available until a real build has answered a real question on it. If yours is not on the list, say so and it goes on it.
Neither. It sits in front of whatever you already use and answers the questions your own pages already cover, which is most of the repeat volume. What it cannot answer goes to your inbox, your contact form or your existing live chat, with the question and the pages it looked at attached. Your helpdesk keeps doing the part that needs a person.
It says the site does not cover that and offers the handover you configured. The question is logged in your own admin either way, which turns the things your site fails to explain into a list you can actually work through.
It is short for Done For You Employee. Answering the same forty questions about shipping, sizing and returns is a job, and the point of the product is that you do not have to do it or hire for it. It reads your store, answers what your own pages already cover, and passes you only the questions that actually need you.
No, and it will not pretend to. DFYe answers from your published content: policies, product pages, shipping terms, opening hours, FAQs. A question about one specific order needs your order system and a person, so that is exactly the kind of question it hands over, with the conversation attached. If what you want is a bot that issues refunds and edits orders inside your shop, that is a different and more expensive kind of product, and you should buy that one instead.
Because the only thing it ever reads is your store. Your product descriptions, your delivery terms, your returns window, your tone, in every language you publish in. There is no shared model of your industry and no persona document to fill in. When you edit a page the passage behind it updates, so what it says about you changes when you do.
One-time or annual, three sizes: 89 euro once or 39 a year for a single site, 249 or 99 for five, 599 or 249 for twenty-five. There is no free tier and no per-message fee, because you bring your own model key and the running cost is yours.