An AI hire that knows your business

One answer from the assistant, including an answer to a suggested question the visitor clicks. What the visitor types costs nothing.

Trial and Starter answer with GPT-6 Luna. On Growth and Scale you choose GPT-6 Luna or Claude Haiku 5.5, and every answer is still one message. On Enterprise you can pick any model, and the heavier ones count as several messages: Gemini 3.5 Flash as 5, Claude Sonnet 5.5 as 20 and Claude Opus 5.5 as 50. The rate is shown next to each model when you choose, and nothing changes without your decision.

The assistant keeps answering at no extra cost, up to twice your allowance, and you get an email and a banner in your dashboard. After two allowances it pauses until the next period, and customers can leave their details if your contact form is on. You can ask for a bigger plan in your dashboard, and top-ups ($9 per 1,000 messages) arrive when billing goes live.

It answers only from the material you give it, and an answer from your website carries a link to the page it came from. When the answer isn’t there, it says so instead of guessing, and if the customer wants to get in touch, it can offer your contact form. The question lands in your dashboard so you can fill the gap.

No. Order status isn’t on any page, and the assistant only knows stock levels from the last time it read your site, so they can be out of date. It says so and can offer your contact form instead.

Yes. When someone on your team has the dashboard open with “Take live chats” on, a customer’s request for a person goes to them: they take over the chat and reply themselves, and the assistant stays quiet meanwhile. If nobody picks it up within 90 seconds, or nobody is taking chats, the customer can leave their details in the contact form, if you’ve turned it on. Replies from people don’t use messages from your plan. It works on every plan.

English, Polish, German, French and Italian. It replies in the language the customer writes in, and the chat window uses the language of their browser or the one you set in the dashboard.

PDF (scans too), Word (.docx), Excel (.xlsx), CSV, TXT, Markdown and photos in JPG, PNG or WEBP, up to 50 MB per file. The text of your files counts toward your plan’s characters of knowledge, together with the text of your website, notes and Q&A pairs; there is no separate allowance for files. Text in scans and photos is read by an AI model, on every plan at no extra cost. We don’t keep the file itself, only the text read from it, which you can check and correct in your dashboard.

On servers in Germany (EU). To make your content searchable, our embedding provider converts its text into search vectors. Each question is converted on our own server in Germany and goes to that provider only as a fallback. To write an answer, the question and the matching excerpts of your content go to the provider of the AI model that writes it, and for that model we route only to providers that don’t collect request data. The DPA and the list of subprocessors are linked in the footer. Your pages are never used to train models.