An AI Assistant That Advises Your Car Dealership Around the Clock

Yes. We train the assistant on your vehicles, equipment, prices and availability so it can provide information on model, condition, mileage and price. The data basis is refreshed regularly; a source counts as stale after twice its refresh interval and is then picked up. Every index run writes a new version and only switches over at the end — if it fails, the previous state keeps answering instead of leaving a gap. Answers stay bound to your sources: sourced answers name their source, and an answer without a used citation is recorded as a gap. The knowledge base page explains the details.

Not in the same instant, but without you doing anything. Website sources carry a refresh rhythm, seven days by default; it can be set from one day up to 365 days, and the value zero switches it off. The run happens at night. If a change should take effect sooner, trigger the read manually under Knowledge and then check the state of the source to see how many sections were produced. Each run writes a new version and switches over only at the end: if it breaks off, the previous state keeps answering, and a run that wrote nothing deletes nothing either. Five versions are kept. More under knowledge base.

Yes, as long as the document has a text layer. The knowledge base accepts files up to 10240 KB and splits them into overlapping sections so a passage does not end mid-sentence; a PDF is read through its existing text layer. A scanned sheet without a text layer deliberately yields nothing — there is no text recognition from the image, and the reason is shown as a message on the source instead of silently missing. For the question, at most five search hits plus your text blocks then go to the model, numbered, and sourced answers name the excerpts they rely on. For prices, terms or deadlines that have to be exact, the maintained question-and-answer pair is the better place: it goes out verbatim. More under knowledge base.

A test drive becomes binding only once your dealership confirms it; the assistant takes the request for it. That runs through the appointment tool type — it writes and therefore never runs unconfirmed: the proposal states vehicle, day and time in plain text in the chat, only the visitor's click creates the request, and once you confirm it in your view, the visitor receives a confirmation email, if they gave an email address. On request the assistant names free times from your stored kinds of appointment — time windows, duration, lead time and buffer — minus the appointments already requested or confirmed in XICBOT; it does not compare them with an external calendar, and a time it names remains a suggestion. The assistant enters the test drive straight into your calendar only with a calendar connection set up in the Custom package. A request is only possible if its kind is known to your dealership, the lead time fits and the moment falls into an approved window; occupancy is checked under a lock, and a double click creates no second test drive. An hourly run sends the reminder for the confirmed appointment by email, if an address is on file. Workshop and service appointments take the same route, only with a different kind. More under booking assistant.

The assistant checks occupancy under a lock: the check and the entry happen in one go, so a time slot is not given out twice. Against a double click there is a bolt in the database rather than in the program flow — the same request submitted twice does not create a second request. An appointment request becomes an appointment when your business confirms it. Without a calendar connection the assistant counts the appointments already requested or confirmed in XICBOT; what is only in an external calendar it does not see. With a calendar connection set up in the Custom package it reads free times through the calendar-availability form and writes the entry through the calendar-booking form; then the free times offered follow your calendar — including appointments that came in by phone or email. The check under a lock does not apply on this path: whether a slot taken in the meantime is rejected at booking is decided by your calendar. XICBOT sends each booking with an idempotency key by which your calendar can recognise a booking submitted twice.

Yes, both live in the same tool type as the appointment request: appointment takes appointment requests, reschedules and cancels. Each of these acts writes and therefore requires confirmation, and the proposal names the act as its own sentence — cancel, reschedule or request is spelled out before anything is clicked. The text of the proposal is stored before the confirmation and not generated anew: what gets confirmed is exactly what was read. A new time has to meet the same rules as a new request — known kind of appointment, matching lead time, approved window, occupancy checked under a lock — and a rescheduled appointment is a request again until your business confirms it. If a case turns delicate, the assistant is meant to call the handoff tool type; that sets the conversation to escalated and notifies a person in your company.

No. A binding financing offer is hard-blocked in the assistant's rule set, together with legal advice, medical information and the assessment of allergies. Instead it captures the enquiry in structured form — vehicle, term, down payment — and hands it as a lead to your sales team, with the conversation as context. Conditions, creditworthiness and contract details are checked and confirmed by a person in your company. Figures, prices and deadlines that are not in your sources are replaced with a notice along the way, and if it finds nothing in your knowledge base, it says that nothing on this is on file instead of estimating an instalment.

No. It takes the recurring questions off your sales people and is meant to pass matters on as soon as they get concrete. Standard questions are answered from your knowledge base, and the answer is bound to the sections that were found; if it finds nothing, exactly one sentence appears in the chat instead of a guessed statement. Appointment requests are taken by the appointment tool type, enquiries collected by the lead type with name, email, phone and request, together with the files the visitor uploaded earlier. If a matter turns urgent or delicate, it is meant to call the handoff type; that sets the conversation to escalated, creates the lead and notifies a person in your company. Your advisors take over a prepared conversation with the full transcript instead of an anonymous note.

The room for that is drawn tight. For every question at most five search hits from your knowledge base, plus your text blocks, go to the model, numbered, and sourced answers name the excerpts they rely on. Figures, prices and deadlines that are not in your sources are replaced with a notice. If the search finds nothing or the model declines, the assistant says that nothing on this is held in the knowledge base — and the case is logged as a knowledge gap, as is an answer without a used excerpt. Its rules forbid it to invent contact details, file numbers or appointments, or to pose as a human being. An error cannot be ruled out entirely with any language model; hence the binding to sources, the visible citation and the handover to a person.

For a dealership with vehicle advice plus appointment and lead functions the Pro package fits: setup from 2,900 € net, ongoing operation from 149 € per month net including hosting in Germany, care and updates. The vehicle inventory enters the knowledge base as its own source, so the assistant answers from your data and not from memory; Pro allows 50 knowledge sources and 2,000 documents for this. If the inventory list is to come continuously from a system of your own, that leads to the Custom package with tool and API control from 4,900 € net setup. We clarify the scope in the free initial consultation; all packages are listed under pricing. All services can be cancelled monthly; there is no minimum term.

No. Your data is not sold and not passed on to third parties. The subprocessor register lists the self-operated language model gateway in Germany; behind it, the answers are generated by a specialised model provider, also outside the EU on the basis of standard contractual clauses. Its purpose is generating the answers, what is passed on is the conversation history and the excerpts found for it in your knowledge base, and the legal basis is Article 28 GDPR. A visitor's address is never stored in clear text but as a salted fingerprint per account; for IPv6 the device part is dropped beforehand. Conversation analytics evaluate only what arises in your own account: no browser identification is read for it, and no profiles are built as a precaution.