AI replies to customers — from your terms, nothing else
Questions about delivery, payment and stock arrive all day and repeat. Oxta answers them in two seconds, taking the wording from your own terms rather than from the internet.
How it works
You upload your terms — delivery, payment, returns, warranty — as one file or several. Oxta splits them into fragments and looks up the right one when a question arrives.
- A question arrives
A message from Telegram, a form on your site or an inbox enters the scenario.
- Search your terms
The search runs over your dataset, not over the model’s general knowledge. The fragments it finds are the only material the answer may be built from.
- Answer, or hand it to a human
If your texts do not contain the answer, the AI says so and the conversation stays with a person. An invented answer is worse than silence.
What you need to connect
- Your terms as text or files: delivery, payment, returns, opening hours.
- The channel the questions come from: Telegram, a form on your site, or email.
- Optionally your own AI provider key, if you would rather pay the provider directly.
What a run looks like
Each run shows not only the answer but the exact fragments of your terms it was built from. So it is easy to see why an answer came out the way it did and fix the source rather than tweak the wording.
When you update the terms, your sample cases are replayed: if a routine answer got worse, you hear it from Oxta and not from a customer.
What it costs
One answer is one run plus one AI operation. The free tier covers 1,000 runs and 25 AI operations a month; Starter covers 10,000 runs and 300 operations.
Bring your own provider key and those operations are billed by the provider directly, without touching your Oxta allowance.
The essentials, briefly
Can the AI make something up?
The answer is assembled only from fragments of your own texts. With no matching fragment, the scenario says there is no answer and hands the conversation over.
Does the AI provider see customer data?
Only the question text and the retrieved fragments of your terms are sent. The providers and the transfer rules are set out in the Data and AI document.
How long does setup take?
Usually about ten minutes: upload the terms, wait for processing, then dry-run a few of your real questions.
Can it answer in several languages?
Yes. The reply follows the language of the question while the source stays your own text.