No made-up answers
The model answers only from topics in your base. No topic, no improvising: the assistant says it doesn't know and hands the chat to a person.
The assistant answers visitors only from your knowledge base: terms, prices, delivery, how to order. If the answer isn't there, it says so honestly and calls a manager. It can run on your own server — then chat history and customer contacts stay with you.
The model answers only from topics in your base. No topic, no improvising: the assistant says it doesn't know and hands the chat to a person.
The assistant handles repeat questions — delivery, payment, opening hours. A manager steps in where a person is needed: an order, a dispute, an unusual request.
When a visitor leaves a phone number in the chat, it goes into a list of leads with a status instead of sinking in message history.
We gather answers from your site, price list, terms and customer chats. You approve every topic — it's the only thing the assistant is allowed to say.
3–5 daysWe connect the model, set up the greeting, tone and rules for handing over to a manager. We add the widget to your site.
2–4 daysWe run real customer questions through it, including awkward ones. Where an answer is off, we fix the base rather than hope the model gets it right.
3–4 daysWe open the assistant to visitors, connect notifications and show your managers how to take over a chat.
1–2 daysAnswers from your knowledge base, calls a manager when it doesn't know
Processing leads, documents and mail; agents inside your systems
From $990: knowledge base, setup, website widget and manager handover. You pay the model provider separately, for actual usage — by message volume (tokens). If you need AI inside email, documents or your CRM, that's AI integration, from $1,990.
From 2 weeks. The knowledge base takes the longest: the tidier your terms and price list are, the faster the launch.
It answers only from approved topics. If there's no matching topic, it says it doesn't know and calls a manager. Most errors come from outdated information in the base, so you can edit the base in the admin panel without a developer.
The assistant can run on your own server — then the knowledge base, chat history and contacts stay with you. The model provider only receives the question and the matching excerpts from the base; with a local model, nothing leaves at all. That's how our product Mimir works.
No, and we don't promise that. It takes care of repeat questions and hands difficult conversations to people. Your managers get less routine and more of the conversations where they're actually needed.
Ari asks three questions — the team comes back with scope, timeline and price.
What you're building, what already exists and how to reach you.
Within one working day — by email or in Telegram.
Agreed before any work starts.