How it works
One conversation, one ticket, one decision, one job done. This is the whole path.
Front desk: the conversation
- 1
It introduces itself.
Kolli opens every conversation by saying it is an AI and on whose behalf it acts. No setting can switch that off.
- 2
It listens.
The customer describes the problem in their own words, in Hungarian. Typed or spoken.
- 3
It asks back.
Kolli confirms the topic and collects what the ticket needs: which system, what the screen says, since when, which username. Two questions at most. Never a password.
- 4
It answers, if a reviewed answer exists.
Word for word. If the customer can solve it themselves, it shows how and files nothing.
- 5
It files the ticket.
Only after an explicit yes. The ticket number is read back at once.
Back office: the work
- 6
It reads.
The new ticket, the customer's history, your knowledge base.
- 7
It writes a decision brief.
A summary that fits a phone screen: cost, size of the change, risk, who is affected, how to roll it back, what we promise to others. Then a recommendation: approve, decline, ask.
- 8
A person decides.
Approval is the ticket's state, not a message. Recorded, timestamped, attributed. It cannot be skipped and it cannot be talked round.
This is the human step
- 9
It does the work.
Only what the approved ticket allows, with a fixed set of tools. Before every change it re-checks the ticket's state.
- 10
It drafts the reply.
In the customer's language, with no internal detail. A person sends it.
- 11
It summarises.
On closing, two short texts: what this was, and how we solved it. Hungarian and English. A person approves; then the customer sees them.
Why so strict
Because a sentence sent to a customer cannot be unsent, and a change to a system cannot be made not to have happened. Kolli prepares everything. The outgoing word and the action on a system are approved by a person. The system enforces it; good intentions do not.