How Torben works
The building blocks behind every answer and action, and why Torben keeps its context small.
Every time Torben handles a request, four things decide what it does.
| Building block | What it decides |
|---|---|
| Channels | Where Torben answers, and whether it answers, drafts, or only watches |
| Knowledge | What Torben is allowed to say |
| Playbooks | How Torben handles a kind of request, step by step |
| Capabilities | Which actions Torben may take, and which need a person's approval |
One request, end to end
A message arrives. A customer writes on a channel where Torben is on.
Torben picks a Playbook. It matches the request to the Playbook written for that kind of request.
It looks things up. It searches the knowledge bases picked for that channel and calls the actions the Playbook names, such as an order lookup.
It acts or asks. Actions set to run automatically run. Actions that need approval wait for a person.
It replies or hands over. Torben answers in the customer's language, or hands the conversation to your team with a short summary.
Small context, better answers
A general chat assistant gets everything you paste into it. Torben loads only what one request needs: the matching Playbook, the right knowledge base, and the actions that Playbook allows.
Less context makes Torben more focused, and it makes each answer cheaper to produce. It also limits what Torben can do on any one request to what you allowed.
Torben only does what you set up
- It answers only from knowledge bases you chose for that channel.
- It acts only through actions that an active Playbook names.
- It runs each action the way you set it: automatically, or after approval.
Read more in Safety and control.