An agent is a job description the model reads before every message. Three short steps and it can answer clients.
Pick a role, then make it yours
The wizard opens on the role templates - Customer support, Researcher, Marketer and the rest. Picking one fills in a personality, a set of responsibilities, an escalation rule and the tools that role needs. It is a starting point, not a commitment: change any of it.
Starting from scratch is the last card. Take it when none of the roles is close.
Say who it is
Name and Job title are what clients see at the top of the chat. Use a first name and a role people recognise - "Mia, Customer Support Lead" reads better than "Support Bot v2".
Personality is the one field worth writing yourself. Two or three sentences on how it should sound and what it should never do:
Warm but efficient. Answers in two or three sentences, never uses corporate filler, and says plainly when something isn't possible rather than hedging.
Give it something to know
An agent with no documents answers from the model's general knowledge, which is not your returns policy. Upload what it needs under Knowledge: a policy, a price list, an FAQ. Processing takes a moment; once a document is ready the agent searches it before answering and cites what it used.
Try it, then publish
The live preview on the right of the editor is the same runtime a client gets. Ask it the three questions you get asked most. If an answer is wrong, the fix is almost always a document or a line of personality - not a different model.
Publish when it sounds right. You get a shareable link and an embed snippet, and only then can anyone outside your organisation reach it. A draft agent is invisible.
What comes next
A published agent answers when it is spoken to. To have it work on its own - research, drafts, a weekly post - give it a scope of work.