Shape an agent in an afternoon

A new NetShow agent is a starting point, not a finished employee. The creation flow gives it a name, a face, a voice and a first set of instructions. What turns it into something that does one job well is the hour or two you spend shaping it afterwards. This post walks that road in order, from a blank agent to one you would trust with a stranger.

Start with one job

The most useful agent does one thing clearly. Before you open anything, finish this sentence: "This agent is the ___ for ___." Front desk for a dental practice. Bookings for a studio. First questions for a law office.

The editor asks for exactly that. On the Identity tab, the Role field carries its own hint: what job does this agent do? Our guide puts it plainly: "Front desk for a dental practice" is more useful than "Assistant". A role that narrow tells the agent what to answer and, just as usefully, what to leave alone.

Create it: five steps, one page

The creation flow is Dashboard → Create your agent, five steps on one page, with a preview panel that shows the agent as you go. You type a Name and a Description (a Tagline is optional), then choose its face, its presence and voice, its instructions and capabilities, and review the lot before anything is written.

For the instructions, the flow offers three sources: a Persona blueprint compiled from what you already typed (marked Recommended), Start from a template, or Write your own. Whichever you pick is saved as version 1, and versions only grow: nothing is overwritten later.

Press Create agent and the page hands you straight to your agent's test drive, so the first thing you do is talk to it.

Name and role

After creation, the place you come back to is the agent editor. On your agents list, each agent you own has a small wand button titled Unified Editor. Its first row of tabs, Who Is Your Agent, holds Identity, Personality and Knowledge.

On Identity, Agent Name is the name visitors see and Role is the job in a few words. The Save button at the top saves the tab you are looking at, so press it before you move on.

Instructions, and what it must never do

The Personality tab is where the agent's character lives:

• Soul MD is a free-form description of its character and operating principles.
• Tone sets the register.
• Goal and Background say what it is working for and the context it works in.

Soul MD is also the natural place for the lines it must not cross. Write them as plainly as you would for a new hire: never promise a delivery date, never give medical advice, always hand a complaint to a person. The agent's Edit screen adds guardrails of its own: allowed and blocked topics, and custom rules.

One honest edge: the instruction body written at creation is kept as its own versioned record, and editing that body from your dashboard is coming. Today, the editor tabs above are how you shape the agent after creation.

A face

The face is chosen at creation, in two lanes: Pick a face from the ready-made bases, or Generate a face from the agent's name, where the same name always gives the same face. Both preview live and cost nothing, so take your time in that step.

Swapping the face of an agent that already exists is not something you can do from your own dashboard today. If the face matters to you, choose it while you create. Change how it looks covers what can change afterwards, including the card picture, which you can regenerate from the Edit screen.

A voice that speaks first

A NetShow agent is built to talk. In the editor, the second row of tabs is Channels. Open Realtime, and under Realtime Model & Voice the Realtime Voice menu sets the voice it speaks with.

Then give it a first line. Back on Personality, turn on Enable Welcome Message, which the page describes as "Automatically send a greeting when a user starts a session." Keep it short, say who is talking and ask one easy question.

The visitor's microphone stays theirs: the browser asks for permission first, and a visitor who says no can type instead. Give it a voice that speaks first has every button.

Hear the result

At the top of the editor, next to Save, Chat opens a typed conversation with the agent and Talk opens a spoken one. Ask one question your role should answer and one your Soul MD should stop. Read the answers, not the settings. If an answer does not reflect what you just wrote, say it more plainly, save, and ask again.

FAQ

Do I need to write instructions from scratch?

No. The creation flow recommends a Persona blueprint compiled from the name and description you typed, and it also offers a template you can edit after creation.

Can I change the face later?

The card picture, yes, from the Edit screen. The living face of an agent that already exists cannot be swapped from your own dashboard today, so choose it at creation.

Where do I write what my agent must never do?

In Soul MD on the Personality tab, in plain sentences. The Edit screen adds allowed and blocked topics and custom rules.

Is anyone else talking to it while I shape it?

No. A new agent is private until you publish it, so the editor is the safe place to get it wrong a few times.

Your next step

The guide behind this post walks the Identity, Personality and Knowledge tabs of the agent editor, button by button, and shows how to test each change before anyone else sees it. It is a four-minute read, and it is the order to work in: the role on Identity, the rules in Soul MD, the knowledge your agent answers from, then Chat and Talk to hear the result. Keep the editor open beside it and save one tab at a time.

When the answers sound right, the same guide points you to the voice and the first line, and then to publishing. Nothing reaches a stranger until you press that button.