> ## Documentation Index
> Fetch the complete documentation index at: https://docs.watx.in/llms.txt
> Use this file to discover all available pages before exploring further.

# Create an agent

> Start blank or from a role, name it, and understand the plan limit before you switch it on.

New agents are made from **AI Agents & Bots → New agent**. Only owners and admins can create, edit or delete one; everyone else can read the studio.

<Steps>
  <Step title="Choose a starting point">
    **Blank agent** gives you an empty form. A role fills in the tone, the ground rules, the skills and the handoff wording that job usually needs.
  </Step>

  <Step title="Name it">
    The name is what you call it in this list. What it calls itself to a customer is the **Agent name** on the Persona tab, and it is a separate field.
  </Step>

  <Step title="Press Create agent">
    You land in the studio. The agent is **switched off** — nothing reaches a customer until you review it and turn it on in Routing or Behaviour.
  </Step>
</Steps>

## The roles you can start from

A role is a set of defaults, not a type. Once created, an agent from a role is an ordinary agent: every field stays editable, nothing re-applies the role later, and no field records where it came from.

| Role                 | What it is set up for                                                         | Skills switched on                                                        |
| -------------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------- |
| **Customer support** | Answers from your knowledge base and hands over anything it cannot source     | FAQ & support, Human handoff                                              |
| **Sales assistant**  | Qualifies new enquiries, answers product questions and books the next step    | FAQ & support, Lead qualification, Product recommendations, Human handoff |
| **Appointments**     | Takes booking requests, states availability and shares your booking link      | Appointments, FAQ & support, Human handoff                                |
| **Order tracking**   | Looks up an order, reports its status and explains the returns policy         | Order status, Returns & refunds, FAQ & support, Human handoff             |
| **Lead qualifier**   | Collects the details your team needs before a human picks the conversation up | Lead qualification, FAQ & support, Human handoff                          |

Each role also sets a tone, a reply length, a set of ground rules, the phrases that always hand off, and the handoff and fallback wording. It deliberately does **not** set the channels — that is a decision about your channels — and never switches automatic replying on.

<Note>
  A role can switch on a skill that needs an integration. **Appointments** is the one to watch: without Google connected it cannot check a calendar or book anything, and falls back to sharing the booking link you give it. See [Skills](/ai-agents/skills).
</Note>

## How many agents you can have

| Plan       | Agents    |
| ---------- | --------- |
| Starter    | 1         |
| Growth     | 5         |
| Enterprise | Unlimited |

A meter under the page title reads **2 of 5 agents on your plan**, and turns amber when it is full. **Paused agents count** — a cap you could dodge by switching an agent off would not be a cap. When you are at the limit the **New agent** button is disabled and its tooltip says how many your plan includes; [upgrading](/getting-started/plans-and-billing) is the way past it.

## Duplicating

The row menu has **Duplicate**. The copy carries the persona, voice, skills and their settings, the escalation wording, the channels and the agent's selection of knowledge documents and actions. It does **not** carry:

* the name — the copy is called *\<name> copy*;
* its position in the list — a copy goes to the bottom;
* **on/off** — a copy is always off;
* **test mode** and the test numbers — those point at one person's phone for one rollout.

A duplicate counts against the plan limit like any other agent.

## Deleting

**Delete** on the row menu removes the agent and its settings. Conversations it answered stay in your inbox but stop being attributed to it, and a thread it was handling is re-routed to whichever other agent covers that channel. The knowledge library and the custom actions are shared by the whole workspace and are not touched.

## Turning it on

Two switches decide what a new agent does, and both start off:

* **Agent enabled** (Routing, or the switch on the list row) — the master switch. With it on, the agent can be asked for a draft in the inbox.
* **Answer inbound messages automatically** (Behaviour) — the agent replies to customers by itself.

Before you turn the second one on, read the readiness checklist at the top of the studio and try the agent in the [test drawer](/ai-agents/testing).