> ## 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.

# Testing and Build with AI

> Walk through a flow in a phone preview before anyone else does, and draft one from a sentence.

## The simulator

**Test** in the editor toolbar opens a phone preview of the flow as it stands — including
unsaved changes. It starts at the entry node and walks forward exactly as the engine
would: messages appear, buttons are tappable, list rows open, and a text box takes the
place of the customer typing.

The restart button clears it and starts again from the entry node.

<Warning>
  **The simulator never calls anything real.** No message is sent, no payment link is
  created, no endpoint is called, no tag is written. Everything with a side effect is
  reported as a line telling you what *would* have happened.
</Warning>

What it stands in for:

| Node                                                                  | In the preview                                                                                           |
| --------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- |
| **Template**                                                          | Shows which template goes out. WhatsApp renders the approved copy, so the preview cannot.                |
| **Take a payment**                                                    | "Would send a payment link for …". No link is created.                                                   |
| **API request**                                                       | "Would call `GET https://…` — not sent from the preview."                                                |
| **Wait**                                                              | Says how long it would wait, then carries straight on.                                                   |
| **Ask location** / **Ask for a file**                                 | Explains what WhatsApp would show, and lets you type anything as a stand-in for the pin or the file URL. |
| **Tag contact** / **Segment contact** / **Set attribute**             | Shown as a line saying what would be written. Nothing is written.                                        |
| **Connect flow**, **Handoff to agent**, **Hand to AI agent**, **End** | Shown as the end of the run, saying where it went.                                                       |

One limitation worth knowing: an **If / else** can only be evaluated in the preview when
it branches on a **flow variable**. A rule about a tag or a contact field has no real
contact behind it at design time, so it takes the false path.

Use the [runs screen](/flows/waits-and-runs) for the real thing — it carries the engine's
own event log for every run that actually happened.

## Build with AI

Describe the flow you want and get a draft on the canvas. There are two ways in.

**From the flow editor.** The prompt bar sits in the toolbar between the name and the
buttons. Suggestions to start from: "Greet new customers, ask what they need, and route
them to sales or support", "Take a food order: menu buttons, then ask for the delivery
address".

<Warning>
  **A draft replaces the whole canvas.** A generated flow is a whole graph; merging it
  into an existing one would leave two disconnected halves. On a flow that already has
  nodes you are asked to confirm, and told what is about to go — "6 nodes, replacing your
  3". An empty canvas skips the confirmation.
</Warning>

**From the flows list.** The **New flow** dialog has the same prompt at the top. The
draft travels to a newly created flow, and the flow row is only created once a draft
comes back — so a failed generation leaves nothing behind.

Either way, **nothing is saved**. The draft is editor state. You press Save, and it goes
through exactly the same validation as a flow you drew by hand. If you do not like it,
do not save it.

### It costs AI credits

Each generation spends [AI credits](/ai-agents/credits-and-byok) from the workspace
wallet, metered separately from the automation drafter so the two can be told apart. A
workspace running on its own provider key is billed by that provider instead.

### What it repairs, and what it leaves you

A flow is a graph, so a generated one can be wrong in ways an automation cannot. Each of
these is fixed before the draft reaches you, and each is reported as a note you can read:

* **Two nodes with the same key** — one is renamed.
* **A path pointing at a node that was never created** — it is left unconnected, so you
  see the gap on the canvas and join it up.
* **A missing or unknown entry node** — it falls back to the first node.
* **A node type this version does not know** — dropped rather than drawn as a card no
  form can edit.

And what it deliberately will not do:

* **It never invents an id.** Tags, segments, flows, agents, product ids and template
  names come back blank and are listed as things you still have to pick.
* **It never activates anything.** A draft is always a draft.

The toast after a successful generation says what you got — "6 nodes · 2 questions · 1
payment" — and a payment node is called out separately, because it is the one thing in a
generated flow that moves real money.

The same feature exists for [automations](/automations/templates-and-ai).