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

# Conversation Builder

> Structure your agent's demo into stages with topics, starters, and outcomes.

The Conversation Builder structures the demo into a sequence of **stages**, each with **topics** and **starters** that guide the agent and give visitors easy ways to engage. Open the **Conversation** tab in the Navattic Agent editor to access it.

## Stages

Stages represent phases of the conversation. The Opening stage is always first and maps to your presentation message. Add as many additional stages as your demo needs:

| Stage type    | Default goal                       |
| ------------- | ---------------------------------- |
| Qualification | Confirm prospect's intent and fit  |
| Discovery     | Uncover pain points and priorities |
| Present       | Demonstrate the product in action  |
| Explore       | Give the prospect time to explore  |
| Objections    | Acknowledge and resolve concerns   |
| Close         | None                               |
| Custom        | Define your own goal               |

Each stage has a **goal** field that tells the agent what to accomplish before advancing. Stages marked **Allow exploration** let visitors ask off-topic questions; the agent addresses them and then returns to the stage's topics.

<Note>
  Knowledge resources are available to the agent in all stages except Opening. If a visitor asks a
  question the agent cannot answer by navigating the demo, it draws on the knowledge base to
  respond.
</Note>

To add a stage, click **Add stage** in the Conversation tab. Drag stages to reorder them.

## Topics

Topics tell the agent what to say or ask within a stage. Each topic has a type:

* **Question**: Something the agent asks the visitor (e.g., "How large is your team?"). Questions can be mapped to a [custom property](/tracking/forms/properties) so the answer is saved to the visitor record automatically.
* **Presentation**: A narrated walkthrough of a specific flow from your environment. The agent navigates the flow while explaining it.
* **Suggestion**: A flow the agent can offer to show when relevant, without requiring it.

Each topic also has a mode:

* **Read exactly**: The agent delivers the text you provide verbatim.
* **Use as prompt**: The text serves as context; the agent phrases it naturally.

To add a topic, open a stage and click the topic type you want from the **Add topic** panel.

## Starters

Starters are clickable suggestion chips that appear at the bottom of the chat panel during a demo session. They give visitors quick ways to ask questions or request demos without typing freeform text.

Each stage supports up to five starters. Click **Generate** to let Copilot draft starters based on the stage's topics and flows, or add them manually.

When the demo advances to a new stage, the starters update to reflect that stage's topics.

<Note>
  Starters don't appear until after the agent finishes its intro greeting. The text input submit
  button is also disabled during the intro, so visitors can't send a message before the agent is
  ready to respond. Both become active automatically once the intro greeting ends.
</Note>

## Outcomes

Outcomes let you configure what the agent presents to a visitor when the right moment in the conversation arrives: a button to a landing page, a calendar to book a call, or a closing message. Each outcome fires when its conditions are met.

To add an outcome, open the **Conversation** tab in the Navattic Agent editor and click **Add outcome**.

**Outcome types**

| Type     | What the visitor sees                                      |
| -------- | ---------------------------------------------------------- |
| CTA      | A card with a title, description, and a link button        |
| Calendar | A scheduling widget to book a meeting directly in the demo |
| Message  | A card with a title and description, no button             |

The button URL on a CTA outcome and the scheduler URL on a Calendar outcome accept personalization variables. See [Personalizing links](/agents/build/settings#personalizing-links).

**Conditions**

Each outcome has one or more conditions that must all be true for it to fire. You can mix condition types:

* **AI-evaluated**: Write a natural-language prompt describing what the conversation must show (e.g., *"The visitor is at an enterprise company and has asked about security compliance"*). The agent evaluates this against the full conversation.
* **Conversation length**: Fires after the visitor has sent a set number of messages.
* **Flow started**: Fires after the agent has navigated a set number of flows.
* **Step viewed**: Fires after a set number of steps have been shown.

**Eligible stages**

You can restrict an outcome to specific stages so it only fires when the conversation is in the right phase. Leave all stages selected to allow the outcome to fire at any point.

<Note>
  Only one outcome fires per session. Once an outcome fires, subsequent conditions are not
  evaluated.
</Note>