Flow Step Types: Elvin AI

Elvin AI steps let a flow send quick messages, open the Elvin widget, launch Discoveries, or use AI to classify and transform text collected earlier in the flow. Use these steps when the flow needs to offer contextual help, hand the user over to Elvin, or react to open-ended input that cannot be handled with fixed conditions alone.

Toast Notification

Displays a short notification in the Elvin widget area.

Use this to send a quick message to the user at a specific point in the flow, such as a confirmation, a heads-up, or a next step, without opening the full widget.

Type your message directly into the text field. What you write is what the user sees, with no processing or reformatting applied.

Urgency

Urgency controls how the notification is presented.

Normal: a standard notification appearance.

High: a more prominent visual treatment to draw the user's attention.

Trigger Elvin Chat

Opens the chat tab of the Elvin widget so the user can start a conversation with the Elvin copilot.

Use this when a flow reaches a point where the user might need open-ended help that a guided flow cannot anticipate.

Instructions for Elvin

Instructions for Elvin: an optional field where you define the opening message Elvin greets the user with.

This lets you set context for the conversation instead of dropping the user into a blank chat. For example, you could pre-frame the topic based on where the user is in the flow.

Trigger Discovery

Launches an existing Discovery from within the flow.

Discoveries are a separate Product Fruits feature. If you are not familiar with them, read about Discoveries first.

Select discovery

Select discovery: choose from your already-published discoveries.

Launch mode

Launch mode controls how the discovery appears.

Toast notification: appears as a notification in the Elvin widget area, which the user can then choose to open.

Fully expanded: opens immediately in full.

Expand Elvin Widget

Opens the Elvin widget and lands on a specific tab.

Use this when you want to direct the user to a particular part of the widget, such as the knowledge base, chat, or any other configured tab, without making them find it themselves.

Tab selector

The tab selector only shows tabs that are currently configured for your Elvin widget.

If a tab you expect to see is not listed, check your Elvin widget settings.

AI Classifier

Reads a piece of text from earlier in the flow and routes the user down different branches based on what the content means.

This is useful when the flow needs to react to something open-ended, such as a user's typed response or a value returned from an API Call, where you cannot predict the exact input in advance.

Input to classify

Input to classify: the text the AI will evaluate.

Select either a previous step's output or an enrollment variable.

Branches

Branches: define the categories you want to classify into.

Each branch is described in plain language, for example User is frustrated or User is asking about billing.

The AI assigns the input to whichever branch it best fits.

Fallback

Fallback: always present regardless of how many branches you define.

The flow follows the fallback path if the input does not clearly match any branch. Always connect the fallback to something. Leaving it unhandled will stall the flow for inputs that do not fit your categories.

AI Transform

Takes a piece of text from earlier in the flow, rewrites or restructures it according to your instructions, and saves the result as a variable for use in later steps.

Use this to summarize a long input, reformat a value, extract a specific detail in natural language, or prepare content for display.

Input to transform

Input to transform: the text to process.

Select either a previous step's output or an enrollment variable.

Output variable

Output variable: give the result a name.

The transformed value is saved as {{ flow.vars.<name> }} and available in all subsequent steps.

The output can be plain text or JSON, depending on what your instruction produces.

Describe the transformation

Describe the transformation: write a plain-language instruction for what the AI should do.

For example: Summarize this in one sentence, Extract the company name, or Reformat this as a bulleted list.

Error branch

Error branch: when toggled on, the step splits into a success path and an error path.

Use this if the transformation is critical to the flow and you need control over what happens if the AI fails to produce a usable output.