Steps are grouped by what they do. Some are available only where they make sense — pipeline steps on a pipeline workflow, for instance.
Let Kai answer — hands the conversation to Kai, which answers from your content. Usually the last step in a flow.
Give Kai a task — asks for a specific job rather than an open conversation.
Pass to agent — hands the live conversation to a custom agent. The agent has to be published.
AI ticket tagging and AI ticket assignment — classify and route without a rule you have to maintain.
Write a message — send text from the workflow.
Show expected reply time — sets expectations using your configured reply time.
Collect customer reply — waits for them to write something.
Collect data — asks a question and stores the answer in an attribute.
Schedule a meeting — offers a booking rather than a conversation.
Rate conversation — asks for a satisfaction rating.
Add a tag, change the priority, assign it, change its status and type, set a data attribute, add an internal note, or open the customer's latest open ticket instead of starting a new one. You can also create, snooze, close, or archive a ticket outright.
Custom API action — calls your own endpoint and can use what comes back.
Dynamic buttons — build the options from an API response, so the choices reflect live data.
Run custom action — triggers something in your own app through the SDK.
Pass to workflow and Delay — hand over to another flow, or wait.
Open a link, start a classic form, branch on a condition, or offer buttons for the customer to choose from.
On pipeline workflows, there are steps to move an entry to another stage, set a field on it, add it to a second pipeline, create a ticket from it, email the contact, or notify a teammate.
