Conversational Flows

Conversational Flows let you script a structured, step-by-step conversation with a node-graph editor instead of relying on a free-form prompt alone.

By default, an agent answers openly using its system prompt, business info, knowledge base, and tools. A flow is for when you need the conversation to follow a defined path — collect specific fields in order, branch on the caller's answers, and only end or transfer at the right moment.

When to use a flow

Reach for a Conversational Flow when the conversation has a required structure or a fixed outcome. Stick with a free-form prompt when the agent just needs to answer questions naturally.

Use a flow when:

• You need to collect specific fields in a specific order (for example, an appointment-booking flow: service → date → time → contact details → confirmation).
• The path should change based on what the caller says (branching logic).
• Certain steps must complete before the call can end or transfer.
• You want tight, predictable control over each turn.

Use a free-form prompt when:

• The agent mostly answers questions grounded in your knowledge base and business info.
• The conversation is open-ended and doesn't follow a set sequence.
• You don't need to guarantee a particular ordering of steps.

The flow editor

The Conversational Flow tab is a node-graph editor. You lay out the conversation as a set of nodes connected by transitions, and you can define variables that carry information across the conversation.

Nodes

Each node represents a step in the conversation. You define nodes for the parts of the conversation you want to control, connect them with transitions, and the flow moves from one node to the next as the conversation progresses.

Transitions

Transitions are the connections between nodes — they define where the conversation can go next. Use them to lay out both a straight-line path (step after step) and branches, so the conversation can take different routes depending on the caller's responses.

Variables

Variables let the flow capture and carry information across the conversation — for example, the details you collect at each step. Define the variables your flow needs so later nodes can use what earlier nodes gathered.

Transfer and End nodes

Two special node types control how the conversation wraps up:

• Transfer — hands the call off to a human.
• End — ends the conversation.

Place these at the points where the flow should conclude or escalate, so the agent only transfers or ends once the required steps are done.

Block interruptions

Each node has a per-node block interruptions setting. Normally a caller can barge in — talk over the agent — but on a node where you don't want the caller to interrupt (for example, while the agent reads back a confirmation or important disclosure), enable block interruptions so the agent finishes that step first.

How the flow advances

The flow is a state machine that advances one step per turn. On each turn, the flow evaluates the current node and its transitions, then moves to the appropriate next node based on the conversation so far. This step-by-step progression is what keeps a structured conversation — like an appointment-booking flow — on track from one node to the next.

Example: an appointment-booking flow

A booking flow is a common use case. You might lay out nodes that:

1. Greet the caller and ask what they'd like to book.
2. Collect the service, then the date, then the time — capturing each into a variable.
3. Collect contact details.
4. Read back the details on a node with block interruptions enabled, so the confirmation isn't cut off.
5. Branch to a Transfer node if the caller needs a human, or to an End node once the booking is confirmed.

Because the flow advances one step per turn, each field is gathered in order and the call only ends or transfers at the node you designated.

Next steps

• Agents — where the Conversational Flow tab lives, alongside identity, business info, and the system prompt.
• Tools — add transfer-to-human, end-call, and other callable tools your flow can use.
• Agent Teams — when you need multiple agents that hand off to each other rather than a single scripted flow.
• Knowledge Base — ground the agent's answers in your own content.