Flow Editor

Overview

The Flow Editor serves as the main entry point to building and visualizing your agent. An agent's flow defines its behavior while it executes. A Flow is a graph of interconnected Behaviors and Actions that execute against the agent's Context.

Preparing to add a Prompt AI Behavior to the flow with no other panels open

Behaviors

Behaviors (sometimes called nodes) represent major steps in your agent's flow execution and are rendered in the top-level flow graph. Double-clicking a behavior results in its configuration flyout appearing from the left:

Common Behaviors

The following behavior types are common across all types of agents and represent the fundamental building blocks of AI automation:

Behavior	Description
Execute Logic	Author visual rules & logic and control the flow execution
Search	Search AI resources
Prompt AI	Prompt AI based on the context and translate the LLM completion into actions
Call API	Call an API to retrieve or modify data in external systems
Call Function	Call a function to derive actions from the context

See the full documentation of core behavior types

Customer Agent Behaviors

Customer Agents have additional behaviors:

Behavior	Description
Assigned	Special behavior that executes when a conversation is first assigned to the agent
Send Message	Send a message to the customer and optionally wait for a response
Set Timer	Set a timer that will fire if no customer response is received
Route	Route the conversation to another agent or a live agent queue
Close	Close the conversation

See the full documentation

Actions

Behaviors execute Actions. Actions are therefore lower-level and not visible in the top-level flow graph. They are, however, very important to the agent's execution as they are used to move between behaviors and trigger important side effects like setting fields and sending messages.

The Flow Editor is used to visually configure actions, but some actions can also be generated programmatically by Functions and API Calls.

Preparing to add another action in an Execute Logic behavior

Common Actions

The following actions are used to build any type of agent:

Core Action	Description
Change Location	Change location to another behavior in the flow
Set a Field	Set a field to a value
Unset a Field	Unset a field
Add to a Field	Increment a counter
Set a Buffer	Store state in an opaque buffer

Change Location is particularly important because it controls the flow. Change location actions are visualized as edges in the flow graph. See the full reference for more information on core actions.

Customer Agent Actions

Customer agents utilize additional actions to interact directly with customers. Here are some of the most important customer agent actions:

Customer Agent Action	Description
Send a Message	Send a message including full rich-messaging support
Set a Catalog Message	Send a message from the Message Catalog
Start Typing Indicator	Display a typing indicator to customers on supported platforms

See the full reference for more information.

Context

An agent's Context is a less visible but nonetheless critical aspect of flow execution. The context contains all the data (i.e. state) that the agent executes against. Behaviors and actions modify the context as they execute.

Any decisions or actions configured in the Flow Editor are evaluated against the context. Similarly, the context is a primary input to functions in the Function Editor, such as prompt building functions that translate the context into prompts to send to an LLM.

The shape of the context varies based on the type of agent. But all agent contexts include the following data:

Search results - Accumulated by Search behaviors during the current flow execution
API results - Accumulated by Call API behaviors during the current flow execution
Session field state - The state of session fields that can be checked and set by visual and programmatic actions
Buffer state - Opaque buffer state used by Functions

Additionally, the context will also include the primary data model that the agent operates against. In the case of customer agents this includes the full conversation record.

More information:

Customer Assistant Context Overview
CustomerAssistantContext Reference Documentation