Flow block in Connect Customer: Agentic CX
This topic defines the flow block for Agentic CX.
Note
The Agentic CX block is only available in Amazon Connect Customer instances, and only in certain Regions. For the list of Regions, see Agentic CX Designer.
Description
-
The Agentic CX Flow block connects a contact to an Agentic CX application built using Agentic CX Designer. Use this block to hand the conversation to the application and pass data from your flow into it as context variables. The block routes the contact to a branch based on the exit condition that the application returns. An exit condition is a signal that the Agentic CX application returns to indicate the outcome of the conversation.
-
For more information about building Agentic CX applications, see Agentic CX designer.
Prerequisite
Before you can use this block, you must build an Agentic CX application in Agentic CX Designer. The block's Application dropdowns only list applications that already exist in your Agentic CX Designer environment.
Supported channels
The following table lists how this block routes a contact who is using the specified channel.
| Channel | Supported? |
|---|---|
Voice |
Yes |
Chat |
Yes |
Task |
No |
No |
Flow types
You can use this block in the following flow types:
-
Inbound flow
-
Customer Queue flow
-
Customer Hold flow
-
Customer Whisper flow
-
Outbound Whisper flow
-
Transfer to Agent flow
-
Transfer to Queue flow
How to configure this block
When you add an Agentic CX Flow block to your flow, you configure the following properties.
Application
Select the Agentic CX application to handle the contact. Each application defines its own conversation logic, AI models, tools, and knowledge bases.
-
Select application — Choose the workspace, application, and alias from the dropdown. The dropdowns list the workspaces, applications, and aliases configured in Agentic CX Designer.
-
Set application — Enter the IDs of the workspace, application, and alias manually, or set them dynamically using contact attributes.
Speech recognition engine
Select the speech recognition engine that converts customer speech to text.
-
Engine — Choose the speech recognition engine. Set manually or set dynamically using a contact attribute. Default: Amazon Connect agentic voice. Available options:
-
Amazon Connect agentic voice (recommended) — A next-generation speech experience with enhanced automatic speech recognition (ASR), built for natural agentic voice conversations. Recommended for most use cases.
-
Amazon Transcribe — A fully managed automatic speech recognition (ASR) service that converts real-time speech to text.
-
Speech to speech — Processes customer audio and generates responses directly as speech, without an intermediate text transcription step.
-
Audio configuration (voice only)
Audio filler — (Optional) Enable audio filler to play while the AI processes a response. This reduces perceived latency for the customer. When enabled, configure the following settings:
-
Audio type — Select the audio filler sound to play. Set manually or set dynamically using a contact attribute. Available options:
-
Melody - Chipper Chime
-
Melody - Curious Crawl
-
Melody - Rising Ripple
-
Melody - Patient Ping
-
Melody - Pondering Pong
-
Typing - Kinetic Keys
-
Typing - Quiet Qwerty
-
-
Audio delay (ms) — Delay in milliseconds before audio starts. Set manually or set dynamically using a contact attribute. Min: 500, Max: 5000, Default: 2500.
-
Minimum play duration (ms) — Minimum duration in milliseconds for audio playback. Set manually or set dynamically using a contact attribute. Min: 1000, Max: 5000, Default: 3000.
-
Response buffer (ms) — Buffer time (pause) in milliseconds after the audio filler is played and before sending the response to the user. Set manually or set dynamically using a contact attribute. Min: 200, Max: 1000, Default: 500.
Idle chat timeout (chat only)
Configure how long the system waits for customer input before timing out. This setting is available under Optional settings.
-
Idle chat timeout — The maximum time to wait for customer input before following the Idle chat timeout branch. Set manually or set dynamically using a contact attribute.
Context variables
Pass key-value pairs to the Agentic CX application as context variables. These values are available in the Agentic CX application and can be used to personalize responses or provide additional context.
-
Key — Select from preconfigured context variables defined in the Agentic CX application, or enter a custom key using free text.
-
Value — Enter a value manually or set dynamically using a contact attribute.
Note
Maximum of 10 context variables per block.
Configuration tips
-
Keep conversations focused — Design your Agentic CX application with clear exit conditions. This makes sure contacts are routed appropriately when the conversation completes or when a handoff to a live agent is needed.
-
Use interim messages — For voice experiences, configure interim messages in your Agentic CX application to fill silence while tools (such as API calls or knowledge base lookups) are executing.
-
Test thoroughly — Use the Agentic CX debugger to trace conversation paths and verify that exit conditions route contacts to the correct branch.
-
Model selection affects latency — Choose faster models (such as Amazon Nova Micro) for latency-sensitive voice interactions. More powerful models can be used for complex reasoning tasks where slight delays are acceptable.
Branches
This block supports the following branches:
| Branch | Description |
|---|---|
Default |
The agentic conversation completed successfully. The Agentic CX application reached an exit condition indicating the task was fulfilled. |
Error |
An error occurred during the conversation (for example, the Agentic CX application could not be reached or encountered an internal error). Connect this branch to a fallback experience, such as transferring the contact to a queue or playing a message before disconnecting. |
Idle chat timeout |
The customer did not provide input within the configured Idle chat timeout. This branch applies to chat only. |
Escalation |
The Agentic CX application determined that the contact should be escalated to a live agent. Connect this branch to a Transfer to queue or Set working queue block to route the contact to a live agent. |
How it works
When a contact reaches the Agentic CX Flow block, the contact is routed to the Agentic CX application configured in the block. The Agentic CX application manages the conversation. This includes the following:
-
Automatic speech recognition (ASR) for converting customer speech to text.
-
AI-powered conversation orchestration using the models and tools you configured.
-
Text-to-speech (TTS) for delivering AI responses back to the customer.
The conversation continues until one of the following occurs:
-
The Agentic CX application reaches an exit condition (Default branch).
-
An error occurs (Error branch).
-
The customer stops responding on a chat contact (Idle chat timeout branch).
-
The Agentic CX application determines the contact should be escalated to a live agent (Escalation branch).
After the block completes, the contact flow continues along the appropriate branch.
Contact attributes
The Agentic CX Flow block sets $.AgenticCX.ContextVariables, which
contains any context variables returned by the Agentic CX Designer application. You
can reference these in subsequent blocks. To reference an individual variable, use
$.AgenticCX.ContextVariables.<variableName> (for example,
$.AgenticCX.ContextVariables.orderId).