Skip to main content
The Agent Breez node defines an independent conversational agent with its own persona, instructions, and behavior.
An agent can route into multiple Agent Tools, allowing it to take autonomous actions when needed. This node has one primary output connector:
  • Flows — Accepts multiple Agent Tools.
Depending on configuration, the node may also display additional connectors for Hear, Think, and Speak.

​
Main Settings

​
Inherit Persona (Ears, Brain, and Voice)

Determines whether this agent inherits its Speech-to-Text (Ears), LLM (Brain), and Text-to-Speech (Voice) settings from the previous agent or Flow Start node.
  • Enabled - The agent hides the Hear/Think/Speak connectors and uses inherited models.
  • Disabled - The agent exposes Hear, Think, and Speak connectors (identical to Flow Start), allowing you to assign unique models for this agent.

​
Inherit Chat History

Controls whether this agent receives the previous conversation history when activated. If enabled, the following setting appears:

​
Seamless Continuation Mode (BETA)

When used with Inherit Chat History, the agent behaves as a continuation of the previous agent, treating all past responses as its own.
When disabled, the agent is aware that it is taking over mid-conversation. This is useful when splitting responsibilities across multiple agents while maintaining continuity.

​
Agent Instructions (required)

Define the agent’s tone, behavior, purpose, constraints, and personality.
These instructions guide how the agent should think and respond throughout the conversation. You may reference workflow variables using @variableName.

​
Initial Agent Action

Determines what the agent does immediately when activated.

​
None (Wait for User)

The agent does nothing until the user speaks.
No additional settings appear.

​
Speak Exact Text

The agent speaks the provided text exactly as written. Additional fields:
  • Allow Interruptions - Allow users to interrupt the agent’s initial action.
  • Save Initial Response to Chat - Stores the agent’s initial response in the conversation history (not the instruction itself).
  • What to Say (required) – The exact sentence the agent will speak.

​
Follow Instructions (with chat history)

The agent generates an initial response using your instructions with full chat history available. Additional fields:
  • Allow Interruptions - Allow users to interrupt the agent’s initial action.
  • Save Initial Response to Chat - Stores the agent’s initial response in the conversation history (not the instruction itself).
  • Allow Agent To Execute Tool Call Immediately On Start
    ⚠️ When enabled, the agent may execute tools (e.g., API calls, DB lookups) as part of its first action. Ensure your instructions clearly specify when tools should be used. If disabled, the agent will only speak without executing any tools.
  • Initial Action Instructions (required) – Guidance on what the agent should say or do when starting.

​
Follow Instructions (fresh start)

Same as above, but the agent receives no chat history when crafting its initial message. Additional fields match those for Follow Instructions (with chat history).

​
Act Autonomously

The agent independently determines its first action — speaking, thinking, or performing a tool call — based on its instructions. Additional fields:
  • Allow Interruptions - Allow users to interrupt the agent’s initial action.
  • Save Initial Response to Chat - Stores the agent’s initial response in the conversation history (not the instruction itself).
  • Allow Agent To Execute Tool Call Immediately On Start
    ⚠️ When enabled, the agent may execute tools (e.g., API calls, DB lookups) as part of its first action. Ensure your instructions clearly specify when tools should be used. If disabled, the agent will only speak without executing any tools.

​
Endpointing Settings

Controls how the system detects when the user has finished speaking.

​
Use Session Defaults

Inherits endpointing timing settings from Flow Start.

​
Customize for this agent

Overrides defaults and displays:
  • Fast Turn Delay (semantic/provider)
  • Slow Turn Delay (VAD only)
  • Max Semantic Delay
These settings function identically to the Endpointing Settings described in the Start node documentation.

​
Show Advanced Timing

Displays fine-grained controls for VAD behavior, sensitivity, and interruption rules, including:
  • VAD Activation Threshold
  • VAD Min Speech Duration
  • Prefix Padding
  • Allow Interruptions
  • Min Interruption Duration
  • Min Interruption Words
These settings function identically to the Advanced Timing section in the Start node documentation and behave the same way.