Say

The Say node forces the agent to speak a specific output. It is typically used inside Agent Paths to produce deterministic speech or override the agent’s default behavior.

Main Settings

Speak Mode

Controls how the Say node generates speech.

• Speak Exact Text – Speaks the provided text verbatim with no LLM processing (fastest).
• Follow Instructions (with chat history) – Uses the LLM to generate speech based on your instructions and the existing conversation history.
• Follow Instructions (fresh start) – Uses the LLM to generate speech from your instructions without inheriting chat history.
• Agent Decides (based on chat) – The agent autonomously determines what to say based on chat context and its instructions.

Note: All modes except Speak Exact Text require an LLM response and therefore take longer.

What To Say

Shown only when Speak Mode = Speak Exact Text

The exact sentence the agent will speak.
Supports variable insertion using @variableName.

Speak Instructions

Shown only when Speak Mode = Follow Instructions modes

Instructions that guide what the agent should say.
Supports variable insertion using @variableName.

Save Speech to Chat

Determines whether the agent’s spoken output is saved in the chat transcript.

• On (default) – The message is added to chat history.
• Off – The message is spoken but not saved (useful for announcements that shouldn’t affect future reasoning).