Talkif Docs
Flow Builder

Agent Nodes

Configure Main Agent and Agent nodes to handle conversation segments and transitions.

Agent nodes are the conversation handlers in your flow. They define what the AI says, how it behaves, and when it hands off to another agent.

Main Agent

Every flow has exactly one Main Agent. It's the entry point — the first node that handles the call when it connects.

The Main Agent is marked with a crown badge on the canvas and cannot be deleted.

Configuration

FieldDescription
System prompt (role_message)Defines the agent's personality, role, and behavioral rules. This is the "who you are" instruction.
Task prompt (task_message)Defines what the agent should accomplish in the conversation. This is the "what you do" instruction.
TransitionsConditions under which the conversation transfers to another Agent node.

Example system prompt

You are a friendly customer support agent for Acme Corp.
Speak clearly and concisely. Never make promises about
refunds without checking the customer's order status first.

Example task prompt

Greet the caller, identify their issue, and route them
to the appropriate department. If you can resolve the
issue directly, do so.

Agent nodes

Agent nodes handle specific conversation topics or tasks. They have the same configuration as the Main Agent — system prompt, task prompt, and transitions.

Use additional Agent nodes when your flow needs specialized handling for different topics. For example:

  • A Billing Agent that handles payment and invoice questions
  • A Technical Support Agent that troubleshoots product issues
  • A Scheduling Agent that books appointments

Transitions

Transitions define when and how the conversation moves from one agent to another.

Each transition has two properties:

PropertyDescription
TargetWhich Agent node to transfer the conversation to
ConditionA natural language description of when this transfer should happen

Example transitions

ConditionTarget
"Customer asks about billing or payments"Billing Agent
"Customer needs technical help with their device"Technical Support Agent
"Customer wants to schedule an appointment"Scheduling Agent

The LLM evaluates transition conditions during the conversation and triggers the transfer when a condition is met. Agents can transition to any other Agent node, including back to the Main Agent.

Context strategies

When a conversation transitions between agents, you control how much conversation history carries over.

StrategyBehavior
APPENDCarry the full conversation history to the next agent. The new agent sees everything that was said.
RESETStart with a clean slate. The new agent has no knowledge of the previous conversation.
RESET_WITH_SUMMARYSummarize the previous conversation and pass the summary to the next agent. Balances context preservation with token efficiency.

When to use each strategy

  • APPEND — When the next agent needs full context (e.g., escalation to a supervisor)
  • RESET — When topics are completely independent (e.g., switching from sales to a post-call survey)
  • RESET_WITH_SUMMARY — When the next agent needs background but not every detail (e.g., transferring to a specialist after initial triage)

On this page