Overview
Access the Transfers tab in the agent settings. Each rule has a trigger condition and a destination: another AI agent or a human team member.
How rules work
When the agent processes a message, active transfer rules are injected into the prompt context. The model analyzes the conversation and, if it identifies that a condition has been met, executes the transfer automatically. The transfer is intelligent: the model does not look for exact words — it understands intent. A rule like “when the customer shows frustration with the service” is triggered both by “I’m angry” and “this is outrageous” or “I want to cancel everything.”Destination types
Another AI agent
The conversation is passed to a specialized agent. Useful for creating service flows with agents specialized by topic (support, sales, finance).
Human attendant
The conversation is passed to a team member in the inbox. The agent stops responding and the attendant takes control.
Create a transfer rule
Write the condition
The Instruction field is where you describe in natural language when the transfer should happen.Examples of effective instructions:
Select the destination type
Choose between AI Agent or Human.
- AI Agent: select which workspace agent should receive the conversation.
- Human: select which team member should receive the conversation in the inbox.
Enable or disable
The Active rule toggle controls whether the rule is being evaluated. Inactive rules are saved but not injected into the agent’s context.
Manage existing rules
In the rules list, each card displays:- The destination type (“AI” or “Human” badge)
- Active/inactive status
- The instruction excerpt (up to 100 characters)
- The name of the destination agent or attendant
| Action | How to do it |
|---|---|
| Enable/disable | Toggle on the card, without opening the form |
| Edit | Click the pencil icon |
| Delete | Click the trash icon and confirm |
Intelligent handoff
The transfer to a human does not just pass control — it also:- Notifies the attendant in the inbox with the conversation context up to that point.
- Preserves the history — the attendant sees all messages exchanged with the agent.
- Stops the agent — the agent no longer responds while the human is in control.
- Allows returning to the agent — the attendant can close the service or return it to automation.
Limits and restrictions
Each agent supports a maximum of 10 transfer rules. When the limit is reached, the new rule button is automatically disabled.
- The instruction field is required — a rule without an instruction cannot be saved.
- When selecting the AI Agent type, you must select a destination agent different from the current agent.
- When selecting the Human type, you must select at least one team user with workspace access.
- Deleted rules cannot be recovered — but you can recreate a rule with the same content.
Best practices
- Create a generic fallback rule to ensure conversations the agent cannot resolve always reach a human.
- Order rules from most specific to most generic — the model evaluates all rules and uses judgment to choose which to apply.
- Temporarily disable rules instead of deleting them — useful for tests or periods without available attendants.
- Test rules in the Playground by simulating the situations described in the conditions.
Common use cases
Escalation by topic
Transfer to specialized agents: a technical support agent, a sales agent, and a finance agent — each with their own rules and knowledge base.
Explicit request
Always honor when the customer explicitly asks to speak with a human. Keep this rule always active.
Attempt limit
After N exchanges without resolution, transfer to advanced support. Helps avoid frustrating loops.
Business hours
Combine with time logic in the prompt: during business hours, transfer to human; outside hours, log the contact and notify about follow-up.
Next steps
Test the agent
Simulate conversations in the Playground to validate that transfer rules fire correctly.
Agent overview
Review all agent configuration blocks to ensure it is ready for publication.