Deploy
Transfer between AI agents
Give specialized agents permission to continue the same conversation.
An AI handoff changes the agent instructions handling an existing session. For example, reception can pass an appointment request to a scheduling specialist. The caller stays in the conversation; no staff phone line is dialed.
Configure the target
- Create the specialist agent in the same workspace. Keep its task narrow and connect only the tools it needs.
- Publish the target agent. A draft alone cannot serve as a published handoff target.
- Open the source agent’s Agent handoffs & subagents section and allow the target.
- If using a conversation flow, add an Agent Transfer node and choose the allowed agent. For a subagent step, configure the approved subagent and its task.
- Tell the source agent what triggers the handoff and how to introduce it. Publish the source changes.
What carries over
The runtime preserves conversation context, durable memory and completed-action receipts so the target can continue without asking the caller to repeat everything or redoing a successful booking. Target-specific flow variables are reset for the new flow.
In-call handoffs use the active voice model and compatible provider connection for that call. Choose compatible same-workspace agents. A handoff is not a mechanism to change a running call from Mistral to GPT or Gemini.
Example instruction
Verify
Test a caller who provides their name, pet name and preferred time before the handoff. Confirm the target retains the relevant facts, follows its own instructions and uses authorized tools. Test an unavailable/unpublished target and confirm the source explains the fallback. Check that a completed action is not repeated after a handoff.
For a staff member on a telephone, configure human call transfers instead.