A transfer hands a live caller to a person. The agent decides to do it, and from that moment the conversation stops being ours: the caller is talking to your colleague and the agent is gone. This page is the operator’s view — what the caller hears, what the person answering sees, what it costs and how it reads back afterwards. To build one, see the transfer operation.

​
The two modes

The builder chooses cold or warm. Everything below is what each one feels like from the outside. While the transfer is dialling, the caller’s microphone and speaker are both switched off. They hear hold music and nobody is listening to them — which is deliberate, so a “hello?” into a ringing line is not transcribed and not billed.

​
What the person answering sees

They see your workspace’s number, not the caller’s. Caller ID on an outgoing call is a property of your carrier connection, not of the individual call, so there is no way to present the caller’s number to your colleague instead. This is the same on all four carriers and cannot be changed per transfer.
Give your colleagues something to go on inside the call instead: a warm transfer, where the agent says who is calling before handing over, is the answer to “who is this?” that caller ID cannot give here.

​
When it does not connect

Every failure reaches the caller as one of a small set of plain lines. A carrier status code is never spoken, and a fault of ours is never dressed up as a fact about the person being called. A declined warm transfer never leaks the decliner’s words. Whatever your colleague said about why goes onto the call’s timeline and nowhere else; the caller hears the fixed line above. After a failure the tool’s on_failure decides what happens next: continue brings the caller off hold and back to the agent, which then has the wording above to work with, or end_call ends the call.

​
What it costs

A transferred call that nobody hangs up is held open until the platform’s three-hour cap on any call. Talqing bills nothing for that time; your carrier does. It ends as soon as either human hangs up.

​
Timing

The ring window and the briefing share the 120-second budget, because the caller is on hold for both. The default 30 + 90 lands exactly on it. Raising ringing_timeout toward its 120-second maximum eats the margin and can drop the held caller entirely — a worse outcome than coming off hold early. If you need a long ring, shorten what the agent says in the briefing.

​
How a transfer reads back

Three places record it, and they answer different questions. The briefing is deliberately not in the transcript. It is the agent talking about the caller to a third party, not to them — putting it in the transcript would replay it into the caller’s next conversation as if they had said it. It is the only record of what was said about a customer to someone else, so treat it as a compliance record rather than telemetry. The recording covers the call up to the handover. What the two humans say to each other afterwards is not recorded — the agent, and with it the recorder, has left. A transferred call is still analysed like any other, so call analysis can tell you why calls escalate.

​
What transfers are not

  • Not agent to agent. Moving between agents is a handoff, a different mechanism with different context rules.
  • Not a queue or an IVR. A transfer dials one number. Where that number rings is your phone system’s business.
  • Not templated. The destination is a literal phone number frozen when the tool is published, and it is refused at validation if it is anything else. That is anti-fraud design: a destination a model could compose is a destination a caller could talk it into.
  • Not paused or resumed. There is no hold-and-return button; the caller is on hold only while the transfer is being attempted.

​
Next

Build a transfer tool

The operation, its fields, and one tool per destination.

Close reasons

Where transferred sits among every other way a call can end.