Answering machine detection

Detection listens to each answered line and decides whether a person or a machine picked up. People are connected to the agent. Voicemail, phone trees and call screeners are dropped before the agent hears them.

Detection works on the words spoken

When a line is answered, its audio is transcribed word by word as it comes in. Machine phrases such as "leave a message", "press 1" or "please state your name" end the line right away. A short greeting followed by a pause, such as "Hello?" or "Ridgeline Roofing, this is Dana", is a person, and the line is bridged as soon as they stop talking.

The agent hears the person about 1 second after they finish saying hello (median, our network). The speech_ended_to_bridge_ms field on each event gives the time for that call.

Detection runs on every answered line in a dial session and on single calls with amd=true. It is part of the per-minute rate, and lines dropped as machines are never billed.

Results

Every answered line gets exactly one result.

Result	Meaning
`human`	A person answered. The line is bridged to the agent.
`machine_voicemail`	A voicemail greeting, personal or carrier default.
`machine_ivr`	A phone tree or auto attendant, for example "press 1 for sales".
`machine_screener`	A call screening service asking the caller to say their name and reason for calling.
`machine_silent`	The line was answered and nobody spoke within 5 seconds.

Settings

Set detection on the dial session or on a single call. The default is to hang up on machines. Use notify when your app handles machines itself, for example to log the voicemail greeting.

Parameters	Type	Description
`amd`	boolean	Turns detection on. Default true for dial sessions and single calls.
`amd_on_machine`	hangup \| notify	hangup drops the line. notify keeps the line open, sends amd.completed with the result and leaves the next step to your app. Default hangup.

POST/v1/calls

curl https://api.liveleg.dev/v1/calls \
  -H "Authorization: Bearer $LIVELEG_API_KEY" \
  -d to=+13035550112 \
  -d from=+13035550184 \
  -d agent_id=agt_207 \
  -d amd=true \
  -d amd_on_machine=notify

const res = await fetch('https://api.liveleg.dev/v1/calls', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.LIVELEG_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    to: '+13035550112',
    from: '+13035550184',
    agent_id: 'agt_207',
    amd: true,
    amd_on_machine: 'notify',
  }),
});
const call = await res.json();

import os, requests

call = requests.post(
    "https://api.liveleg.dev/v1/calls",
    headers={"Authorization": f"Bearer {os.environ['LIVELEG_API_KEY']}"},
    json={
        "to": "+13035550112",
        "from": "+13035550184",
        "agent_id": "agt_207",
        "amd": True,
        "amd_on_machine": "notify",
    },
).json()

Events

Each answered line sends amd.completed with the result and the action taken. For a person, call.bridged follows with the same speech_ended_to_bridge_ms.

Event

{
  "id": "evt_5Rt09Kd3",
  "type": "amd.completed",
  "created_at": "2026-11-03T15:42:07.390Z",
  "livemode": true,
  "data": {
    "call_id": "cal_8Jq2mT4x",
    "leg_id": "leg_a1",
    "result": "human",
    "speech_ended_to_bridge_ms": 980,
    "action": "bridge",
    "agent_id": "agt_114"
  }
}

With amd_on_machine=notify, machine results include the words heard on the line.

Event

{
  "id": "evt_5Rt0A1x8",
  "type": "amd.completed",
  "created_at": "2026-11-03T15:44:12.006Z",
  "livemode": true,
  "data": {
    "call_id": "cal_Z71nP0qe",
    "leg_id": "leg_b4",
    "result": "machine_voicemail",
    "action": "notify",
    "transcript": "Hi, you've reached Hollis Family Dental. Please leave a message after the tone."
  }
}