Listens to the greeting
Takes a copy of the audio the moment they pick up, without touching the call.
NeerBat analyzes your outbound call audio in real time and automatically detects whether a voicemail, a human, or a ringback tone answered — without relying on tones or silences.
Takes a copy of the audio the moment they pick up, without touching the call.
Breaks down the acoustic fingerprint and compares it against thousands of real greetings.
Hangs up the voicemail or hands over the person, before the first "hello".
By the time the agent hears "leave your message after the tone", 20 seconds are gone. NeerBat listens to those first seconds for them and decides before passing the call.
Classic detectors measure silences and greeting length. A voicemail starting with "Hello?" fools them, and so does a person who answers quietly. They fail exactly where it matters.
It analyzes the timbre of the audio, not its rhythm. A recording compressed by the phone network sounds different from a live voice, even saying the same words. That cannot be faked.
Agents stop spending their shift talking to machines. Every voicemail NeerBat cuts is one more real contact attempt in the same working time.
One script installs the agent on any Linux server. Your dialplan and your dialer stay untouched: the agent connects over AMI and observes. All you need is AMI access to Asterisk — it can run on the same server or a separate one. Generate your API key from the dashboard, add it to /etc/buzon/monitor.conf and you are done.
When the other side picks up, the agent takes a copy of the audio — it does not touch the call, it just listens in parallel. The first seconds are enough: nobody needs a full greeting to know it is a recording.
The audio travels to the engine, which breaks it into its acoustic fingerprint (MFCC) and compares it against thousands of already classified real greetings. It returns an answer with a confidence percentage, not a blind yes or no.
If it is voicemail, the call hangs up by itself and the number is flagged to retry at another time. If there is a person, it reaches the agent while the greeting is still going. All of this happens before the alert even shows on their screen.
A person answered. The call goes to the agent immediately.
It is a voicemail or answering machine. It hangs up and the number is retried at another time.
Still ringing: nobody has answered yet. The agent keeps listening.
NeerBat does not dial or manage customers. It is a piece that adds to the dialer you already have — Asterisk or Vicidial — and does one thing: tell you who answered. Your operation keeps working the same, with less wasted time.
Built for production from day one, with security and monitoring included.
Classifier trained on thousands of real recordings. It tells voicemail from humans with high accuracy, without relying on silences or DTMF tones.
The engine processes audio in parallel with independent workers. Detection arrives before the voicemail finishes talking.
Each Asterisk agent authenticates with its own key. No shared global keys. Instant revocation from the dashboard.
See detections in real time, call history, credits used and agent logs from one central web dashboard.
Agent written in pure C. No Node, Python or external dependencies on the telephony server. Works with Vicidial and any dialplan.
No plans or monthly fees. Top up when you need it and pay only for each analyzed call attempt. Credits never expire.
The service runs on prepaid balance, not plans. Top up whenever you want and use the credits at your own pace. No monthly fees, no contracts, no agent limits.
Sign up on the dashboard and get your API key. No contract, no monthly fee.
Top up any amount from this same site. Credits never expire and are deducted per analyzed attempt.
We walk you through setting up the agent and connecting it to your Asterisk. You do not have to figure it out alone.
Each analyzed call is deducted from your balance. You see usage live and we warn you before it runs out.
Every classification is available through the API so you can write it into your CRM, your dialer, or wherever you track your work. NeerBat does not replace your system: it feeds it.
{
"id": "1746532991.4823",
"numero": "56912345678",
"resultado": "MACHINE", // MACHINE | HUMAN | RINGBACK
"confianza": 0.97,
"duracion_analisis_ms": 1480,
"accion": "reintentar_otra_hora",
"fecha": "2026-08-28T12:03:13-04:00"
}
You can also receive them via webhook: every time a call is classified, we notify the endpoint you give us. That way your CRM finds out instantly, without having to ask.
Tell us how many calls you make per month and we will tell you how much balance you need. We help you get the agent running on your servers.