VICIdial AI AMD setup in 10 minutes, with no new servers

Why add AI AMD to VICIdial

VICIdial's built-in AMD is Asterisk's AMD() application. It decides from silence and word timing: how long before the first word, how many words, how long the pause after. That works on long, chatty greetings. It fails on the calls that matter most: a person who says "Hello?" and waits, a short personal greeting, a call screening assistant or a recorded "Hello?" robot. Tuning amd.conf moves those errors around but does not remove them. If that sounds familiar, VICIdial AMD not working covers the usual causes.

The usual workaround is to turn AMD off and let agents listen to voicemail all day. That keeps the live people but burns agent time on greetings, carrier messages and fax tones. Neither choice is good, and neither is necessary.

AI answering machine detection listens to the sound of the answer instead. nomoVM plugs into the same place in VICIdial where AMD() sits today, and sets the same AMDSTATUS variable that VICIdial already reads. Everything after the decision stays the same: VD_amd.agi, AA statuses, your voicemail drop settings, your agents and your carrier.

Before you start

Check these on each dialer server. None of them should need new software on a current ViciBox install:

• Root shell access to the dialer server
• Asterisk 13 or newer: asterisk -rx "core show version"
• Python 3.6 or newer: python3 --version (ViciBox 11 and 12 already have it)
• Outbound HTTPS on port 443 from the dialer to amd.nomovm.com
• A campaign with AMD already on, using Routing Extension 8369

There is no additional infrastructure required. You do not add servers, buy hardware or change carriers. nomoVM listens to the audio your dialer already has and streams the first seconds of each answered call over an encrypted WebSocket.

The same steps work on GoAutoDial, scratch VICIdial installs and multi-server clusters, as long as Asterisk is version 13 or newer. You do not need to upgrade VICIdial first.

Here is roughly how the 10 minutes break down:

Step-by-step: add AI AMD to VICIdial

1. Create a key

   Sign up for nomoVM. The Free plan covers 30,000 calls a month and needs no card. Open WS keys and create one key per dialer server, named after the server, so the dashboard shows each server's calls separately. Copy the key when it is shown. It starts with amd_live_.

2. Install the script

   Download nomovm.agi.py from the WS keys page of your dashboard and copy it to the Asterisk AGI folder on the dialer:

   scp nomovm.agi.py root@dialer1:/var/lib/asterisk/agi-bin/
   ssh root@dialer1
   cd /var/lib/asterisk/agi-bin
   chmod 755 nomovm.agi.py
   python3 --version

   The script uses only Python's standard library, so there is nothing to pip install.

3. Test the connection

   Before any live call touches it, run the test mode with your key:

   python3 /var/lib/asterisk/agi-bin/nomovm.agi.py --test wss://amd.nomovm.com/ YOUR_KEY

   It streams a few seconds of silence, like a real call where nobody speaks, and should print AMDSTATUS=MACHINE AMDCAUSE=SILENCE in about 2 seconds. If it prints FALLBACK instead, jump to troubleshooting.

4. Add extension 8370

   VICIdial sends answered AMD calls to extension 8369. Leave it alone and add a copy, 8370, in /etc/asterisk/extensions.conf, in the same context as your 8369 (usually [default]):

   exten => 8370,1,AGI(agi://127.0.0.1:4577/call_log)
   exten => 8370,n,EAGI(nomovm.agi.py,wss://amd.nomovm.com/,YOUR_KEY,4500)
   exten => 8370,n,GotoIf($["${AMDSTATUS}" = "FALLBACK"]?native)
   exten => 8370,n(vdamd),AGI(VD_amd.agi,${EXTEN})
   exten => 8370,n,AGI(agi-VDAD_ALL_outbound.agi,NORMAL-----LB-----${CONNECTEDLINE(name)})
   exten => 8370,n,Hangup()
   exten => 8370,n(native),AMD(2000,2000,1000,5000,120,50,4,256)
   exten => 8370,n,Goto(vdamd)

   Copy the VD_amd.agi and agi-VDAD_ALL_outbound.agi lines from your own 8369 rather than from this page. Some installs pass extra options, and those should stay exactly as they are.

   What each line does: the first logs the call for VICIdial. The EAGI line runs nomoVM with a 4500 ms client timeout and sets AMDSTATUS and AMDCAUSE. If the result is FALLBACK, the call jumps to the native label and runs Asterisk's AMD() instead. Either way it ends up at VD_amd.agi, which routes the call just as it does today. Reload and check:

   asterisk -rx "dialplan reload"
   asterisk -rx "dialplan show 8370@default"

5. Switch one campaign

   In the VICIdial admin, open Campaigns and pick one campaign to start with. Change Routing Extension from 8369 to 8370. Leave AMD send to VM exten and your voicemail message settings as they are. Submit.

6. Watch the first calls

   Open the nomoVM dashboard. Calls appear within seconds, each with its verdict, its cause and the time from answer to verdict. Most calls are decided about 2 seconds after answer; only unclear calls listen up to 3.2 seconds.

Verify it is working

Give the campaign an hour or a day of normal traffic, then check three things:

• Dashboard: verdicts and causes look sensible. Listen to a few recordings on the Call logs page to hear what nomoVM heard. Recording is optional; you can turn it off or keep recordings from 1 to 90 days.
• VICIdial: compare the campaign's AA count and agent-connected count with the days before. You should see more "Hello?" calls reaching agents and fewer greetings.
• Agents: ask them. They notice quickly when fewer people are dropped and fewer voicemails come through. Calls screened by Google or iPhone assistants show up in the dashboard with the cause SCREENING.

For context, on 400 real outbound calls checked by a person, none of them used to build nomoVM, it sent 94.9% of live people to agents, and 88% of calls sent to agents were live people. The accuracy page explains how that was measured. When you are happy with the results, switch the remaining campaigns.

Roll back in one step

Set the campaign's Routing Extension back to 8369 and submit. That is the whole rollback. Extension 8370 can stay in the dialplan doing nothing until you want it again. Nothing about your carrier, your agents, your lists or VD_amd.agi was changed.

Troubleshooting

The test prints FALLBACK AUTH

The key is wrong or was copied with a missing character. Create a new key in the dashboard and run the test again. Check the same key is in the EAGI line of extension 8370.

The test prints FALLBACK NOSERVER

The dialer cannot reach amd.nomovm.com. Check DNS on the server and that outbound connections on port 443 are allowed by the server firewall and any network firewall in front of it.

FALLBACK QUOTA or SUSPENDED in the Asterisk log

Check your plan and account status in the dashboard. Calls keep working through Asterisk's AMD() while you sort it out.

No calls appear in the dashboard

The campaign is still on 8369, the dialplan reload did not take, or the extension is in a different context from 8369. Run dialplan show 8370@default and check the campaign's Routing Extension.

"python3: not found"

Install Python 3 from your distribution's packages. On openSUSE-based ViciBox: zypper install python3.

Agents notice a delay before the customer speaks

Verdicts take about 2 seconds after answer, similar to a typical AMD setting. If calls in the dashboard show longer times, check the network path from the dialer. For more on silence after answer, see dead air calls.

Multi-server clusters

Repeat steps 2 to 4 on every dialer server, each with its own key. A campaign can place calls on any dialer, so 8370 must exist everywhere before you switch the campaign.

What it costs and where to go next

nomoVM bills one call per verdict on an answered call. There are no per-server or per-agent fees, and servers are unlimited. The Free plan covers 30,000 calls a month as a hard limit and is never billed. Starter is $99 a month for 1,000,000 calls, Growth $199 for 3,000,000 and Scale $499 for 10,000,000. See pricing for details, and VICIdial AMD for how the integration works.

Ready? Create your free account and keep the VICIdial AMD setup guide open while you work.