Skip to main content
This guide walks through setting up a Telzino AI agent on any PBX other than NetSapiens, such as Metaswitch, BroadSoft or Avaya. The agent connects to your PBX through SIP registration, like any other extension. The example used throughout is a receptionist agent called Trudy for the organization Acme Organization.
Before you start, complete Prepare Your Telzino Account (Any SIP PBX): it creates your account, adds the reseller and organization, and unlocks the features the agent needs. You only do it once per organization.
If your PBX is NetSapiens, follow Set Up Your First AI Agent on NetSapiens instead.

1. Create your first agent

Agents are built from a template, which pre-fills their instructions for a specific job.
1

Start a new agent

On the Resellers and Organizations page, click the organization name. On the Agents page that opens, click New Agent.
The Agents page, with the New Agent button

The Agents page, with the New Agent button.

2

Pick a template

Pick the template that matches the agent’s job and click Select. This guide uses Receptionist Agent. Other options include Support Agent (CRM), Appointment Scheduler, Message Taker Agent, Blank Slate Agent, OpenTable Reservation Agent, Translator Agent and Outbound Calling Agent.
The template library, with a Select button on each option

The template library, with a Select button on each option.

3

Fill in the agent details

Fill in the fields:
  • Agent Name: the name the agent uses on calls. For example, Trudy.
  • Company Name: used in greetings. It defaults to the organization name.
  • Transfer Destinations (optional): who callers can be transferred to, one rule per line in plain English, as below.
Start every transfer number with sip:. It can be an extension (sip:1001) or a full phone number (sip:12127773456). For registered extensions, pbx: and ext: also work.
Click Create Agent.
The Receptionist Agent configuration, with two transfer rules

The Receptionist Agent configuration, with two transfer rules.

4

Enable the agent

The agent is listed as Disabled. Click the Status toggle to enable it, then click Edit to open the agent.
Enabling the agent with the Status toggle

Enabling the agent with the Status toggle.

2. Connect the agent to your PBX

SIP registration lets the agent send and receive calls like any other VoIP device on your PBX.
Before you start, get the SIP domain, username and password for this agent from your PBX administrator, and have them allow both of Telzino’s registration addresses in the PBX firewall: 143.47.120.76 and 161.153.52.190 (US), or 192.18.150.64 and 147.15.129.207 (Canada).
1

Start a new registration

The agent opens on the SIP tab, which shows “No SIP Registration configured.” Click New Registration.
The SIP tab, with the New Registration button

The SIP tab, with the New Registration button.

2

Enter the SIP details

Fill in the fields:
  • SIP Domain: your SIP server domain.
  • Username: the SIP username, usually the extension. For example, 3001.
  • Auth Username (optional): only if it differs from Username.
  • Password: the SIP password.
  • Proxy (optional): the outbound proxy host, if your platform uses one.
  • Use DNS SRV: turn on only if your server publishes DNS SRV records. This hides the Port field.
  • Transport: UDP, TCP or TLS, to match your server.
  • Port: usually 5060.
  • DID, Destination User and SIP User-Agent (optional): only if your PBX administrator asks for them.
The Create SIP Registration form

The Create SIP Registration form.

3

Verify the details

Click Verify to check the details. If a value is wrong, a red message appears under that field. Correct it and click Verify again.
Verify flags a proxy host that does not resolve

Verify flags a proxy host that does not resolve.

4

Create the registration

When verification passes, click Create. Wait a couple of minutes, then click the refresh icon next to Status. It shows “Successfully registered”.
A successful SIP registration

A successful SIP registration.

3. Configure the agent

All settings are on the Configure tab. Click Advanced at the top of the page to show every option, and click Save changes at the bottom after each section. The bar there reads “Unsaved changes” until you do.
1

Basics

Open the Configure tab. Under Basics, fill in the fields:
  • Name: the agent name.
  • Description (optional): an internal note to tell agents apart. For example, Receptionist Agent.
  • Timezone: the business’s local time zone. For example, America/New York.
The Configure tab, Basics section

The Configure tab, Basics section.

2

Greeting

Under Greeting & Instructions, fill in the fields:
  • Static greeting: the first thing callers hear. It is pre-filled from the organization name. Add any legal disclaimer or call recording notice here.
  • AI personalized greeting (optional): greets known callers by name when the agent finds them in a connected CRM or system.
  • Let callers interrupt the AI greeting (optional): lets callers talk over the greeting.
Static greeting and AI personalized greeting

Static greeting and AI personalized greeting.

3

Agent instructions

Review Agent instructions. They are pre-filled from the template. Add or change instructions for this business, or click Improve with AI to have them rewritten.
Agent instructions pre-filled from the template

Agent instructions pre-filled from the template.

4

Business information

Under Knowledge & Tools, give the agent information about the business:
  • Google business listing: search for the business (at least 3 characters, adding the city helps) and pick it, or paste a Google Place ID. Useful for small businesses without a website.
  • Reference links: website pages the agent reads, one URL per line.
  • Live knowledge search (RAG) (optional): searches a knowledge base on every call. Knowledge bases are created from the organization’s Knowledgebase button.
Searching for a Google business listing

Searching for a Google business listing.

Add this line to the Agent instructions so the agent uses those sources:
5

Functions

Under Functions, click Add function, then click Add next to each tool the agent needs: Webhook, Send Email During Call, Transfer to Agent, Slack Notification, Schedule, Listings Search or End of Call Action.
The list of functions the agent can use

The list of functions the agent can use.

6

Transfers

Under Transfers, check the Transfer rules (rules entered when the agent was created are kept), make sure Call transfer is on, and set the options:
  • Confirm before connecting: checks the destination can take the call, so callers do not land in voicemail.
  • Keypad (DTMF) transfers: lets callers also dial the destination on their keypad.
  • Ring time before giving up (seconds): how long a confirmed or web widget transfer rings before the agent gives up, 10 to 120. The default is 25. It shows while Confirm before connecting or Allow transfers from the web widget is on. When Warm transfer is on, it is in the warm transfer settings instead.
  • Allow transfers from the web widget: lets the website version of the agent transfer calls.
  • Transfer announcement: the agent announces each transfer. Leave Announcement text blank for the default line (“Please hold while I transfer your call”) or type your own.
Call transfer options

Call transfer options.

7

Warm transfer (optional)

Turn on Warm transfer, so the agent briefs the person first and connects the caller only if they accept (by saying yes or pressing 1). See Warm Transfer for details. Fill in the fields:
  • Warm transfer rules: which transfers should be warm. For example, “When transferring to customer service, first brief the destination about the subject of the call, then transfer to sip:8002”.
  • Hand-off introduction (optional): what the agent says to the person receiving the call.
  • Hold music (optional): click Upload file to add an MP3 or WAV of up to 4 MB.
  • Pass through caller number: shows the caller’s number to the person receiving the call.
  • Ring time before giving up (seconds): how long the destination rings before the agent takes the caller off hold, 10 to 120. The default is 25. The same setting applies to confirmed and web widget transfers.
Warm transfer rule and hand-off introduction

Warm transfer rule and hand-off introduction.

8

Save

Click Save changes.

4. Choose the voice and sound (optional)

The defaults work well. Change these to match the business.
1

Language

Under Voice & Audio, set Language: English - US, Spanish - US, French - FR, or Multilingual if callers may use more than one language.
Language options

Language options.

2

Voice

Set Voice (Katie is the default) and click Preview to hear it. For a multilingual agent, also set Spanish voice and/or French voice, which the agent switches to when the caller speaks that language. A French voice also needs a French variant: Canadian (fr-CA) or European (fr-FR). Adjust Speaking speed with the slider.
The voice list

The voice list.

3

Pronunciations

Under Pronunciations, add any names the agent may say wrong. Enter the Word or name (for example, Colonel) and how it Sounds like (for example, KUR-nel), click Apply to generate the pronunciation, then click Add.
A saved pronunciation

A saved pronunciation.

4

Filler while waiting

Under Sounds while talking, turn on Filler while waiting so the agent says a short phrase while it looks something up. Set Wait before speaking (for example, 1.5 seconds) and up to 10 Filler phrases, one per line.
Filler while waiting, with three phrases

Filler while waiting, with three phrases.

5

Other sounds

Optionally, turn on Thinking sounds (for example, Keyboard Typing), Background audio (for example, Office Ambience at 30%) and Slower list reading. Leave Turn-taking timing at its default values unless you need to fine-tune how long the agent waits before replying.
Background audio, slower list reading and turn-taking timing

Background audio, slower list reading and turn-taking timing.

6

Save

Click Save changes.

5. Set call handling, recording and follow-up

These settings decide how calls are screened, recorded and reported.
1

Call handling

Under Call Handling, set the options the business needs:
  • Max concurrent calls: leave blank for unlimited.
  • Detect automated callers: hangs up on robocallers and phone menus.
  • End calls with echo politely: ends calls on lines the agent cannot understand.
  • Detect mobile callers: checks whether the caller is on a mobile phone.
Call Handling and caller screening

Call Handling and caller screening.

2

Recording and privacy

Under Recording & Privacy, turn on Record calls to keep recordings in Call Logs for a configurable amount of time. Turn on Redact personal info in transcripts for healthcare and other regulated businesses.
Recording and Privacy

Recording & Privacy.

Where the law requires it, mention call recording in the static greeting.
3

Call summary email

Under After the Call, fill in the fields:
  • Send to: the email addresses that receive a summary and transcript of every call, separated by commas.
  • Include recording link: adds a link to the call recording. The link expires after 30 days.
  • Email server: leave as the default, or choose the business’s own server.
  • Summary instructions (optional): what every summary must include. For example, “Always include the customer’s account number as the first item.”
  • Skip calls where the caller never spoke: on by default, so no summary email is sent when the caller said nothing. The call is still saved in call logs. Count keypad presses as a response keeps the summary when the caller only pressed keys.
  • Skip calls shorter than (seconds) (optional): no summary email is sent for calls shorter than this, whether or not the caller spoke.
Call summary email settings

Call summary email settings.

4

Models

Under Models, choose the Language model (LLM) (GPT-4.1 is the default) and the Speech-to-text model (Deepgram is the default; Cartesia is also available). The text-to-speech model follows the voice you chose.
The models offered here can be forced or limited for each organization in Feature Administration (see Unlock the features you need).
Choosing the language model

Choosing the language model.

5

Save

Click Save changes. An “Agent updated successfully” message confirms the agent is ready to test.
The configuration is saved

The configuration is saved.

6. Test the agent

Talk to the agent before it takes real calls. You need a computer with a microphone.
1

Open the test playground

Open the Test Agent tab and click Launch Test Playground. A new browser tab opens.
The Test Agent tab

The Test Agent tab.

2

Start a conversation

Click Start Conversation and allow microphone access.
The test playground

The test playground.

3

Talk to the agent

Talk to the agent as a caller would. Check that it plays the greeting, then ask it to do real tasks, such as “Can you transfer me to Joe?”. The live transcript shows what the agent heard and said. Click End Session when you are done.
A live test conversation

A live test conversation.

If the agent mishears a name or answers wrongly, adjust the instructions or add a pronunciation, save, and test again.

7. Connect integrations (optional)

Integrations let the agent work with other systems, such as a CRM, calendar or SMS. They must be Available in Feature Administration first.
1

Open the Integrations tab

Open the Integrations tab. The available integrations are listed: Google Calendar, Outlook Calendar, Twilio SMS, TeamMate, Model Context Protocol (MCP) and Inject Agent on Web Page.
The Integrations tab

The Integrations tab.

2

Add an MCP connection

To connect a system through MCP, click Model Context Protocol (MCP), click + next to Connections, then fill in the fields:
  • Connection Name: a friendly name. For example, CRM.
  • MCP Endpoint URL: the full URL of the MCP server.
  • Transport Type: SSE or HTTP, to match the endpoint URL.
  • Authentication Headers (optional): the JSON headers the server needs, such as an API key.
  • Initialization Options (optional): leave as {} if not needed.
The MCP connection settings, with authentication headers hidden

The MCP connection settings (authentication headers hidden).

The connection details depend on the MCP server and are usually in its documentation. An AI assistant such as ChatGPT or Claude can help you fill them in.
Authentication headers are secrets. Do not show them in screenshots, recordings or chat.
3

Test and save the connection

Click Test Connection. The server’s tools are listed. Leave every tool on, or untick the ones the agent should not use, then click Save Connection. The MCP card shows a green Connected badge, and your agent is fully set up.