Agent Core

Agent Core is the left-hand pane of the agent editor - the column that stays on screen whichever tab you have open on the right. It tells Vocily who the agent is, how it should behave, what it says first, and which variables it can use while talking to a customer. Because one agent can run across multiple channels, Agent Core should describe the agent’s core job and behavior. Channel-specific setup, such as phone numbers, web widgets, and WhatsApp assignment, lives in Deploy.

The pane folds. Use the control in its top-right corner to collapse it to a strip of two icons - Greeting and System Prompt - and click any icon to unfold it at that box. It folds on its own when Lily takes half the screen.

What Agent Core controls

Agent Core is the pane you use for the agent’s most important conversation settings:

Greeting

Write the opening message. It is spoken in the agent’s primary language.

System Prompt

Write the agent’s role, response style, and conversation flow in one prompt.

Default Instructions

Turn on ready-made rules for tone, repetition, numbers and honesty instead of writing them into the prompt.

Variables and defaults

Add reusable variables such as customer names, booking details, or other values the agent should personalize.

References to tools

Reference connected tools and API data inside the prompt when the agent needs them.

Write the greeting

Greeting controls the opening of a voice call, including browser Talk Test and widget voice. Choose how the opening is written:

Mode	What to enter
Custom	The exact words the agent should speak.
AI-generated	Nothing extra. The agent creates an opening from its system instructions and prepared call context.

For Custom, keep the message short, identify the business, and give the caller an obvious next step.

Hello {{customer_name}}, thanks for calling. How can I help you today?

For AI-generated, put any desired opening style in the System Prompt. There is no separate greeting instruction field. The custom message supports {{variables}} and is preserved when switching modes, but is spoken only in Custom mode. Generated greetings use variables already prepared in the system context. Both modes use the primary language and selected voice.

Set when the greeting starts

Open the settings icon beside the greeting mode dropdown. Start greeting after accepts whole seconds from 0 to 120; zero starts immediately. If the caller speaks during the delay, or while an AI greeting is being generated, the opening is skipped and the agent responds to the caller. If they stay silent, the selected greeting starts after the delay. AI generation begins then and adds a short model-dependent delay. These opening controls apply to voice calls. They do not schedule a greeting in chat or WhatsApp.

Write the system prompt

Use System Prompt for instructions shared across every conversation: the agent’s role, business context, response style, boundaries, and steps to follow. You can use your own headings or plain text. Existing prompt text stays intact, including any headings you already wrote.

You help customers find and book appointments.
Ask one question at a time. Never invent availability.
Find the appointment type and preferred time, check availability, confirm the booking,
and summarize the next step.

The Rich/Raw editor supports headings, bold text and lists. Configure allowed languages in the Language pill so your prompt agrees with the platform’s language rules.

Add instructions for a channel

Choose Add channel instructions when a channel needs different behavior. Available contexts: Inbound phone, Outbound phone, Web voice, Web chat, and WhatsApp. Only the matching addition is appended to the shared system prompt. For example, outbound phone instructions can explain why the agent is calling, while WhatsApp instructions can request shorter messages. These additions do not replace the shared prompt or change where the agent is deployed. Leave a channel blank or remove it to use just the shared prompt there. Tool references and variables work in channel instructions too. A tool mentioned only in the WhatsApp addition is not activated by that mention in a phone call; its configured channel availability still applies.

Turn on Default Instructions

Default Instructions are ready-made rules you switch on instead of writing them into the prompt. Open the Default Instructions pill on the right of the System Prompt heading, turn on what you want, then choose Save. On the Settings tab, switch between Voice and Chat to see the settings for calls or for chats. You have two options for every setting:

Turn it on. Vocily adds its text for you.
Copy it and improve it. Select the code icon on a setting, copy the text, edit it and paste it into your prompt. Leave that setting off so the rule is not added twice.

Settings

Tab	Setting	What it does	Applies to	On for new agents
Behaviour	No Repetition	Says a repeated question or answer in different words. Facts stay the same	Calls and chats	Yes
Behaviour	Respectful Address	Uses the respectful form of you, in languages that have one	Calls and chats	No
Behaviour	Agent Gender	Uses the right word forms for the gender you pick, in languages that need it	Calls and chats	No
Default Tone	Default Tone	Pick one: Professional + Conversational, Empathetic, Energetic, or Firm and Polite	Calls and chats	Yes, Professional + Conversational
Settings, Voice	Natural Filler Words	A light filler about once every 2 to 3 sentences	Calls	No
Settings, Voice	Number Reading	Says money, dates, times and phone numbers in words	Calls	Yes
Settings, Voice	Smart Matching	Accepts a close mishearing of a name or product	Calls	Yes
Settings, Chat	Answer Length	Makes answers about 30% shorter or longer	Chats	No
Settings, Chat	Emoji Use	Never uses emojis, or at most one in a message	Chats	No
Settings, Chat	Number Reading	Writes money, dates and phone numbers neatly with digits	Chats	Yes
Settings, Chat	Smart Matching	Accepts a close misspelling of a name or product	Chats	No
Trust & Safety	AI Disclosure When Asked	Says it is an AI when asked if it is a person	Calls and chats	No
Trust & Safety	Scope Boundaries	Answers only from its prompt and knowledge base, and does not guess	Calls and chats	Yes

Calls means phone calls and web voice. Chats means web chat and WhatsApp. Agents created before Default Instructions keep every setting off until you turn one on.

How the rules reach the agent

Default Instructions are added to the prompt when a conversation starts. The rules go above your prompt, and No Repetition, Natural Filler Words and Number Reading for calls also add a one line reminder below it. They never change the text you wrote. Each conversation gets only the rules that fit it. A call gets the rules for calls and a chat gets the rules for chats, so the rule for writing numbers neatly never reaches a phone call. The same goes for channel instructions: only the ones for that channel are sent. If your prompt says something different from a default instruction, your prompt wins.

Text that is sent word for word does not follow these rules: a greeting you typed, WhatsApp templates, and the lines a tool speaks. To keep a line from your prompt word for word when No Repetition is on, write it as Say exactly: "This call is recorded."

Smart Matching is not for identity checks. It accepts a name that is close, so do not rely on it when you need to be sure who is on the line.

Changes to Default Instructions are saved with your draft, like the prompt, and go live when you publish. See Versions.

Set the agent’s timezone

Select the clock button beside Default Instructions to open Current Time Awareness, then pick the agent’s timezone. Every call and chat is told the current local date and time in that zone, so the agent understands “today”, “tomorrow”, “in 2 hours”, business hours and scheduling windows. Calendar tools also use it when the caller does not name a timezone. New agents start on your account’s timezone:

Account billed in	Starting timezone
Rupees (INR)	India, `Asia/Kolkata`
Dollars (USD)	US Pacific, `America/Los_Angeles`

Agents created before this setting keep India time until you change it. The timezone is saved with your draft and goes live when you publish.

Use variables and defaults

Variables let one agent personalize conversations with values that can change by customer, campaign, API result, or test run. Add a custom variable with double braces:

{{customer_name}}
{{appointment_type}}
{{order_id}}

When Agent Core sees a variable in the greeting or prompt, it lists that variable under Prompt Variables & Defaults. A reference alone does not declare a variable. Select Declare beside an undeclared reference so campaigns can ask for its value. Use that area to:

Select Declare when a referenced variable has not been created yet. You can give it an optional fallback value.
Select Add to create a variable directly.
Edit or delete variables that should no longer be used.

Default values are fallbacks. If Vocily receives a real value for the variable during a conversation, the live value should be used. If no value is supplied, the default keeps the agent from sounding incomplete. For a deeper explanation, see Variables and default values.

Reference tools and API data

Agent Core can also reference the agent’s tools and the data they return, so the agent knows they are available in the conversation. The editor renders each reference as a pill as you type. There are four formats:

{{custom_variable}}             a custom variable
<tool_name>                     grants the prompt access to a tool
{{api.tool_name.api_field}}     data returned by a during-call API tool
{{pre.tool_name.api_field}}     data fetched by a before-the-call API tool

Use these only after the tool exists in Tools. Keep the instruction clear: tell the agent when to use the tool, what information to collect first, and what it should say if the action cannot complete.

Save and test

Select Save Changes after editing Agent Core. Saving stores the draft you are editing - it does not change what customers hear. Publish the draft and make it live when you are ready (see Versions). After saving, test the most important parts:

In Talk Test, staying silent plays the selected greeting once, with the configured delay.
Speaking during the delay gets a response to your words instead of the scheduled greeting.
Both Custom and AI-generated sound natural and use variables correctly.
The agent’s language set behaves as expected - it greets in the primary language and follows a caller who switches.
The agent follows the flow in System Prompt instead of skipping steps.
Any tool references are only used when the matching tool is configured.

Use Test to try the agent by chat, browser voice, or a real phone call before sending customers to it.

Let Lily write it

Lily can rewrite the greeting, shared system prompt, and individual channel instructions. To change generated openings, ask Lily to update the system prompt. An explicit greeting text rewrite selects Custom when accepted; the text and mode change are reviewed together. Review the suggested text in Lily’s chat and select Save to write the accepted changes to the draft. You can also ask her to switch greeting mode or set the delay. Those setting changes are saved directly to the draft. For example: “Start the greeting after 5 seconds.” Lily does not publish the draft or make it live.

What’s next

Set the languages the agent speaks in Language.
Configure the model and response parameters in Intelligence.
Add tools and knowledge bases in Tools.
Choose the agent’s TTS voice in Voice.
Connect the agent to channels in Deploy.

Get Started

Building Agents

Test

Conversations

Call Campaigns

Integrations

Phone Numbers

WhatsApp

Account & Workspace

What Agent Core controls

Greeting

System Prompt

Default Instructions

Variables and defaults

References to tools

Write the greeting

Set when the greeting starts

Write the system prompt

Add instructions for a channel

Turn on Default Instructions

Settings

How the rules reach the agent

Set the agent’s timezone

Use variables and defaults

Reference tools and API data

Save and test

Let Lily write it

What’s next