Skip to main content
An agent is the AI that talks to your customers. You build and configure it in the agent editor, where you define how it behaves, what it knows, what actions it can take, and which channels it uses.

​
Create an agent

Open Agents and select Create Agent. You get two doors:

Build with Lily

Recommended. Lily creates an empty agent, opens it, and interviews you - the name, what the agent is for, then the greeting, purpose, response style, and call flow. You review everything she writes before it lands.

Start from scratch

Give it a name and nothing else. Language, speech, model, voice, and post-call analysis start on platform defaults, and you configure the rest yourself.
Either way the agent is created immediately and opens in the editor.
Start from scratch gives you an outline, not a finished prompt. It fills each section with a placeholder describing what to write there. Replace it before the agent takes calls.

​
How the agent editor is laid out

The editor is two columns side by side, plus Lily when you open her.

Left - the agent editor pane

Always visible. The greeting, the three prompt sections, and prompt variables - with the four configuration pills across the top.

Right - the mode switcher

A single row of three tabs. One tab’s settings show at a time.
Either column can be folded to a strip of icons using the control in its top-right corner, and clicking an icon on a folded strip unfolds that column at the thing you picked. When Lily takes half the screen, one column folds automatically - narrowing her opens both again.

​
The four pills

A row of four pills sits at the top of the editor pane. They are the agent’s model choices, read left to right in pipeline order - and each one is narrowed by the one before it, so nothing you pick later can invalidate an earlier choice.

Language

The language a call starts in, the languages it may switch into, and when switching is allowed. Everything else is filtered by this, so it comes first.

Transcriber

The speech recognition model, keyword boosting, turn-taking, and a backup.

Intelligence

The language model, Temperature, Max Tokens, and a backup model.

Voice

TTS provider, voice, speaking speed, emotion, preview, and a backup voice.
Each pill opens a menu for the choice itself, and a gear for its finer settings.

​
The three tabs

Settings

Live-call behavior, duration and silence limits, webhooks - and what the call leaves behind: Custom Analysis and Saved Artifacts.

Tools

What the agent can do and what it can read: built-in tools, your own API tools, Cal.com actions, and knowledge bases.

Deploy

Where customers reach the agent. Three sub-tabs: Phone, Web, and WhatsApp.
Tabs appear in that order, and Deploy is the only tab with sub-tabs. Every other tab is one continuous panel you scroll.
If you used Vocily before, several things moved.
  • Intelligence and Voice are no longer tabs. They are pills in the top row, alongside the two new ones: Language and Transcriber.
  • The Conversation Language dropdown in Agent Core is replaced by the Language pill, where an agent has a primary language and a set of languages it may switch into. Greeting Language and Auto Language Rules are gone.
  • Capabilities is now Tools, with knowledge bases on the same tab.
  • Runtime is now Settings, and the Outcome tab has been folded into it - Custom Analysis and Saved Artifacts are at the bottom of Settings.
  • Transfer and the closing message are now the Transfer Call and End Call tools on the Tools tab.

​
Build with Lily

Lily, your Catalyst is Vocily’s agent assistant, and the fastest way to get an agent built and working. She lives at the top of the sidebar, opens as a dock beside whatever page you are on, and can read and rewrite the agent you have open - the prompt, the model, the voice, the call settings, the knowledge bases, the tools. Everything she proposes is staged as a reviewable change on the draft; nothing is applied until you accept it.

Lily, your Catalyst

How the interview works, what she can change, and what she will not touch.

​
One agent, multiple channels

Each agent can work across phone calls, WhatsApp, Web Talk, and Web Chat. You configure the agent once, then use Deploy to connect and tune the channel settings it needs.
Designing for WhatsApp is different from designing for a phone call - message length, templates, and the 24-hour window all shape the experience. Vocily keeps the same agent model, while Deploy lets you configure the channel-specific setup for each place the agent runs.

​
Saving and going live

Saving stores your work in the draft you are editing - customers keep getting whatever is live until you say otherwise. Going live is three steps:
  1. Save your draft.
  2. Publish it, which freezes it as a numbered version.
  3. Make live - the only step your customers feel.
Editing an agent therefore never interrupts the one answering calls, and you can go back to an earlier version at any time. See Versions.

​
Test before you rely on it

At any point you can test the agent directly in the app - by chat, by voice in your browser, or by placing a real phone call - before you send real customers to it.

Start with the agent editor pane

The persona and the prompt are the foundation of every agent.