Configure an agent

Once you create an agent, you open it up and click on the Configure tab. The Configure tab has three parts: General, Tools, and Models. All three together govern and control the entire behavior of the agent during a conversation.

General

Set the agent's Name, the display name visible to users during interactions.
Select a Timezone to provide the agent with accurate time-based context.
Set the primary Language, and add any Additional languages the agent should understand or respond in.
Attach a Knowledge base so the agent can draw on your documents when answering. Use Select a knowledge base to choose a collection.
Write the First message, the opening line the agent delivers at the start of every session.
Fill in the Prompt with the base instructions that shape the agent's personality, scope, and behaviour.
Edits are tracked as you type. An unsaved-changes banner appears when there is uncommitted work; save before navigating away.

Attaching a knowledge base enables retrieval for that agent. Keep your collections tidy and up to date in the Knowledge area, since the agent answers from whatever you've published there.

Tools

Tools extend the agent beyond conversation into action.
In the Tools section, browse System tools, the built-in capabilities grouped under Conversation, Telephony, Integration, and Other.
Toggle a tool on to enable it. A confirmation appears and the tool is added to the What this agent can do panel.
Select any enabled tool to open its configuration, adjust settings as needed, and save.
To add custom capabilities, click Add tool and configure a callable webhook or a client-side integration.
If there are no active tools yet, you will see "No tools enabled" with an invitation to Activate a tool

Some tools require additional configuration before they function. Transfer call, for example, requires an outbound number to route to. When you enable such a tool, a configuration panel opens automatically for the required fields..

Webhook tools connect to your backend systems. Client tools interact with the browser interface in real time. Writing a precise description for each tool is the most important thing you can do, the agent reads it to determine when the tool is appropriate, checkout Tools and Webhooks.

Models

The Models section is where you select the underlying AI models that power your agent. It contains two parts: Voice configuration for spoken conversations and Chat configuration for text-based chat.

Which section applies depends on how your agent is deployed:

Audio only, for phone calls and voice sessions. Only voice configuration applies.
Text only, for chat widgets. Only chat configuration applies.
Both, configure both sections.

Set up whichever applies to your agent, and you can always come back and add the other later.

Voice configuration

For agents that speak, choose a Voice Playbook and tune the underlying voice models.

Under Choose Playbook, select a speech flow:

STT LLM TTS for a cost-effective, highly controllable pipeline. Speech is transcribed to text, sent to the language model, then converted back to speech.
Realtime TTS for low-latency speech generated directly from the model's text.
Realtime for near-instant, end-to-end voice conversations.

Configure the relevant models for your playbook:

LLM, the brain that understands and writes replies.
Speech to Text, which turns what the caller says into text.
Text to Speech, which gives your agent its voice.
Realtime, which powers instant, end-to-end voice conversations.

Pick the Language, the Agent Voice, and the Pace that controls how fast the agent speaks.

Start with STT LLM TTS when you want fine control and predictable cost, and move to a Realtime playbook when low latency matters more than tuning each stage.

Chat configuration

For agents that handle text chat, pick the Chat model, the model that powers typed conversations in your chat widget. Choose one from the list and save. This is separate from the voice models above, so a chat agent gets its own model without any of the speech setup.

Test your agent

Click Test Agent from the agent header.
A live preview opens in a side pane. Interact with the agent exactly as a user would.
Close the pane when done.

Testing requires the agent to be deployed and live. If the Test Agent button is unavailable, deploy your agent first.

When the agent behaves the way you want, move on to Deploy an agent.