Create an assistant

An assistant is the configured brain of every call. It combines an AI model, a voice, a system prompt and first message, optional knowledge, and tools such as call forwarding or recording.

Before you start

Make sure your team workspace is set up and you know:

The use case — inbound receptionist, outbound campaign agent, or both.
The voice and language you want the assistant to speak.
Any knowledge the assistant should ground its answers in (docs, FAQs, or a website).

Create the assistant

Open Assistants from the sidebar.
Select New assistant and give it a clear name.
Choose a template if one matches your use case, or start from scratch.

Choose the voice

Configure how the assistant sounds and understands speech:

AI model — the reasoning engine that drives the conversation.
Transcriber — converts the caller's speech to text in real time.
Voice — the spoken voice your assistant uses on the call.

Start with the defaults, then adjust after you hear a few test calls.

Write the prompt

Give the assistant its instructions:

System prompt — role, goals, tone, and guardrails. Use variable mentions to inject team-specific values.
First message — what the assistant says when the call connects.

Keep the prompt focused on one job. Shorter, specific prompts produce more reliable calls.

Attach knowledge and tools

Knowledge bases — link documents, FAQs, tables, or crawled websites so the assistant can retrieve accurate answers during the call.
Tools — enable calendar booking and CRM stage updates (through connected integrations), plus toggles for end call, record, and transcript.
Forwarding number — set a number to live-transfer to a human. See Call forwarding.

Test and publish

Use Test voice to talk to the assistant in your browser.
Review the transcript and evaluation, then refine the prompt or voice.
Publish the assistant so it can be attached to a phone number or campaign.

Next, connect a number so the assistant can take real calls: Add an inbound number.