Create and configure an agent

1. Create the agent

Open Voice AI → Agents, select Create agent, enter a recognizable name, and choose the engine that matches your use case.

2. Configure the conversation

Setting	What it controls	Recommendation
Greeting type	Fixed text or generated first turn	Use fixed for regulated or exact openings
Greeting line	First words the caller hears	Include company and reason for calling
Fallback greeting	Used if the preferred greeting cannot be produced	Keep it short and context-safe
Agent prompt	Role, goal, boundaries, and conversation policy	State success criteria and escalation rules
Knowledge bases	Approved business information available to the agent	Attach only current, published sources

Realtime engines also have their own prompt configuration. Treat it as an independent configuration rather than a copy of the Classic prompt.

3. Configure speech

Select the provider-specific voice and language. For Classic, also verify the TTS provider, TTS speed, and transcription languages. For Vani Ultra, configure the supported Vani voice and optional pronunciation vocabulary. For Ultra Realtime, select the Gemini Live voice and language.

4. Set call behaviour

Enable Interruptions when callers should be able to speak over the agent.
Enable Background music only after listening to a real phone call.
Set a hard end-call limit to protect callers and usage.
Keep debug options off for normal production calls.

5. Configure analysis

Choose default, structured, or advanced analysis. If downstream automation depends on a value, prefer structured fields and test missing or ambiguous answers.

6. Test before assignment

Save the agent and place a controlled test. Confirm:

greeting and language;
correct company facts;
turn-taking and interruptions;
safe handling of unknown answers;
call ending behaviour;
transcript, recording, outcome, and analysis.

Only then assign a number from Phone numbers. An outbound campaign may use one number or a rotation pool of up to ten numbers.