Creating Custom Personalities

Personalities in Cekura define how your AI agents behave, communicate, and interact with users. They encompass the agent’s tone, knowledge base, response style, and voice characteristics, allowing you to create unique and purposeful AI experiences.

Getting Started

1. Access the Dashboard

1. Log in to your account at dashboard.cekura.ai
2. Navigate to the Personalities section from the sidebar menu

2. Creating a New Personality

You have three options for setting up personalities:

1. Use Pre-defined Personalities: Choose from Cekura’s curated collection of optimized personalities
2. Fork a Pre-defined Personality: Copy one into your project and change only what you need — everything else is inherited
3. Create Custom Personality: Design a completely new personality from scratch

Pre-defined personalities are shared across all workspaces, so they cannot be edited directly. Fork one when you want to keep its voice and accent but change its behavior — for example to raise the idle timeout. Forking is also the quickest route to a custom personality, since you don’t have to re-pick the voice settings.

To create a custom personality:

1. Click the “Create Personality” button in the Personalities section

2. Fill in the following required fields:

Basic Information

• Personality Name: Give your personality a unique, descriptive name
• Personality Description: Briefly describe the personality’s purpose and characteristics

Behavior Configuration

• Personality Prompt: Define detailed instructions for how the AI should behave and respond
• Temperature: Set the creativity level (default: 0.7)
• Model: Choose the AI model that powers your personality

Voice Settings

• Background Noise: Control background noise during conversations. The dropdown offers several options including built-in noise presets and a Custom URL option — select “Custom URL” and enter a direct link to an audio file (e.g., https://<your-domain>/background.mp3) to play any sound file as background noise during the simulation.
• Voice ID: Select a unique voice identifier
• Voice Provider: Choose your preferred voice service provider

Advanced settings

Open the Advanced Settings section of the personality form for these. They control how the testing agent behaves during a call, and cannot be set from an evaluator — instructions and expected outcomes only control what the testing agent says, so anything in this section has to be configured here.

Setting	What it controls	Default
Idle Timeout (seconds)	Seconds of silence before the testing agent prompts (“Are you still there?“)	10
Max Idle Prompts	How many times it prompts before ending the call	3
Interruption Level	How aggressively the testing agent talks over your agent	Preset per personality
Start Wait Seconds / stop speaking plan	Fine-grained control over when the caller starts and stops speaking. Ignored when an Interruption Level preset is set	—
Network Simulation	Packet loss, jitter and latency, for degraded-network testing	Off

Making the testing agent wait in silence. If your test needs the testing agent to say nothing while your agent works — a long lookup, hold music, an IVR menu — raise Idle Timeout past the expected pause. Adding “stay silent” or “do not respond” to the agent description or the evaluator’s instructions will not work: the idle timeout fires regardless.

Both idle fields must be set together, and both must be 1 or greater — the idle prompt can be pushed out but not switched off entirely.

Updating a personality over the API

The same fields are available programmatically:

• Fork Personality — copy a personality into one of your projects
• Update Personality — change any setting on a personality your organization owns, including a fork you just made

To customise a pre-defined personality, fork it and then update the copy:

Update the fork with a longer idle timeout

{
  "message_plan": {
    "idle_timeout_seconds": 45,
    "idle_message_max_spoken_count": 1
  }
}

Next Steps

After creating your personality, you can:

• Assign it to specific agents
• Test interactions
• Monitor performance
• Make adjustments as needed