Create an Agent
- Go to the Agents section from the left navigation bar.
- Click the Create Agent button in the top right, or if you’re a new user, click Create your first agent in the center of the page.
- Fill in basic info: agent name, language, and description.
- Configure a provider and connection type.
Basic Info
A label to identify this agent within Cekura. Choose something descriptive so you can easily distinguish it from other agents (e.g. “Support Bot Production” or “Sales Agent v2”).
Language
The primary language your agent speaks. Cekura uses this to run simulations and generate evaluators in the correct language.
A comprehensive overview of your agent’s purpose, behaviour, and operational guidelines. See the Agent Description section below for details on what to include and how to structure it.
Agent Description
Agent Description is a comprehensive overview of your voice conversational agent’s purpose, behaviour, and operational guidelines. It encapsulates all the instructions, prompts, and rules you provide to your AI agent.By clearly articulating your agent’s description, you enable us to generate automatic, accurate evaluators.
Most customers include their
agent’s main prompt here.
Type of Agent
Single Prompt Agent
Single Prompt Agent
For Single prompt agents, you can copy the agent’s main prompt here.
Multi State Agent
Multi State Agent
If you have multi-state agent, you can build a description in a json structure.
Note - We are very flexible with description so even if you copy paste all your prompts and paste as one huge prompt, we can process. JSON is the recommended approach.
Note - We are very flexible with description so even if you copy paste all your prompts and paste as one huge prompt, we can process. JSON is the recommended approach.
Retell Workflows
Retell Workflows
If you have an agent with Retell workflows, you can export the agent’s details as json from Retell.
Go to agents -> select the agent -> click on export button (this will download a .json file) -> copy the content of the file and paste it in agent description field.
Go to agents -> select the agent -> click on export button (this will download a .json file) -> copy the content of the file and paste it in agent description field.

Vapi Workflows
Vapi Workflows
If you have an agent with Vapi workflows, you can export the agent’s details as json from Vapi.
Go to workflows -> select the agent -> click on
Go to workflows -> select the agent -> click on
code button on top right -> copy the content shown and paste it in agent description field.
You can follow the same approach for other providers.
Configuration
The Configuration section connects your Cekura agent to the actual AI platform powering it. This is what allows Cekura to simulate calls, run evaluations, and observe live traffic against your real agent.Provider
Select the platform your agent is built on. Cekura supports a wide range of providers out of the box. You can explore different providers below and go to their integration specific docs to know more.Custom
LiveKit
Pipecat
ElevenLabs
Chirp
Retell
VAPI
Cisco Webex
Bland AI
Synthflow
Kore.ai
Trillet
Genesys
Agentforce
Agora
Connection
Choose how Cekura connects to your agent to run simulations and collect data. The available connection types depend on your agent’s channel — voice or chat.Voice Connection
Telephony
Telephony
Connect via a phone number. Cekura will dial your agent directly over the PSTN network, making it the simplest way to test a phone-based voice agent.

SIP
SIP
Connect via a SIP trunk. Suited for enterprise telephony setups where your agent is reachable through a SIP endpoint rather than a regular phone number.

Chat Connection
SMS
SMS
Connect via SMS. Enter the phone number your agent sends and receives chat messages on.

WhatsApp
Connect via WhatsApp. Enter the WhatsApp-enabled phone number your agent sends and receives chat messages on.Configure Twilio credentials in Settings → Provider API Keys for WhatsApp integration.See WhatsApp Testing Guide for detailed setup.

WebSocket
WebSocket
WebSocket enables real-time, two-way chat between your agent and the LLM, allowing dynamic and continuous conversation testing.
Enter your WebSocket URL
Add the WebSocket endpoint your chatbot is reachable on (e.g.
wss://your-chatbot-endpoint.com).
Provider Chat
Provider Chat
Connect directly through your provider’s native chat interface — no phone number or messaging service required.
Provider chat is configured automatically using your voice agent settings. However, Retell and Bland AI require some additional configuration — you’ll need to provide a separate chat agent ID for those providers.

Additional Settings (Optional)
These settings are available when editing an existing agent. To access them, go to the Agents section in the left sidebar → click on the agent you want to configure.Knowledge Base
Knowledge Base
Upload reference material your agent uses, so Cekura can generate more accurate evaluators grounded in your content.We support two options for Knowledge Base:
- Upload a document — drag and drop or select a file directly. Supported formats: PDF, TXT, JSON, MD, CSV, XML.
-
Connect an external source — click Add Connector to sync knowledge base content from an external source such as Google BigQuery or a Website Scraper.

Mock Tools
Mock Tools
Define mock input/output data for tool calls so Cekura can simulate your agent’s tool usage during testing.There are two ways to add mock tools:
- Auto-Fetch — automatically pulls your tool definitions directly from your provider, so you don’t have to add them manually.
-
Add Tool manually — create a mock tool yourself and define its behaviour using Input/Output Mappings in JSON.

Dynamic Variables
Dynamic Variables
Dynamic variables are values injected at call time during scenario generation — letting you parameterise your test scenarios with real-world data like names, addresses, or account IDs instead of hardcoding them.Click Add Variables to define a new variable. Each variable has a name (e.g. 
Read more about Dynamic Variables.
address, name) and a default value that Cekura uses when running simulations.