Tarush AgarwalApr 3, 2026
Cekura - Retell Integration
2views
Cekura - Retell Integration
Tarush Agarwal
Learn how to integrate your Retail voice agents with Cekura. This guide covers agent setup, automatic prompt syncing, generating test scenarios, and analyzing call performance metrics.
https://docs.cekura.ai/documentation/integrations/retell/testing
Cekura - Retell Integration
Tarush Agarwal
Learn how to integrate your Retail voice agents with Cekura. This guide covers agent setup, automatic prompt syncing, generating test scenarios, and analyzing call performance metrics.
https://docs.cekura.ai/documentation/integrations/retell/testing
Chapters
00:00
Introduction to Secura Integration
An overview of building and integrating voice agents using Secura's retail platform.
00:21
Connecting Your Retail Agent
02:22
Configuring Integration Settings
03:08
Generating Test Scenarios
08:13
Running Test Evaluations
08:53
Analyzing Call Results
10:11
Conclusion and Next Steps
Transcript
00:00
Hey.
00:00
Today I'm walking you through Secura's Retail Integration.
00:04
A lot of teams are building voice agents on retail and we have built first class support for it.
00:12
To get started, simply connect your retail agent directly.
00:16
In Sakura we call it Creating an agent.
00:19
Give it a name, a language and a description of your agent.
00:26
For the description you can paste your agent prompt directly.
00:31
You can see my retail agent here.
00:32
I'm using a single prompt retail agent and for this you can simply copy paste the prompt over here.
00:42
If you have a multi prompt conversation flow you can always export it as a JSON copy paste that JSON onto Sakura.
00:54
One thing worth noting, we auto sync your prompt with retail so any future updates you make there will reflect here automatically.
01:05
Next, choose Retail as the configuration provider and enter your retail agent id.
01:14
You also need to paste your retail API key You also get two integration toggles worth knowing about auto fetching, your production calls for monitoring and analysis and the prompt auto sync we just mentioned.
01:33
Once configured, you unlock a few testing modes, the retail WebRTC, chat based testing, telephony and a few others for chat.
01:47
You will need to convert this retail agent into a chat agent and paste this agent ID over here.
01:59
If you don't care about testing in chat, simply skip this.
02:06
Next Mock Tools One of the most important features for realistic testing if your agent uses tool calls to fetch data, say which appointments a patient has, booking an appointment, etc.
02:24
You don't want to hit live APIs during tests.
02:29
Hit auto fetch and Secura pulls in all the tools your agent is configured with.
02:35
Note no change takes place in your agent at this step.
02:42
In my case you can see get patient info, check appointments and book appointments are the three functions enable mock tools so Secura intercepts and handles these tool calls during a live run.
03:02
Note at this point your retail agent is modified and you would want to come back and disable this to restore the original state.
03:12
Once happy with testing, Next up is dynamic variables.
03:16
Secura automatically fetches the dynamic variables your agent needs.
03:22
In this case we picked up the agent name right away.
03:27
You may have others Secura missed and you can always add those manually With the agent setup.
03:35
Let's go to the evaluator section and generate test cases.
03:41
I'll create a new folder, Pick an agent of choice, add the number of scenarios I want, and hit generate.
03:57
you can always add extra instructions.
03:59
For example, if you only want scenarios around booking appointments, you can specify that.
04:07
What's unique about our retail integration is for each test case we generate the full scenario that the test agent will play out.
04:17
Example, booking, canceling, etc.
04:20
The dynamic variables your agent needs, in this case the agent name.
04:25
And lastly the mock tools values align to that specific scenario.
04:30
So if a test case involves canceling an appointment on January 15, the mock data will have exactly that appointment.
04:40
Once generated, you can review and accept the ones you want.
04:46
It's an AI agent, so you can always talk to it and modify it as needed.
04:53
Next up, we'll run these test cases.
04:57
Select all the evaluators you want to run, hit run and choose your testing mode.
05:03
Chat, phone or retail WebRTC?
05:08
Mine is a retail outbound agent.
05:16
You'll note we automatically hit trigger retail to make phone calls to us.
05:23
You can see the calls going in progress now.
05:27
While that's running, let's.
05:28
Let's look at an older result, five tests and an 80% pass rate.
05:36
You can see a good summary of what happened here.
05:41
Let's look into the failed call.
05:43
On the right you'll see the call recording itself, the call ID from retail, and the entire transcript including all the tool call details.
05:57
On the left is Secura's evaluation.
06:01
You can see the call is failed because the expected outcome metric failed.
06:06
Reason being the agent asked for patient id, not the patient name and same for date of birth.
06:12
You'll also see the transcription accuracy metric.
06:16
Retail score 100%.
06:18
Here we fetch the transcript from retail.
06:21
We also have the ground truth of what we said during the call and compare the transcripts.
06:29
Latencies and other metrics are there too.
06:32
Plus the full raw call JSON from retail so you can build any custom metrics on top.
06:40
So WebRTC, chat, auto outbound calling, dynamic variables, mock tools and detailed result View all first class in retail.
06:51
All in the UI docs are linked below.
06:54
Give it a try.
Comments
Comments
No comments yet