Skip to main content

​
Overview

If your voice agent is built on LiveKit Agents and uses AssemblyAI for speech-to-text, connect it to Cekura with the LiveKit integration. Cekura joins a LiveKit room as the caller, your agent transcribes the caller with AssemblyAI, and every run is scored with your metrics.

​
Prerequisites

  • A LiveKit agent deployed to LiveKit Cloud or your own LiveKit server, with an explicit agent name so Cekura can dispatch it.
  • LiveKit API key, API secret, and server URL.
  • AssemblyAI STT configured in the agent, through LiveKit Inference or the AssemblyAI plugin.

​
Use AssemblyAI STT in your LiveKit agent

No separate AssemblyAI API key is needed; usage is billed through LiveKit Cloud.
See the LiveKit AssemblyAI STT guide for the available models and parameters.

​
Connect the agent in Cekura

1

Select LiveKit

In Cekura, go to Agents, create an agent, and select LiveKit as the provider.
2

Enter the LiveKit credentials

Enter your LiveKit API Key, LiveKit API Secret, LiveKit URL, and the Agent Name your AssemblyAI-powered agent is registered under. See LiveKit (Automated) for every field.
3

Run evaluators

Open an evaluator, click Run, and select WebRTC under Voice connections. Cekura creates a room per run, dispatches your agent, and plays the scenario as the caller.

​
Tips

  • Use LiveKit tracing to send your agent’s own transcript, tool calls, and latency to Cekura alongside each run.
  • Include accents, background noise, and interruptions in your test profiles to check how AssemblyAI transcription and turn detection hold up.
  • If the agent never joins the room, confirm the Agent Name matches the name the agent registers with and that the agent worker is running.