Home
Contact usLog inSign up
Release notes

Voice AI Testing and Observability Now Available with Cekura for Telnyx Voice AI

15, Jul 2026

Voice AI testing and observability is now available for Telnyx AI Assistants, powered by Cekura. Teams can now run pre-production simulations across diverse caller personas, define custom evaluation metrics, and monitor production conversations to test instruction-following, tool calls, and conversational quality, all from inside Mission Control.

What's new

  • Test Agents: Create caller personas with defined personalities, languages, and instructions. Each test agent is assigned a real SIP endpoint and phone number, and calls your AI Assistant like a real customer, allowing you to validate behavior before production deployment.
  • Evaluators: Define test scenarios with expected outcomes, first messages, extra instructions, max duration, and language. Run evaluators on demand or schedule them as cron jobs for continuous regression testing.
  • Custom Metrics: Create metrics to evaluate assistant quality across three eval types: Binary Workflow Adherence (true/false, failure means the run failed), Binary Qualitative (true/false, good to have), and Continuous Qualitative (graded on a scale of 5). Metrics support simulation, observability, and sampling modes.
  • Generate and Improve Evaluators: Auto-generate evaluator scenarios from your assistant configuration, and batch-improve existing evaluators with natural language instructions (for example, "add DTMF tool to all evaluators").
  • Scheduled Test Runs: Create cron jobs to run evaluators on a schedule, with personality overrides and timezone support, so regression testing happens without manual triggers.
  • Results and Call Playback: View test results with call detail cards, play back audio from test calls, and inspect transcripts to diagnose where the assistant succeeded or failed.
  • Export Evaluators: Export evaluator configurations as CSV or Excel for sharing, version control, or external review.

Why it matters

  • Voice AI teams ship agents to production without a reliable way to test them first. Cekura closes that gap: simulate calls, define what success looks like, and catch regressions before customers do.
  • Testing happens inside the same Mission Control UI where you build and deploy the assistant. No separate tool, no separate login. The Cekura integration is a proxied API layer inside the Telnyx AI Assistants service, so a Cekura project is created on demand the first time your organization uses the test framework.
  • Evaluators run on schedule, not just on demand. Cron-based regression testing means every assistant change gets validated against your test suite before it touches live traffic.
  • Custom metrics let you define quality on your terms. Whether it is "did the agent greet the caller" or "did the agent follow the tool-call workflow correctly," the eval framework adapts to your definition of a good call.

Example use cases

  • Pre-production validation: simulate 20 caller personas across 5 languages calling your support agent before deploying changes to production.
  • Regression testing: schedule a nightly cron job that runs your evaluator suite against the latest assistant config to catch instruction-following drift.
  • Quality monitoring: enable observability metrics on production calls to track tool-call accuracy, greeting adherence, and conversation quality over time.
  • Team review: export evaluator results as Excel and share with product or QA leads who do not have Mission Control access.

Getting started

  1. In Mission Control, go to AI > AI Tests > Test Agents.
  2. Click Create New Test Agent and configure a caller persona: name, language, personality, and phone number. Your assistant must have a phone number assigned first.
  3. Navigate to AI > AI Tests > Evaluators and click Create Evaluator to define a test scenario with instructions, expected outcome, and assigned test agent.
  4. Go to AI > AI Tests > Metrics and click Create Metric to define how the evaluator grades call quality (Binary Workflow, Binary Qualitative, or Continuous Qualitative).
  5. Run evaluators on demand, or create a cron job under AI > AI Tests > Scheduled Jobs to run regression tests on a schedule.
  6. Review results under AI > AI Tests > Results with audio playback and transcript inspection.