rokan.

Rokan and Cekura

Cekura tests voice agents with simulated calls before launch, red-teams them, monitors live calls and suggests fixes for the failures it finds. Rokan works per customer after launch: why the agent failed at that customer, the fix written as an edit or a code branch, and a replay of the fix on every customer's past calls before it ships. Cekura calls your agent through its real voice stack; Rokan's replay does not. The two can be used together.

Updated . Cekura described from cekura.ai and its docs on that date.

What Cekura does

Cekura's headline is "Test, Monitor and Self Improve voice agents". Before launch it can "Run thousands of synthetic conversations before go-live", red-team for jailbreaks and off-script behaviour, and benchmark the same scenarios across providers and models. In production it offers "Live drift detection across every call". Its improvement loop "flags issues → reproduces in simulation → suggests fixes automatically".

Its Deep Research feature audits a sample of production calls and reports failure modes ranked by urgency, each with a short suggested fix and example calls (Cekura docs). Cekura connects to Vapi, Retell, ElevenLabs, Pipecat and LiveKit, or takes calls from your own backend.

If you need to test how your voice agent behaves on a real call, Cekura covers that, and Rokan does not.

What Rokan adds

  • The why, for one customer. Rokan reads one customer's failing calls with that customer's own rules, hours and config, and says why the agent failed there and not elsewhere.
  • The fix, written. A prompt or config edit, or a code change as a branch in your repo, ready for review.
  • A replay on every customer. Before the fix ships, Rokan replays it on every customer's past calls. If it would break a call that works today at another customer, it is blocked or scoped to the one customer.
  • Your approval. A person on your team approves every fix.

When Cekura alone is enough

  • You are before launch, or about to change models or providers, and need to see how the agent sounds and behaves on real calls.
  • Your failures are about latency, interruptions, speech recognition or tool calls. Those need a real call, and Rokan's replay does not make one.
  • Cekura's suggested fixes are enough for your team to apply and test across customers on its own.

When you'd add Rokan

  • The agent fails at one customer and works at the others, and the cause is in that customer's rules or setup.
  • You want the fix written, including code fixes as a branch, not only a suggestion.
  • A fix for one customer's calls keeps breaking another customer's calls.

How to send your calls to Rokan

The calls Cekura monitors usually come from your voice provider, so the simplest path is to export the same calls from Vapi or Retell and send those. Rokan reads both directly. If your calls reach Cekura from your own backend, Cekura's Calls API can list the production calls it holds (Cekura docs); save them as JSON.

Send 20 to 50 calls, from more than one customer if you can, plus your prompt and config. If Cekura already flagged some calls, include those call IDs so we start there. The steps are on the trial page.

What Rokan does not do

Rokan's replay is a model playing the caller's side against your prompt and rules. It does not call your tools or your voice stack, so it cannot catch audio, timing or transcription problems. Rokan is also early: it has no paying customers yet.