What's new in EvaliQA, October 9, 2026

Every new workspace now starts with a working demo project you can explore right away, voice evaluation runs on platform credits, and scores and costs read the same everywhere in the app.

October 9, 2026 to October 9, 2026

A ready-made demo project

New workspaces include ShopWave, a demo project with a real AI support agent, a quality plan and a red-team plan, their datasets and finished example runs to explore. Open the ShopWave site from the project to chat with the bot yourself, and add the demo back from Home if you delete it.

Voice speech on platform credits

Voice runs can now use EvaliQA's speech service on credits, so you no longer need your own speech-to-text or text-to-speech key. Speech usage shows in each call's details and in the CSV export.

Voice calls end cleanly

The test caller hangs up after saying goodbye, gives your bot time to finish its last answer, and the transcript only includes lines that were actually played.

Also shipped

  • Voice runs alongside text runs. Voice runs now appear on their test plan with text runs, including trends and run comparison. Pass rates count calls that hit the time limit by their verdict, voice reports show the bot's response latency, and Reports can be filtered by Text or Voice.
  • Context metrics skip replies without context. Faithfulness and contextual metrics now show n/a when a reply had no retrieved context, instead of failing a correct answer with a score of 0.
  • Scores and pass rates that agree. Score bars are colored against each metric's threshold, so a passing score is always green. Project Overview and Home now show a run's pass rate by rows, the same figure as the run page.
  • Eval cost in credits on the projects list. When your workspace runs on platform credits, the projects list shows eval cost in credits, and cost filters on projects and eval runs match the figures you see.
  • Simpler test plan editing. Dataset generation settings now live only in Generate dataset, edited voice plans keep their call settings, and a phone line can serve one inbound voice run at a time so calls do not mix.
  • Online evaluation and report fixes. Switching a metric to Whole session now shows the new session verdict while earlier per-trace scores stay on their traces. PDF report tables wrap long metric names and reasons cleanly.

All changelogs