Skip to content
Vexa

Changelog

What Vexa has shipped

Every release ships with a reason. Here is the full record of what changed, why it matters, and what it changes for your test suites.

  1. v0.8AI

    Turn-level failure pinpointing and latency budgets

    Scoring now identifies the exact turn where a call lost the thread. Set a latency budget per agent and get a breach event the moment any turn exceeds it.

    • Turn-level failure view in the dashboard: click any call to see the turn that first dropped below threshold, with the scorer's reasoning alongside the transcript.
    • Latency budgets: define a maximum acceptable response latency per agent; any turn that exceeds the budget is flagged in the score and triggers a budget-breach event in webhooks.
    • Improved tone scoring model with higher inter-rater reliability on edge cases including aggressive callers, prolonged silence, and abrupt hang-ups.
    • Score history now surfaces a 30-day regression trend per metric so you can see whether turn latency or tone is drifting before it hits a gate threshold.
  2. v0.6New

    Regression alerts on every deploy with a GitHub Action gate

    Vexa now reruns your suites on every deploy and compares scores to the previous release. A regression on any voice path surfaces as an alert and can block the merge.

    • Deploy webhook endpoint: POST to /api/deploys and Vexa immediately triggers a full suite run, then scores each call against the last passing release.
    • GitHub Action (vexa-ai/test-action) published to the Actions Marketplace; add it to any workflow in three lines to gate pushes on regression score.
    • Regression alert emails and Slack notifications summarise which scenario and which voice path regressed, with a side-by-side diff of task-success and tone scores.
    • New deploy history view in the dashboard: every deploy is listed with its overall verdict and a per-voice breakdown so you can see exactly what changed.
  3. v0.4New

    Code-switching scenarios, 12 new voices, and accent expansion

    The scenario library now covers code-switching, low-bandwidth paths and fast-talker rate variation. Twelve new voice profiles added across six additional regional accents.

    • New built-in scenario suites: code-switching (English and Spanish mid-call), low-bandwidth audio with codec artifacts, and fast-talker rate variation at 1.4x to 1.8x normal cadence.
    • 12 new voice profiles added to the core library: South African English, Nigerian English, Singaporean English, Scottish English, Australian regional, and Argentine Spanish.
    • WER estimation now runs per turn, not just per call, so you can pinpoint the exact moment in the conversation where the ASR lost the thread.
    • Custom scenario builder (Pro and Scale): describe a scenario in plain text and Vexa compiles it into a caller script and adds it to your suite within seconds.

Every voice, tested.

Connect one agent and run your first suite of simulated calls in a minute, free.

Start freeRead the docs