> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voxworks.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Week of 28 July 2026

> A rebuilt Dashboard with PDF export, AI script recommendations as a review workflow, voicemail modes and background classifiers.

## New features

**A brand-new Dashboard.** The [Dashboard](/dashboard/overview) has been rebuilt around a filters bar (direction, script and date range, including **Custom range…**) and four sections: **Overview** tiles with comparison against the prior period, **Performance by handler** comparing your AI assistant against your human team, **Objectives** showing how often each script's success criteria were met, and **Call outcomes** showing the mix of outcomes over time. An **Actions** menu in the header starts a new call, batch or script without leaving the page.

**Export the dashboard as a PDF.** An **Export PDF** button on the filters bar lets you tick which sections to include — **Overview metrics**, **Trend chart**, **Performance by handler**, **Objectives**, **Call outcomes** — and download a branded report with your filter settings on the cover. Your selection is remembered for next time.

**Call details now open beside the dashboard.** Clicking a row in the activity list opens the full call in a pane next to the dashboard, the same layout as the Calls page, instead of showing a hover summary. The charts reflow around it, and **View full call** takes you through to the complete record.

**AI recommendations became a real review workflow.** A wand button in the [script editor](/scripts/editor-guide) toolbar turns on **AI recommendations** review mode: every line a suggestion touches is marked across all tabs, and a floating toolbar walks you through them one at a time with the reason for the change and a coloured before-and-after diff. You can **Accept**, **Accept all**, **Dismiss** or **Undo** any decision, and nothing is committed until you save the script — so a failed save doesn't lose your review. Recommendations only ever change voice-step wording and voice-step conditions; everything else is analysed but can't be edited automatically. See [AI recommendations](/scripts/ai-generation).

**A visual editor for post-call emails.** Post-call email steps now open a full-screen **Design post-call email** editor with headings, fonts, colours, alignment, lists, links, tables, containers and padding, plus desktop and mobile previews and an HTML source view. The email step's **Format** dropdown chooses between the **Voxworks System Template** and a **Custom email**. See [Automation steps](/automations/steps).

**Drop dynamic blocks and variables into a post-call email.** A **Blocks** menu inserts ready-made **Objectives**, **Data Collection** and **Transcript** sections that fill themselves in from the call and hide when they're empty, and they can be reordered. Variables appear as removable pills you can click to change, and **Send this design to your account email** sends yourself a test using a previous call's data. The **To** field now also offers a team-member picker.

**Choose what happens when a call reaches voicemail.** Script Settings gained a **Voicemail** selector with three modes: **Hangup** (say nothing), **Leave Voicemail** (the previous behaviour) and **Leave SMS** (text the message to the contact instead). The message now has variables properly substituted, so `{{contact.first_name}}` finally works in both the spoken message and the text. Texting only applies to Australian mobiles — anything else simply hangs up. See [Voicemail message](/scripts/voicemail-message).

**Knowledge base answers can be spoken word for word.** Each question-and-answer pair now has an answer mode: **Instruct**, where the assistant phrases the answer from your response, or **Verbatim**, where your response is read out exactly as written. Knowledge is now only used when a caller's question actually matches a pair, rather than being mixed into every prompt, which keeps the assistant on script the rest of the time. See [Knowledge base](/scripts/knowledge-base).

**Classifiers — background data capture that runs alongside a step.** Voice and tool steps can now carry up to three **Classifiers**: background AI lookups that run at the same time as the step and write their answers into variables, without changing where the call goes next. On a speaking step they run as the assistant starts talking and never hold up the conversation. Only the LLM Data and LLM Custom tools can be used as classifiers. See [Steps](/flows/steps) and [LLM tools](/tools/llm/overview).

**Per-step Key Terms for speech recognition.** As well as script-level keywords, each voice step can list words and short phrases the recogniser should listen for while that step is running — useful for names, products and jargon it usually mishears. Terms share a budget with your script-level keywords, so keep each one to one to three words and only add the ones that matter. See [Keywords](/conversation-dynamics/keywords).

**Scripts can do work before the first word is spoken.** If a script begins with tool, code or sub-flow steps, those now run while the call is still connecting, and the first step that actually speaks becomes the opening line with its variables filled in. That means a script can look a caller up, check opening hours or pick a route before greeting them — and use what it found in the greeting. Previously this produced silence or an invented opening.

**Recordings can be held back when a caller declines.** If the call captures that the caller did not consent to being recorded, neither the call recording nor internal diagnostic audio is stored. See [Recording consent](/security/recording-consent). Speak to the Voxworks team for further custom configuration.

**Transcripts now say why the assistant said something.** Lines that came from a special path are labelled — a safety refusal, a knowledge base answer, an intent pathway confirmation, or a recovery after a failed check — so an unexpected reply can be traced back to its cause. The labels are for you only; the assistant never sees them. See [Call details](/calls/call-details).

## Improvements

**Script changes go live without waiting to be compiled.** Scripts are now prepared as the call starts rather than by a separate build step, so a saved edit is in effect on the very next call. See [Versions and publishing](/scripts/versions-and-publishing).

**Call time is now measured and split by who handled it.** Every completed call now carries an auditable breakdown of its total duration and how much of it was handled by the AI assistant, by a human after a handoff, and by the system, including the time spent ringing before it was answered. This is what powers the new **Performance by handler** table and makes usage reporting easier to check. Applies to calls from 22 July onwards. See [Usage](/billing/usage).

**Call recordings have a more consistent volume.** The assistant's speech now passes through a broadcast-style loudness normaliser before it leaves the system, so its voice sits at a steady level across calls, scripts and voices. Because it's applied before the phone line's own processing, it improves the live call as well as the recording.

**Better transcription on noisy and speakerphone calls.** The caller audio clean-up models have been upgraded, which noticeably improves accuracy on poor lines.

**Automations can write an outcome back onto a call and see your batches.** An automation can now record its own verdict against a call — which then appears in the calls listing and the API — and can look up your batches and list contact counts at run time. See [Automation steps](/automations/steps).