> For the complete documentation index, see [llms.txt](https://docs.voiceb.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.voiceb.ai/product-updates/august-2026-voiceb.ai-product-release.md). # August 2026 — VoiceB.ai Product Release By Alex Bisbe. Sept 9th 2026 August moved VoiceB **from execution to evidence**. July gave the agent hands: tools that act on your systems, a second channel in the widget, a warm-up SMS before every outbound call. August closes the loop on trust. When an agent says a sale happened, the platform now **proves it** — the client's own web form confirms the order, the completion lands as a SOLD outcome, and every call can be audited end to end, audio and transcript side by side, without leaving VoiceB. August also raises the floor on how the agent *sounds*. Not the voice engine — that was already at the ceiling — but turn-taking: when to speak, when to wait, when a "sí" is a confirmation and not an interruption. Every ElevenLabs agent in the workspace now runs on the same natural-conversation defaults. The objective: **outcomes you can hear, verify, and count — on every channel, on every screen.** ## 🚀 Key Highlights * **Call Detail, rebuilt** — a wide, full-screen-capable drawer with the full transcript and an inline audio player, synced turn by turn. Audit any call inside VoiceB, no downloads, no third-party console. * **Form Step Tracking for Digital Journeys** — the widget now detects when a visitor completes the client's signup or contracting form, tracks each step, and maps completion to a SOLD outcome. * **Natural turn-taking for every agent** — Turn V3, soft timeouts with natural fillers, and backchannel-aware interruption handling rolled out to all ElevenLabs agents via API. * **Widget hardening** — mobile positioning and centering fixed, chat/voice choice restored after privacy consent, and repeated click-to-call sessions no longer merge transcripts. * **Async Tools from the console** — the Immediate/Async execution mode introduced in July is now configurable directly in the Tools UI. ## ✨ Call Detail — Audit a Call Without Leaving VoiceB Until now, reviewing a single call meant downloading the recording and the transcript separately, or opening the underlying voice provider's console — slow, and not something you want on screen when a client is watching. Call Detail was rebuilt around one goal: **audit a call end to end, inside VoiceB, without downloading anything.** * **Wide drawer with full-screen toggle** — Call Detail opens as a \~70–80% viewport panel from the right, with an expand-to-full-screen control. The campaign list stays visible behind it. * **Inline transcript, turn by turn** — speaker-labelled agent and user turns with relative timestamps (0:00, 0:11, 0:21…), visually differentiated. * **Inline audio player** — docked at the bottom of the drawer: play/pause, seek bar, current time / total duration, ±10s skip, and playback speed (0.5× / 1× / 1.5× / 2×). Audio streams from VoiceB's own storage via signed URL. * **Synced playback** — the transcript auto-scrolls with the audio and highlights the turn being played. Click any turn to seek the player to that moment. * **Two-column layout** — transcript on the left, metadata, summary, and collected data on the right; stacks to a single column on narrow screens. * **Prev / next navigation** — move through a campaign's filtered call list from the drawer header without closing and reopening the panel. * Download recording and Download transcript remain available. Works for inbound, outbound, and calls that ended in a transfer. Pairs naturally with [Flag calls for review →](https://docs.voiceb.ai/flag-calls-for-review) for building an audit queue. [Understanding Status & Outcomes in VoiceB →](https://docs.voiceb.ai/understanding-status-and-outcomes-in-voiceb) *Image: Call Detail drawer — transcript left, metadata right, player docked at bottom* ### Why this matters Client-facing auditing is where trust is won or lost. An enterprise operator reviewing a Vodafone España or Primagas campaign should never need a second tool — or see a vendor's logo — to check what the agent actually said. Call Detail makes VoiceB the single place where an outcome is both recorded and verified. | Before August | With the new Call Detail | | -------------------------------------------- | ---------------------------------------------------- | | Metadata only; download audio and transcript | Full transcript and audio player in one view | | "Open in provider" to hear the call | Streams from VoiceB storage — no third-party console | | One call at a time, close and reopen | Prev / next across the filtered list | | Read the transcript, then find the moment | Click a turn, the audio jumps there | ## ✨ Form Step Tracking — Proof of Completion In a Digital Journey campaign the agent accompanies a visitor through the client's own web signup or contracting form — advising, answering, guiding. Until August, the agent had no reliable way to know whether the visitor actually finished. Now the widget does. * **Form completion detection** — the widget listens to the form steps you configure and fires a "form completed" event when the visitor submits the final one. When the call ends, the event is matched to the call from the same browser session. * **Multi-step tracking** — add one row per step (HTML tag + element id); each completed stage is tracked, and the last step in the list is treated as completion. * **No code on your site** — the widget listens for submit and click events on the configured elements; only the tag and id of each step are needed. * **Completion → SOLD** — in Digital Journey campaigns, a matched completion marks the lead as converted in the Campaign Dashboard. In other campaign types the event is logged but does not change the outcome. Why it matters: this is outcome-based pricing made literal. The sale is no longer inferred from the conversation — it is confirmed by the visitor's own action on the client's own site, and counted as such. Reliability of completion detection was prioritised over partial heuristics. [Form Step Tracking →](https://docs.voiceb.ai/widget-configuration-guide/form-step-tracking) ![Form Step Tracking configuration in the campaign's Widget settings](https://content.gitbook.com/content/ggIr2RvhGm49D4vbjurT/blobs/92ZWIu2aeuIN7U8z3y6Y/Screenshot%202026%2008%2018%20at%2010.48.03.png) ## ✨ Natural Turn-Taking — Applied to Every Agent Voice quality was already at the platform ceiling. The remaining gap between "a good voice" and "a natural conversation" lives in turn-taking: awkward silences during a tool call, the agent cutting itself off because the customer said "vale". August closes that gap across the whole workspace, via API, not agent by agent. * **Turn model V2 → V3** — contextual end-of-turn detection, reducing clumsy interruptions and unnatural pauses. The single highest-impact change to perceived naturalness. * **Soft timeout enabled** — when the model is slow to respond (typically mid tool call), the agent fills the wait with a natural phrase instead of dead air. * **Backchannel-aware interruptions** — "vale", "ajá", "claro", "sí", "mmm" are now treated as the customer listening, not interrupting. The agent finishes its sentence. * **New defaults** — every newly created ElevenLabs agent inherits Turn V3, soft timeout, and the interruption presets automatically. [Memory — Conversation Recall →](https://docs.voiceb.ai/memory-conversation-recall) · [VoiceB Prompt & Script Guide →](https://docs.voiceb.ai/product-guides/voiceb-prompt-and-script-guide) Background-speech filtering was evaluated but held back pending validation on real noisy calls (street, car), as it can clip quietly spoken sentence openings. Changes were tested on a non-production agent before rollout, with latency, conversation duration, and transfer-to-human monitored post-release. ## 🔧 Improvements & Fixes ### Widget * **Mobile positioning** — the widget launcher now anchors bottom-right on mobile as configured (previously pinned to the left edge). * **Mobile modal centering** — the click-to-call modal is horizontally centered on \~390px viewports (iOS Safari) instead of overflowing left. * **Chat/voice choice after consent** — after accepting the privacy policy, the "Call instead" option is accessible again alongside chat; buttons are spaced correctly and the "Call instead" label is now editable. * **Session grouping fix** — repeating a click-to-call from the same open widget no longer merges separate calls into one transcript; session legs are now restricted to VoIP calls. [Widget Configuration Guide →](https://docs.voiceb.ai/widget-configuration-guide) · [Web Widget & WebRTC VoIP Channel →](https://docs.voiceb.ai/widget-configuration-guide/new-voiceb-web-widget-and-webrtc-voip-channel) ### Tools & agents * **Async mode configurable in the UI** — the Immediate/Async execution mode shipped in July can now be set per tool directly in the Tools console. [Tool execution modes →](https://docs.voiceb.ai/tools-connect-your-agent-to-external-actions/tool-execution-modes) * **Tool configuration errors surfaced** — tool-config failures now show the real error message instead of a minified React error, so misconfigurations are diagnosable in the console. * **Click to Call dynamic variables** — `agent_phone_number` and `customer_phone_number` are now injected on Click to Call, matching the inbound behaviour shipped in July, and protected as built-ins. [Using Variables in Click to Call →](https://docs.voiceb.ai/using-variables-in-click-to-call-to-personalize-conversations) ### Agent Studio * **GPT models updated** — two current GPT options are available alongside Gemini Flash (default) and Claude Sonnet. [Agent Studio →](https://docs.voiceb.ai/agent-studio) * **Prompt Admin error reporting** — configuration errors surface with their real cause instead of a redacted client error. ### Platform & security * **Dependency refresh across all apps** — dashboard, outbound app, and voice widget bumped to current dependency sets (Next.js, React, Clerk, Vite, ElevenLabs and Twilio SDKs, Axios, type definitions), resolving known security advisories. Part of VoiceB's ongoing security hardening program. [Data Journey, Storage & Security →](https://docs.voiceb.ai/legal-and-compliance/data-journey-storage-and-security) ## Final Takeaway August was about **evidence**. [May](https://docs.voiceb.ai/product-updates/may-2026-voiceb.ai-product-release) made sure every lead gets contacted. [June](https://docs.voiceb.ai/product-updates/june-2026-voiceb.ai-product-release) put the operation in the operator's hands. [July](https://docs.voiceb.ai/product-updates/july-2026-voiceb.ai-product-release) gave the agent hands of its own. August makes the result verifiable: the form confirms the sale, the drawer plays back the call, the transcript follows the audio, and the agent that produced it sounds like someone you would actually stay on the line with. An outcome you can't audit is a claim. An outcome you can hear, click through, and match to a completed form is a fact. From August, VoiceB deals in facts. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.voiceb.ai/product-updates/august-2026-voiceb.ai-product-release.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.