Product Updates

📞 WhatsApp Voice Calls

This is the big one. Your assistant can now take and place real voice calls over WhatsApp, on the same numbers you already use for messaging. Your customer taps the call button in the chat they already have with you, and the assistant picks up. No phone dialling, no roaming charges on their side, no separate app.

• Customers calling you: anyone can call your WhatsApp number at any time and your assistant answers automatically, day or night.
• You calling customers: WhatsApp asks for consent first. Send an Ask to call request inside the conversation, and once the customer taps Allow, you can call them for the next 7 days.
• One toggle per number: open a WhatsApp sender, switch calling on, and the platform sets everything up for you. Before it turns on, you see exactly which assistant will be answering.
• Same call, same tools: WhatsApp calls behave like any other call. Recording, transcript, post-call analysis, tools, and knowledge base all work the way they already do.
• Everything in one place: WhatsApp calls show up in your calls list with their own icon, and the call also appears inside the chat thread, so the message history and the call history read as one conversation.
• It tells you when it can't: if a contact isn't callable yet, or a number isn't eligible for calling, you get a plain-language reason instead of a silent failure.

Where to find it: Channels → WhatsApp, open a sender, then Enable calling. There's a built-in WhatsApp Calling guide on the same page covering the rules below.

🎙️ Import Voices Straight from the Voice Libraries

You're no longer limited to the voices already on the list. You can now browse the full voice libraries of the built-in voice providers and import any voice you like into your account with one click. It then shows up in the voice picker for every assistant, like all the others.

💬 Channels and Inbox

• Ignore message reactions: a new per-number setting so a thumbs-up on your message doesn't wake the assistant up and trigger a reply.
• Web widget stays put: the widget settings panel no longer closes on you every time you save.
• Group widgets by campaign: custom dashboard widgets can now filter and group by campaign name.
• Instagram setup instructions: connecting an Instagram account now comes with step-by-step instructions in the app, plus a fix for connection screens that failed to load on some accounts.
• Mobile menu fix: opening plans or credits on a phone now closes the side menu behind it.

🛠️ Improvements and Fixes

• Deleting a lot of records at once: large deletions of calls, leads, campaigns, or conversations now run in the background instead of making you wait on a loading screen or timing out.
• Warm transfers, three fixes: the assistant no longer introduces the caller twice, the briefing to the person receiving the call is no longer skipped, and recordings of transferred calls are now saved correctly.
• Bringing your own number: importing a number no longer fails when a matching setup already exists at your carrier.

🤖 Connect AI Now Builds Your Automations

Connect AI already let you build and fix assistants by chatting with ChatGPT or Claude. It can now build your automations too: the workflows that run before, during and after a call.

You describe what should happen in plain language. The AI picks the trigger, wires the steps, finds your sheets, campaigns and numbers by name, runs a real test, and switches the automation on only if it passes. No builder screens, no field names, no IDs to copy.

💬 What That Looks Like

Things you can now ask for, word for word:

• "After every call, if the caller was qualified, add a row to my Google Sheet with their name, phone and summary."
• "If a customer asks to be called back later, call them back at the time they asked for."
• "Give me a link my website form can post leads to, and call every new lead automatically."
• "During a call, let the assistant look up an order number in my system and read the status back to the caller."
• "Let callers cancel or reschedule their appointment while they're on the phone."
• "Before the assistant offers a discount, text my supervisor and only continue once they approve."

Each one becomes a live, tested automation attached to the assistant you name.

📚 17 Ready-Made Automations

The AI checks a library of proven automations first, so common requests are set up in seconds instead of built from scratch:

• Save qualified leads to a spreadsheet, or log every call to your CRM
• Call the customer back at the exact time they asked for
• Hot lead alerts, SMS or WhatsApp follow-ups, and transcript emails after a call
• Call new leads automatically from a web form, ad platform or spreadsheet
• Confirm or remind before a booking, and cancel or reschedule mid-call
• Recognise returning callers and greet them by name
• Ask a human for approval before the assistant does something sensitive

Start from one of these, then change it by chatting.

🔗 Connects to What You Already Use

Spreadsheets, email, calendars, chat apps, CRMs and hundreds of other services are available as steps. Anything without a ready-made connector is still reachable with a plain web request, so your own internal tools work the same way.

Accounts are connected once on the Automate page under Connections. After that the AI picks your sheet, your pipeline or your page by name from what's actually in your account.

🛡️ It Tests Before It Goes Live

An automation that looks right but quietly does nothing is worse than no automation. So nothing switches on until it has actually run.

• A real test run, not a preview: the automation is executed end to end with a realistic sample of your own data. Only a green run gets attached to your assistant.
• You approve anything that really sends: if the test would send a message, place a call or send an email for real, you're told first and asked to confirm, using your own number or address.
• Safe repairs: if a change to a live automation fails its test, the previous working version is restored automatically so the automation keeps running. You can roll back any change yourself too.
• No silent overwrites: an assistant that already has an automation for the same event won't get it quietly replaced.

🔍 Fix What's Already Running

Ask what went wrong and the AI reads the execution history for you: every run, what each step received and returned, and exactly which one failed and why. Once the cause is fixed, the run can resume from the broken step, so steps that already succeeded are never repeated and nothing gets sent twice. Repairs change only what needs changing and leave the rest of your setup intact.

🛠️ Also Worth Knowing

• Available on your API too: if you'd rather build from your own code than by chatting, the same create, test, update, roll back and run history operations are exposed as endpoints.

Where to find it: open Connect AI from the top bar to link your account, then just ask. Everything the AI builds also appears on your Automate page, where you can open and edit it visually at any time.

💬 Facebook & Instagram Messages, 📣 SMS + WhatsApp Campaigns, 🔐 Two-Factor Authentication

💬 Facebook Messenger and Instagram Messages

Your assistant can now answer people who write to you on Facebook Messenger and Instagram, the same way it already handles WhatsApp and web chat.

• Connect in a few clicks: link your Facebook Page or Instagram account, pick the assistant that should reply, and you're live.
• Everything in one inbox: Messenger and Instagram conversations land in the same inbox as your other channels. You can read the full thread, jump in yourself, and take over from the assistant at any point.
• Open in Meta: one click takes you straight to the original conversation on Facebook or Instagram when you need to handle something there.
• Connection health at a glance: each connection shows its current status, and reconnecting after a password or permission change is a single button.

🔐 Two-Factor Authentication and a Real Security Center

Your account now has proper protection against someone getting in with just a stolen password.

• Two-factor authentication: turn on a second step at login using an authenticator app, or get a one-time code by email if you'd rather not install anything.
• Trusted devices: mark the computer you use every day as trusted so you're not asked for a code on every login. Remove a device any time and it starts asking again.
• Active sessions: see every browser and device currently signed in to your account, with location and last activity, and sign any of them out with one click.
• Security activity: a running log of logins, password changes, and security settings changes, so you can spot anything you don't recognise.
• Identity confirmation: sensitive actions now ask you to confirm it's really you before going through.

Where to find it: the new Security page in your account settings.

📣 Campaigns Now Send Text Messages Too

Campaigns are no longer calls only. You can now run the same lead lists through SMS and WhatsApp.

• Pick a channel: when you create a campaign, choose whether it places calls, sends WhatsApp messages, or sends SMS.
• Schedule it in advance: set a start date and time and the campaign switches itself on when the moment arrives. No need to be at your desk.
• Track every message: sent messages and replies show up on the campaign and on each lead, so you can see exactly what went out and what came back.
• Better date and time picker: choosing when a campaign runs is now clearer and harder to get wrong.

🧭 New Channels Hub

All your messaging channels now live in one place. Voice, WhatsApp, SMS, web chat, Messenger, and Instagram are grouped under a single Channels page, each with its own colour and status, so connecting a new one or checking what's already active takes seconds instead of hunting through menus.

🧠 New AI Models: GPT-5.6 (Luna & Terra)

Your assistants can now run on OpenAI's newest GPT-5.6 models, the latest generation of the AI that understands each request and decides how your assistant responds. Expect sharper reasoning, better instruction-following, and more natural handling of longer, multi-step conversations.

Available in two variants, GPT-5.6 Luna and GPT-5.6 Terra. Both are live now in pipeline mode. Pick either one from the model list when you create or edit an assistant.

🔇 Turn Off Interruptions in Speech-to-Speech and Dualplex

The Allow interruptions setting now works in every conversation mode, not just Pipeline.

Until now, Speech-to-speech and Dualplex assistants were always interruptible: if the caller made any noise while the assistant was talking, it would stop mid-sentence. You can now switch interruptions off in those modes too, so the assistant finishes what it's saying without being cut short by background noise, TV sound, or someone talking in the room.

Where to find it: Assistant settings, Allow interruptions? toggle. Leave it on for natural back-and-forth conversations. Turn it off when the assistant has to deliver something in full, like disclaimers, confirmations, opening hours, or a list of options.

🧠 GPT Realtime 2.1 & 2.1 Mini: Now in Speech-to-Speech & Dualplex & WhatsApp Media

🧠 New Voice Models: GPT Realtime 2.1 & 2.1 Mini

Your speech-to-speech and Dualplex™ assistants can now run on OpenAI's newest voice models. GPT Realtime 2.1 follows your prompt more reliably and sounds more natural on names, numbers, and interruptions. GPT Realtime 2.1 Mini is a lighter, faster option for high-volume or simpler assistants.

Both are live now. Pick either one from the model list in any speech-to-speech or Dualplex™ assistant.

💬 Send Media on WhatsApp

You can now send images, audio, video, and documents in your WhatsApp messages, not just text. Add a caption or filename and send straight from the inbox or the API.

🚀 Build & Manage Your Assistants Right Inside ChatGPT & Claude

This is a big one. You can now connect your account to ChatGPT and Claude and run your whole AI calling setup just by chatting — no dashboards, no setup screens. Tell the AI what you want in plain language and it does the work for you, from start to finish.

What you can do once it's connected:

• Build a new assistant from scratch — Describe what you need ("an assistant that books appointments") and the AI creates it for you: it writes the prompt, picks a natural voice, sets a greeting, and even adds the tools the assistant needs, like checking a calendar and booking slots.
• Fix and improve an existing assistant — Ask it to review an assistant's prompt and settings. It points out what's weak and applies the fixes for you.
• Analyze your calls — "Look at my last call and tell me what went wrong." It reads the full transcript and the call evaluation, explains what to improve, then updates the assistant to match.
• Learn from your history — Ask it to study your last 30 calls, spot patterns, and tune your assistant so it performs better next time.
• Launch calling campaigns — It walks you through setting up an outbound campaign from your leads and starts it for you.
• Handle the day-to-day — Manage phone numbers, leads, and knowledge bases, or send a message — all from the same chat.

🔗 How to connect it (about a minute):

• Open the new Connect AI button in your dashboard and copy your secure connection link.
• In ChatGPT or Claude, go to the connectors settings and choose Add custom connector, then paste the link. A built-in, step-by-step guide with screenshots walks you through both apps.
• Sign in to your account and approve. Your tools appear in the AI app, and you're connected.

It's secure by design: the connection only works after you sign in and approve it with your own account, and every action runs with your existing permissions. Nothing happens that you haven't logged in for.

Try it: open a fresh chat and type "Create an appointment-booking assistant." Within seconds the AI has built it, written the prompt, given it a voice and a greeting, wired up calendar booking, and offered to place a test call.

⚡ Reliability & Fixes

• Smoother call transfers — More reliable transfers between assistants, including when an assistant hands off to book a meeting.
• Faster chat under load — Increased capacity for handling chat conversations during busy periods.
• Independent dashboard tables — Tables on the dashboard no longer interfere with each other when you page through them.

🔌 Connect External Tools with MCP, 🛠 Redesigned Mid Call Tools, ⚡ One-Click Automation Linking

🔌 Connect External Tools to Your Assistant (MCP)

Your assistant can now plug into outside tools and live data through MCP (Model Context Protocol), the open standard for connecting AI to external systems. Instead of building tools one at a time, you connect a whole toolset in a single step. Point your assistant at a compatible server (a CRM, your internal APIs, a knowledge tool) and it can look things up and take action right in the middle of a conversation, on voice calls, the web widget, and chat alike.

• Add a server in minutes, no code — Paste the server address and (if needed) an authentication header. There's nothing to install and nothing to build
• Tools discovered automatically — The moment you connect, every tool the server offers is detected and listed, with its description and parameters, so you can see exactly what your assistant can do
• Pick precisely what's exposed — Turn individual tools on or off per server, so an assistant only gets the abilities you want it to have
• Connect once, reuse everywhere — Assign one server to a single assistant or to many, and manage them all from a dedicated page
• Always know the status — Each server shows a live "connected" indicator, lists its available tools, and gives a clear message (with a one-click re-test) if anything needs attention
• Secure by design — Authentication tokens are stored encrypted, and you stay in control of which tools and permissions each assistant can reach

Popular uses include connecting a CRM to look up contacts and create deals mid-call, querying your own systems for live order or account data, or wiring in a documentation tool so the assistant can answer detailed questions accurately.

🛠 Redesigned Call Actions — Easier to Build, Easier to Understand

The tools that let your assistant do things during a conversation (look up an order, send an SMS, book an appointment, check availability, submit a form) got a full redesign. The whole experience is now built so that anyone can set up a working action, even with no technical background and no coding.

• Visual flow card — A new step-by-step layout shows exactly how an action works, from the information the assistant collects to the response it gets back, all in one clear view
• "How it works" guidance built in — Plain-language explanations sit right next to each setting, so you always understand what an action does and when your assistant will reach for it
• System variables at your fingertips — Drop in live details like the caller's phone number or the current date and time straight into your action, with a handy reference so you never have to guess the right name
• One home for everything — Custom actions and connected MCP servers now live together on a single, redesigned Tools page that's far easier to work with

⚡ Build Call Actions with No-Code Automations

This is the big one. You no longer need an API or a developer to give your assistant a custom action. When you create an action, you can now choose the Automation Platform option, and the platform builds the entire technical connection for you.

• The connection is created automatically — Pick "Automation Platform" and a linked automation is generated and wired up for you instantly. The web address that connects it to your assistant is set and locked behind the scenes, so there are no IDs to copy, paste, or keep in sync
• Build the logic visually — Open the linked automation and drag together whatever steps you need: look up a record, check a calendar, send a message, run several actions in a row. When you're done, you simply finish with a "Return Response" step to hand the result back to the assistant on the call
• Multi-step workflows made simple — Anything that used to require custom code (chaining several systems together, transforming data, conditional logic) can now be assembled with no code at all
• Smart linking for existing automations — If you paste in an automation you've already built, the platform recognizes the workflow on its own and pulls in its details automatically, so connecting it takes seconds

The result: from "the assistant should be able to do X" to a working, live action in minutes, entirely point-and-click.

🎙 Voice & Call Reliability

• Cleaner call endings — Fixed a case where the assistant could repeat itself or hesitate when wrapping up a call
• Voice picker polish — Follow-up fixes to the new voice picker for smoother selecting

📁 Assistant Folders & Labels, New Voice Picker, WhatsApp Human Takeover

🗂 Assistant Folders & Labels — Organize Your Workspace

If you manage more than a handful of assistants, finding the right one just got much easier. You can now organize assistants into folders and tag them with color-coded labels.

• Folder tabs — The assistants list now shows tabs for each folder (plus an "All" view), with a searchable dropdown when you have many folders
• Labels for cross-cutting tags — Each assistant can carry multiple labels (for example "Production", "Testing", "Client X"), shown as badges on the assistant cards and filterable from the list
• Bulk actions — Select multiple assistants and move them to a folder or add/remove labels in one go
• Create on the fly — New folders and labels can be created right from the assistants page, no detour through settings
• Full API support — Folders and labels can be managed via the API, and assistants can be assigned on create or update

🎙 New Voice Picker — Find the Perfect Voice in Seconds

The voice dropdown is gone. Choosing a voice now opens a full picker built for browsing hundreds of options without scrolling through an endless list.

• Filter by gender, age, and accent — Narrow the catalog to exactly the kind of voice you're looking for, with free-text search on top
• Provider tabs — Voices are grouped by speech engine, and each card shows a provider badge so you always know what you're selecting
• Language-aware — The picker follows your assistant's language and warns you before selecting a voice that doesn't support it
• Clone voices in place — The voice cloning option now lives inside the picker, and deleting a cloned voice uses a proper styled confirmation
• Smarter syncing — Selecting a voice automatically sets the matching speech engine on the assistant, so there's nothing to configure twice

🏷 Phone Number Nicknames

Phone numbers can now have a nickname and a color, so "the support line" stops being "+1 555 0182 …which one was that again?"

• Visible everywhere — Nicknames show up in the phone numbers list, assistant cards, call details, SMS history, and every place you select a number (assistant setup, warm transfers, campaign pools)
• Searchable selectors — Find a number by its nickname or by the number itself
• API support — Nicknames are included in all phone number responses and can be set via the API, including for SIP trunks

🤝 WhatsApp Coexistence — Automatic Human Takeover

Coexistence mode gets its biggest quality-of-life upgrade yet: when someone on your team replies to a customer directly from the WhatsApp Business app, the AI now detects it and automatically steps aside for that conversation.

• No more talking over each other — The moment a human takes over from the phone, the AI stops responding in that chat
• Clear audit trail — A system note appears in the conversation explaining why the AI paused, so everyone knows what happened
• Voice notes & images fixed — Audio messages and images from customers now come through reliably in Coexistence mode
• No charges for synced history — When you connect an existing WhatsApp account, old conversations that stream in during setup are never billed

🔔 Webhook & Integration Upgrades

The conversation ended webhook got a major upgrade for anyone building automations on top of chat conversations.

• Media attachments included — Images, audio, and files exchanged in the conversation are now part of the webhook payload
• Conversation ID & timestamps — The payload now carries the conversation ID and per-message timestamps (human-readable and unix), so transcripts can be reconstructed precisely
• Re-send with one click — Every conversation now has a "Re-send webhook" action, and the delivery result (status and response) is stored so you can see exactly what your endpoint received
• Bulk actions — Re-evaluate or re-send webhooks for many conversations at once, straight from the conversations list

🛠 Other Improvements & Fixes

• No more technical data read out loud — In rare cases the AI could speak fragments of raw technical output mid-call when using tools; this is now stripped from both speech and transcripts
• Day separators in conversations — Long chat histories now show date dividers, making it much easier to scan past conversations
• Chat widget button fix — The small launcher button no longer appears clipped on some sites
• Webhook reliability — Fixed a timing issue that could cause the conversation ended webhook to silently skip for some WhatsApp chats
• Template sender display — WhatsApp templates now show their sender more clearly in the templates list

👥 WhatsApp Coexistence, Smart Message Buffering, Engine upgrades

💬 WhatsApp Coexistence — Keep Using the WhatsApp Business App

You can now connect a WhatsApp sender in Coexistence mode, which lets your team keep using the official WhatsApp Business app on their phones at the same time as the AI assistant. No need to migrate your existing chats, contacts, or labels — everything stays where it is, and the AI handles new conversations in parallel.

This is a game-changer for businesses that already have an active WhatsApp Business setup and don't want to disrupt how their team works today.

⏱ Smart Message Buffering for WhatsApp

Real customers don't send one perfectly composed message — they send three quick texts, an image, then a follow-up question. Until now, the AI would reply to each one separately, creating a clumsy back-and-forth. With the new response delay setting, the AI now waits a configurable number of seconds (0–300) before replying, giving customers time to finish their thought.

• Configurable per assistant — Set the response delay individually on each assistant in the assistant form. Leave it at zero for the original instant-reply behavior

📞 New Outbound Call Limits Page

To stay aligned with carrier compliance regulations on shared infrastructure numbers, the new Limits page (now in the user menu, next to "My profile") shows your current daily outbound call allowance, your usage so far today, and your trust tier at a glance.

• Inbound and SIP trunk calls are unlimited — The limits only apply to outbound calls placed through shared platform numbers; bring-your-own carrier accounts are not affected
• Request a higher limit — A request workflow is available directly from the page; verified businesses can be upgraded to higher tiers
• Bring your own carrier for unlimited outbound — A prominent banner links to the BYO Telephony wizard for teams that want full control over their outbound volume
• Smart campaign handling — Campaigns that hit the limit aren't paused or broken; they automatically resume the next day where they left off

🚀 Voice Engine Upgrade — Across-the-Board Quality & Stability

The underlying voice engine has been upgraded. Customers should notice smoother calls, fewer mid-call hiccups, and faster response times across every voice mode.

🛠 Other Improvements & Fixes

• Delete campaigns via API — A new DELETE /campaign/{id} endpoint lets you remove campaigns programmatically; running campaigns are stopped automatically before deletion
• Assistant transfer over API — The assistant_transfer tool now persists correctly when configured via the assistant API, and self-transfers (an assistant trying to transfer to itself) are blocked
• Mid-call tool test button — Fixed the test button on mid-call tool configuration so it sends the correct HTTP method (GET, POST, etc.) instead of always defaulting

🎙 OpenAI Realtime 2, GPT 5.5, Cartesia 3.5, Bring Your Own Telephony 1 click integration and more

🧠 GPT Realtime 2 — Better Prompt Following

Our speech-to-speech and Dualplex mode now includes OpenAI's GPT Realtime 2. The biggest win: the assistant follows your prompt much more reliably — staying on script, respecting your instructions, and handling complex multi-step conversations without drifting.

Combined with our proprietary Dualplex™ technology, this delivers the best voice quality, lowest latency, and most reliable instruction-following available on any AI calling platform today.

🎙 Cartesia Sonic 3.5 — Voice Engine Upgrade

The Cartesia voice engine has been upgraded to Sonic 3.5, the latest version. Expect noticeably more natural-sounding voices, better stability across long calls, and improved consistency on tricky words and names.

The upgrade applies automatically to every assistant using Cartesia voices — no configuration changes needed.

⚡️ GPT-5.5 — New Language Model in Pipeline Mode

You can now select GPT-5.5 as the language model for your assistants in pipeline mode. It's a great all-rounder: fast responses, strong reasoning, and excellent prompt-following — ideal for assistants that need both speed and intelligence without the cost of the flagship models.

Available right now in your assistant's pipeline mode configuration.

📞 Bring Your Own Telephony — Self-Service Carrier Imports

You can now connect your own carrier accounts directly through a guided import wizard — no support ticket, no manual SIP setup. Teams that already have negotiated rates with their preferred provider can plug those accounts in and start dialing in minutes.

• Twilio SIP trunks — Full self-service import with a step-by-step onboarding help panel that walks you through each setting
• Telnyx support — Connect Telnyx accounts using the same wizard experience as Twilio
• Zadarma & DIDLogic — Two more carriers added to the picker, giving you more flexibility on regions and pricing

🎤 Post-Call Re-Transcription for Dualplex™ & Multimodal Calls

You can now automatically re-transcribe completed calls from the original recording at any time. Useful when you want to retry transcription with the most capable STT model.

Re-transcription is available for both Dualplex™ and multimodal assistants, with clear billing transparency built right into the assistant view.

📂 Split-View Detail Panels — Calls & Conversations

The calls and conversations list pages now have a side-by-side detail panel. Click any record on the left, see the full details on the right — no more clicking back and forth. Reviewing dozens of calls or chats is suddenly a much smoother experience.

• Performance indexes added — Navigating between records inside the panel is significantly faster, even on accounts with very high conversation volumes
• Extracted variables section — Variables collected during the conversation are now displayed in a dedicated section on the conversation detail view

💬 Disable & Enable AI Per Conversation

Two new API endpoints let you pause and resume the AI for a specific conversation programmatically — perfect for human handoffs in WhatsApp and chat workflows. When the AI is disabled, your agents take over; re-enable it whenever you're ready to hand back.

💸 Auto Top-Up for Chat Credits

Chat credits can now top up automatically when balances run low, with email notifications keeping you informed of every refill. No more interrupted chat experiences because credits ran dry overnight.

🔄 Telephony & Call Flow Improvements

• Initial message on AI-to-AI transfers — When a call is transferred to an AI agent, the agent now speaks its initial message immediately for a smoother handoff
• Transfer caller details visible — Transfer-to and client phone numbers are now displayed on transfer calls in the call view
• Single-character tool parameter names — Tool parameters can now use single-character names, useful for compact JSON schemas

🛠 Other Improvements & Fixes

• WhatsApp pre-verified number deletion — Removing a pre-verified WhatsApp number now works correctly
• Phone number deletion — Phone number deletion jobs now run reliably, with proper logging when something goes wrong

🧠 Gemini 3.1 Flash Live, GPT Realtime 1.5 in Dualplex

🧠 Gemini 3.1 Flash Live — New Speech-to-Speech Model

You can now run voice calls using Google's Gemini 3.1 Flash Live in speech-to-speech mode. This gives you another top-tier real-time voice option with excellent multilingual support and fast response times.

⚡ GPT Realtime 1.5 — Now with Dualplex™

GPT Realtime 1.5 now fully supports Dualplex™ mode — our proprietary technology for the most natural-sounding AI phone calls. This means you get OpenAI's latest real-time model combined with superior interruption handling, smoother turn-taking, and lower perceived latency.

📊 Per-Turn Latency Metrics

Every conversation now tracks detailed latency metrics for each turn — how long the assistant took to respond, broken down by processing step. Use this to spot slow turns, compare model performance, and fine-tune your setup for the fastest possible calls.

📈 Custom Dashboard Widgets

Build your own dashboard with customizable stat widgets. Choose what data to display, apply percentage calculations, filter by time of day, and format with prefixes, suffixes, and abbreviations — all without writing code.

🛠️ Other Improvements

• Cleaner transcripts — Control tokens and SSML tags are now automatically stripped from call transcripts for a cleaner read, with an optional toggle to show them for debugging
• Soniox for WhatsApp voice notes — WhatsApp audio message transcription now uses Soniox for better accuracy
• Widget auto-open fix — The web widget now remembers when a visitor closes it and won't pop up again during the same session

🔀 Transfer to AI Assistant — Mid-Call Handoffs Between Specialized Agents

Your AI voice agent can now transfer a live call to a different AI assistant — mid-conversation, with zero interruption. The target assistant takes over with its own voice, language model, and transcription provider, so each step of the call gets the right specialist.Why this matters: Instead of building one massive assistant that handles everything, you can now create specialized agents for each stage of a call — a sales qualifier that hands off to a technical expert, a receptionist that routes to a scheduling specialist, or a triage agent that escalates to a support specialist.

• Seamless mid-call handoff — the caller stays on the line while the conversation transfers to a different AI assistant with full context
• Full provider swap — each assistant can use its own voice (ElevenLabs, Cartesia), language model (GPT-4o, GPT-5.4), and transcription engine
• Rich assistant selector — choose the target assistant from a dropdown showing avatar, language, voice, and mode at a glance
• Works with the Flow Builder — add assistant transfers as actions in your visual call flows alongside warm transfers, DTMF, and keypad collection

Example use case: A dental clinic's receptionist AI answers the call, qualifies the patient, then transfers to a scheduling assistant that speaks the patient's language and books directly into the calendar — all in one seamless call.

📞 Warm Call Transfers — Now with Variable Injection

Warm transfers just got smarter. When your AI agent transfers a call to a human supervisor, it can now inject call variables — like the caller's name, reason for calling, or qualification score — into the briefing. The supervisor picks up already knowing who's on the line and why.

🧠 GPT-5.4 Mini & Nano Models Available

Two new language models are now available for your AI assistants:

• GPT-5.4 Mini — faster responses at lower cost, ideal for simple call flows like appointment booking and FAQ handling
• GPT-5.4 Nano — the fastest option for high-volume, low-complexity calls where speed matters most

Select them from the Language Model dropdown when configuring your assistant. All existing prompts and tools work without changes.

🔧 Flow Builder Improvements

The visual flow builder got several quality-of-life upgrades:

• New action types — warm transfer, assistant transfer, DTMF input, and collect keypad are now available as flow actions
• Smarter action dropdown — the dropdown now shows which tools are configured on your assistant, with warnings for unconfigured ones
• Edge delete on hover — remove connections between nodes by hovering over the edge and clicking the delete button
• Import/export support — save your flows as files and import them into other assistants or accounts
• UI fixes — wider edit panel, fixed dropdown clipping, smoother animations

✨ AI Assistant Advisor

Your AI assistants now have their own personal optimization coach. The new AI Advisor automatically analyzes your assistant's configuration and call performance, then gives you clear recommendations to improve call quality — with one-click fixes.

Open any assistant and you'll see a smart banner telling you exactly how many settings could be improved. Click "View & fix" to open the full advisor panel.

💡 How It Works

The advisor runs 41 intelligent rules across three categories:

• Recommendations — Settings that aren't optimal for your use case. Things like voice speed, temperature values, silence thresholds, and endpoint sensitivity. Each recommendation explains why it matters and what to change.
• Troubleshoot — Experiencing issues on calls? Echo, interruptions, the AI not following your prompt, calls cutting off too early? Select the problem and the advisor shows exactly which settings to adjust.
• Best Settings — A curated guide to the optimal configuration for each voice mode (Pipeline, Multimodal, and Dualplex).

🔧 One-Click Fix

Every recommendation comes with a "Fix" button that automatically applies the optimal value — no manual editing needed. You can also hit "Fix All" to apply every recommendation at once. The advisor updates in real time as you make changes.

🆕 New models: Soniox STT, Realtime 1.5, GPT 5.4 & 5.3

🎙️ Soniox — New Speech-to-Text Provider in Pipeline Mode

We've added Soniox as a new speech-to-text engine available in pipeline mode. Soniox delivers improved transcription accuracy across multiple languages with noticeably lower latency, meaning your voice assistants understand callers faster and respond more naturally.

You can select Soniox as your transcription provider when configuring any assistant in pipeline mode — give it a try and compare the results with your current setup.

⚡ GPT Realtime 1.5 — Speech-to-Speech mode Upgrade

Our speech-to-speech mode now runs on OpenAI's GPT Realtime 1.5, bringing major improvements across the board:

• Better call quality — More natural, fluid conversations with improved voice output
• Lower latency — Faster responses for a smoother caller experience
• Improved prompt following — The assistant stays on script more reliably and handles complex instructions better

Select GPT Realtime 1.5 in your assistant's speech-to-speech mode configuration and hear the difference.

🧠 GPT-5.4 & GPT-5.3 — Latest OpenAI Models Now Available in Pipeline Mode

You can now select GPT-5.4 and GPT-5.3 as the language model for your assistants in pipeline mode — OpenAI's newest and most capable models.

• GPT-5.4 — OpenAI's most factual model yet, with 33% fewer factual errors than its predecessor. Significantly better at following complex instructions, staying on script, and handling multi-step conversations. Your assistants will give more accurate, reliable answers and follow your prompts more precisely.
• GPT-5.3 — A strong all-rounder with excellent speed and reasoning balance. Great for assistants that need fast, smart responses.

Both models are available right now — just select them in your assistant's pipeline mode configuration.

🛠️ Other Improvements

• AEC warmup — A new parameter that pre-warms the acoustic echo cancellation, reducing audio artifacts at the start of calls
• Expanded API — New endpoints for campaigns, WhatsApp, warm transfers, keypad collection, filler phrases, voice settings, and provider discovery — build your own integrations with full programmatic control
• Voice preview fix — Cartesia voice previews now play correctly
• Web widget improvements — Fixed mobile height and form input collection on auto-open

🚀 WhatsApp Files & Audio, Cal.com EU + Teams, GDPR Data Retention, Chat Webhooks & Widget Customization

🔥 WhatsApp Gets Smarter: Vision, Audio Transcription & Media Attachments

Your WhatsApp AI agents just got a massive upgrade: they can now see images, transcribe voice messages, and handle file attachments from your customers.

📸 Vision Support — When a customer sends an image via WhatsApp, your AI agent can now analyze it using vision-capable LLMs (OpenAI, Claude, Gemini) and respond intelligently. Perfect for reading documents, analyzing property photos, diagnosing issues from images, or identifying products.

🎙️ Automatic Voice Message Transcription — Customers who send WhatsApp voice notes no longer get ignored. Agents now automatically transcribe audio messages based on your assistant's language settings.

📎 Full Media Attachment Support — All inbound media (images, audio, video, documents) are now securely stored as attachments. A new attachments gallery in the conversation detail view lets operators browse received files. Operators can also send files back to customers from the chat UI.

🔒 GDPR Data Retention

Full GDPR-compliant data retention with configurable retention periods for calls, leads, conversations, and SMS. Set retention rules per account — data is automatically cleaned up on schedule so you stay compliant without lifting a finger.

📅 Cal.com v2 Integration Overhaul

Completely rebuilt for the v2 API with dynamic booking fields and team event types.

• Region selector — choose from EU, US, or custom Cal.com endpoint
• Dynamic booking fields — reads your Cal.com event type's custom fields as-is, no more overwriting your configuration
• Team event types — book into team calendars, not just personal ones
• Full backward compatibility — existing configs continue working without changes
• Smart event picker with grouped personal/team events and auto-fetched booking fields

💬 Chat Conversation Webhooks & Automation

Chat conversations now fire "conversation ended" triggers on inactivity timeout. Payload includes full transcript, extracted variables, WhatsApp sender info, and AI evaluation using your post-call analysis schema. New automation platform trigger for "Conversation ended".

🎨 Widget Customization Upgrade

• Extra large layout — half-screen panel, full-screen on mobile
• Translatable tab labels — rename Voice/Chat tabs to match your brand
• Auto-open on page load — widget expands automatically
• Custom avatar — upload your own widget avatar
• Hide function calls — LLM tool calls hidden from chat by default

🛠️ Improvements & Fixes

• Auto top-up locking — prevents duplicate charges during concurrent top-up attempts
• WhatsApp logo sync fix — resolved an issue where profile logos weren't syncing correctly

🔥 New Call Features: Warm Transfers + Keypad Digits Collection

We’ve added two major capabilities to make AI phone agents more reliable in real-world support and sales scenarios — smoother human handoff and faster, more accurate data capture.

☎️ Warm Call Transfer (Hold + Brief + Connect)

When a caller asks to speak with a human, your AI can now run a true warm transfer: it puts the customer on hold (with music or silence), dials a supervisor/agent, briefs them with context, and only then connects both sides.

Why it’s better

• ✅ Context preserved: agents don’t start cold — they already know what the customer needs.
• 🎶 Professional experience: caller hears a hold message + hold music while the agent is reached.
• 🔁 **More compatible transfers:**Web calls can now transfer to real human phone agents (the initial interaction doesn’t need to be a phone call anymore).
• For SIP trunks without SIP REFER, warm transfer now works by creating a new SIP INVITE and merging calls — no REFER required.

Configurable options Supervisor number, caller ID mode, hold music/silence, hold message, supervisor briefing instructions, and the final “connected” message.

⌨️ Customer Keypad Digits Collection (DTMF)

Customers can now enter digits on their phone keypad during a conversation, and the assistant receives the input instantly and continues naturally.

How it works

• Runs in the background — no awkward pauses
• Auto-submits after a configurable timeout (default 5s)
• Instant submit when the user presses # or *****
• Works alongside voice: customers can speak or type

Perfect for Account numbers, PINs, phone numbers, license plates, order codes — especially when accuracy and privacy matter.

Configurable options Input timeout + submit key (# or *).

These upgrades make your assistants feel more human, more reliable, and easier to deploy in production support environments.

↗️ WhatsApp External Numbers Integration

You can now connect external WhatsApp sender numbers; like your own mobile number or SIP/business-owned lines and use them to fully automate WhatsApp messaging inside the platform.

What’s included

• 📲 Bring your own sender: use your existing number instead of being limited to platform-provisioned WhatsApp numbers
• 🤖 Inbound + outbound automation: the AI can reply to incoming messages and send outbound follow-ups, reminders, and lead qualification messages
• ✅ Simplified onboarding via Meta Embedded Signup with OTP verification (SMS/voice) — no complicated Meta setup guesswork

Where you’ll find it

Create a new WhatsApp sender from the platform and choose the External Number path. Once connected, you can submit message templates for approval to Meta and use your connected number to send and receive WhatsApp messages with AI handling the conversation end-to-end.

This update unlocks a more “native” customer experience — your AI can operate at scale on WhatsApp while keeping your brand identity and sender number consistent.

✍️ AI Prompt Writer, 🧩 Flow builder & 🟢 Whatsapp native integration

This update packs three major upgrades that change how you build assistants end-to-end: prompting becomes conversational, flows become visual, and WhatsApp becomes native — with a setup that skips the usual Meta headaches.

✍️ AI Prompt Writer

Edit prompts by chatting with AI — no manual prompt engineering required.

• Conversational editing: just say what you want (“make it more friendly”, “add return policy info”) and the AI updates your system prompt.
• Smart diff preview: review changes before applying — additions, removals, and edits are clearly highlighted.
• Accept/reject control: apply suggestions one-by-one or in bulk, so you stay in full control.
• Quick actions: one-click improvements like Make concise, Add detail, Improve clarity.
• Variables + schema built-in: manage pre-call variables (e.g. {customer_name}) and define structured post-call extraction fields (string/number/boolean) without leaving the editor.
• Template library: start from proven templates for sales, support, scheduling, surveys.

Now you can refine assistant behavior quickly — without technical skills.

🧩 Flow Builder

Design conversation flows visually with drag-and-drop — perfect for production ready flows.

• Node-based builder (5 node types): Start, Speak, Prompt, Action, End.
• Branching logic: build multi-path conversations (Interested / Maybe / Not Interested) with clear outcomes.
• Settings panel: customize agent personality (name, language, assertiveness, humor) per flow.
• Auto-layout: one click to neatly organize your flow.
• Import/Export JSON: share flows across assistants or teammates instantly.

Build complex, multi-branch scripts without writing long prompts — faster iteration, cleaner structure, easier collaboration.

🟢 WhatsApp Native Integration

Connect your AI assistant to WhatsApp Business in minutes — and let it handle both inbound and outbound messaging automatically.

✅ Setup that’s actually easy

Meta’s WhatsApp onboarding is notoriously complex (business setup, verification, templates, embedded signup, compliance). We’ve turned that into a guided flow so you can go from zero to live quickly:

• Simple onboarding wizard that walks you through the required Meta steps
• Platform-assisted setup that removes the usual guesswork and friction
• You get WhatsApp running without spending days figuring out Meta’s process

🤖 AI-managed inbound + outbound WhatsApp

Once connected, your assistant can:

• Reply instantly to incoming messages
• Send outbound messages proactively (follow-ups, reminders, confirmations, lead qualification)
• Keep a consistent “brain” across channels — same assistant logic, same data, same tone

🔜 What’s coming next

• External numbers support (soon): right now WhatsApp can be activated on numbers from our platform, but bringing your own number is coming next.
• WhatsApp calls support (soon): we’re expanding from messaging to calls so you can handle the full lifecycle in one channel.

Together, these updates make assistant-building dramatically faster—design flows visually, refine behavior by chatting with AI, and deploy natively on WhatsApp for end-to-end omnichannel automation.

🚀 New Models, Chat Automations & Smarter Call Controls

This release brings together powerful new AI models, chat-based automation triggers, and advanced call behavior controls, marking another big step forward for both voice and text experiences on the platform.

🤖 New & Updated AI Models

GPT-5.2 Now Available

We’ve released GPT-5.2, delivering improved reasoning and better instruction-following. For faster responses we recommend to continue using GPT-5.1

GPT Realtime Mini Upgrade

The realtime mini model has been bumped to: gpt-realtime-mini-2025-12-15 This update improves stability and responsiveness in low-latency voice interactions.

New Gemini Native Audio Model (Speech-to-Speech)

Based on Google’s December 12, 2025 release, we’ve added: gemini-2.5-flash-native-audio-preview-12-2025

You can find the new version as Gemini 2.5 Flash Dialog New (Dec 2025) and the previous version as Gemini 2.5 Flash Dialog (Oct 2025)

• Better handling of complex workflows and function calls
• Available in speech-to-speech mode only

⚡ New Conversation Created Trigger

You can now trigger automations the moment a new chat conversation starts on the web widget.

• Instantly capture pre-form input variables such as name, email, phone number, or any custom fields you’ve defined.
• Use this data to create leads, sync contacts to your CRM, send notifications, or start any custom automation flow.
• Works seamlessly with the chat web widget, making text interactions just as automatable as voice calls.

📞 Smarter Call Prompts & Behavior

📨 End Call on Voicemail – Custom Prompt

You can now define a custom prompt that the AI will use when leaving a voicemail. This gives you full control over tone, content, and messaging instead of relying on a generic message.

🔁 Re-Engagement Prompt Setting

When the re-engagement interval is reached, the assistant can now use a custom re-engagement prompt. This allows you to fine-tune how the AI follows up after silence — making interactions more natural and context-aware.

🎧 Multiple Ambient Sound Options

We’ve added multiple ambient sound choices for voice calls.

As always, we’re excited to see what you build with these new capabilities.

⚙️ Automations Update & Calendly integration

This release focuses on making automation setup faster and smoother, while expanding your assistant’s scheduling capabilities.

🔄 Simplified Data Injection for Automations

Setting up automations is now much easier and more intuitive.

• Previously, you had to fully build a flow, run a test call, and only then reference the latest data to configure your automations.
• Now, with a single click on “Load data” when adding a trigger, the platform automatically fetches all custom variables defined in your assistant settings.
• This includes both input variables and post-call evaluation / extracted variables, giving you a complete and accurate data structure from the start.

The result: faster setup, fewer test calls, and a much smoother automation-building experience.

📅 Calendly Integration

You can now connect Calendly calendars directly to your assistants.

• Seamlessly book appointments through your AI flows.
• Ideal for sales calls, consultations, demos, or service-based businesses.
• Works alongside existing calendar integrations, giving you more flexibility in how and where you schedule.

These improvements remove friction from automation workflows and expand your scheduling options — helping you build, test, and deploy assistants more efficiently than ever.

🔊 New Models & Smart Tools Update

This release brings** upgrades** to voice quality, transcription accuracy, and usability across the platform — along with a few handy workflow improvements.

🗣️ Cartesia Sonic 3 (TTS) for Pipeline & Dualplex

Introducing Cartesia Sonic 3, our newest high-fidelity TTS engine with voice cloning support and advanced emotion control.

• 🎭 Toggle SAML tags to adjust tone, emotion, and intensity in your AI’s voice.
• 🎤 Clone and fine-tune voices directly within your assistants for natural, expressive conversations.

✍️ ElevenLabs Scribe v2 Transcriber

We’ve added Scribe v2, a cutting-edge transcription model from ElevenLabs, now available in Pipeline Mode. Expect sharper accuracy, better punctuation, and faster transcription speeds — perfect for complex dialogues and noisy environments.

🎙️ Record Voice Samples in-App

You can now record voice samples directly inside the platform for cloning. No need to upload pre-recorded files — just hit record, speak, and your voice sample will be ready to use in seconds.

🔁 Deepgram Nova-3 Migration

All assistants using Deepgram have been automatically upgraded from Nova-2 → Nova-3, bringing:

• Higher transcription accuracy
• Better noise handling
• Faster response times

🌐 Web Widget Preview Sharing

You can now copy a shareable preview URL of your web widget — no need to embed it on a website first. Perfect for demoing widgets to clients or sharing with your team during testing.

🔄 Bulk Resend Webhooks

Quickly resend multiple webhooks for past calls in one action. Ideal for syncing historical data, debugging integrations, or recovering missed events.

These updates make your assistants more powerful, expressive, and developer-friendly — with the latest models and smarter workflow tools across the board.

🧠 GPT-5.1, campaign level timezone and fixes

🤖 GPT-5.1 in Pipeline Mode

We’ve added GPT-5.1 — smarter, faster, and more context-aware than ever. Enjoy improved accuracy and response times across all your pipeline-based assistants.

🌍 Campaign Timezone Settings

You can now set a custom timezone per campaign, ensuring calls and automations run exactly at the right local time for your audience.

🛠️ General Improvements & Fixes

This update also includes multiple stability, performance, and UI fixes across the platform for a smoother experience overall.

💬 Introducing AI Chat for Web Widget — A Major Step Toward Omnichannel Communication

We’ve just reached a huge milestone in our mission to become a truly omnichannel AI platform for both voice and text interactions.

With this release, your AI assistants can now chat directly with users through a fully integrated web widget, alongside voice mode. This unlocks an entirely new way for customers to engage with your AI — from websites, landing pages, and soon, across multiple messaging channels.

🧠 AI Chat Mode in Web Widget

Assistants can now operate in chat mode, voice mode, or a hybrid voice + chat mode, allowing you to choose the ideal experience for your users.

The web widget can be easily embedded on any website, giving your visitors instant access to your AI assistant for inquiries, bookings, or lead capture.

Coming soon: support for WhatsApp, SMS, and Meta Messenger, making your assistant truly omnichannel.

💬 Conversations View

Just like you can see your Call History, you now have a dedicated section for Chat Conversations — view, search, and manage all customer chat interactions directly from the dashboard.

Each conversation is logged with timestamps, user details, and message history, so you can track performance and customer engagement easily.

🧩 Web Widget Pre-Form Builder

Collect essential lead data before your assistant starts chatting or talking.

You can now build custom pre-chat forms with a visual form builder, fully customizable for your business use case — whether that’s a salon appointment, a medical consultation, or a sales lead.

Supported Field Types

• Text – Single-line text input
• Email – With built-in validation
• Phone – With country selector
• Text Area – For multi-line inputs

Additional Settings

• Required fields (ensure important info is captured)
• Variable names sent directly to your assistant (e.g. customer_name)
• Labels, placeholders, and helper text for a clean, user-friendly interface

🎨 Enhanced Widget Customization

The web widget is now more flexible and on-brand than ever:

• Choose between Modern or Classic styles
• Adjust size, placement, and colors
• Enable voice-only, chat-only, or hybrid interaction modes

💰 Pricing & Availability

• Free when used internally for testing assistants (no balances deducted).
• Public chat mode costs just $0.01 per AI reply, regardless of the model or message size — making it one of the most cost-efficient AI chat options available.

🌟 Custom dashboards, 💬 Assistant chat, 💳 Auto top-up, 📅 1-Click Cal.com integration

We’re excited to roll out a series of powerful updates designed to make managing, testing, and customizing your AI assistant experience smoother than ever.

📊 Custom Dashboards

Say hello to fully customizable dashboards built for your workflow.

• Drag and drop charts, stats, and tables to design layouts that fit your business.
• Display data from calls, leads, campaigns, or any connected integrations.
• Add and resize widgets in real time to focus on what matters most — performance, revenue, or engagement trends.

💬 Live Chat with Assistant

You can now chat directly with your assistant in real time using text — perfect for testing and refining conversation flows without making live calls.

• 💡 Free usage for testing: Chats with your assistant don’t consume any balance, so you can experiment safely.
• 🧠 Great for training, debugging, or reviewing responses before deploying changes to production.
• 🚀 Coming soon: Extend live chat to customer-facing channels like your website widget, WhatsApp, Messenger, and SMS, allowing users to talk to the AI on their favorite platforms.

This makes building and improving assistants faster, cheaper, and more accessible for every user.

💳 Auto Top-Up

No more interruptions due to low balance. The new Auto Top-Up feature ensures your system always stays online and calling.

• Get real-time alerts when your balance is running low.
• Automatically add credits from your default payment method.
• Set thresholds and amounts to stay in full control of your spending.

It’s a simple but powerful way to guarantee uninterrupted service — especially for active campaigns or high-volume operations.

🔗 Cal.com 1-Click Integration

Connecting your Cal.com account is now faster and simpler than ever.

• Use the new one-click connect button to instantly link your calendar.
• Built-in troubleshooting tools help you identify and fix sync issues quickly.
• Perfect for ensuring your scheduling automations and booking workflows stay seamless across assistants and team calendars.

📅 Set Date Format

You can now choose your preferred date format across the platform.

• Select from common regional styles like DD/MM/YYYY or MM/DD/YYYY.
• Improves clarity for reports, logs, and analytics — especially across global teams.

📵 Blacklist, ♻️ Automatic number rotation, 🔁 Transfer call improvements and more

We’ve rolled out several powerful updates to give you more control, visibility, and performance across your calls and assistants.

📵 Blacklist System

Easily block unwanted numbers for both inbound and outbound calls.

• Add numbers to your Blacklist directly from Calls History or Leads (single or bulk).
• Calls to or from blacklisted numbers are automatically rejected with a clear message: “Number is blacklisted.”
• View, manage, or remove entries via the new Blacklist panel.

📞 Campaign Number Pool

Campaigns can now use multiple outgoing numbers.

• Assign several numbers to a campaign.
• Calls are automatically rotated and balanced between numbers for even distribution.
• Built with fast and efficient number rotation algorithm.

🔁 Transfer Call View

A clearer way to track transferred calls.

• See all transferred calls directly in the Calls table with detailed info like parent call, duration, and outcome.
• Available for calls made through dedicated numbers or the “Use your own number” feature. SIP trunk integrated calls cannot be tracked for call transfers.

🤖 Gemini 2.5 Model October version

Introducing the latest Gemini 2.5 model released in October for Speech-to-Speech mode.

• Smarter and more accurate with improved function-calling abilities.
• Will soon be integrated into Dualplex Mode for an even better assistant experience.

📅 Multi-Calendar Support for Assistants

We’ve added Multi-Calendar Tools, making scheduling smarter and more flexible than ever.

What’s new:

• 🧑‍🤝‍🧑 Team & Service Calendars – Assign different calendars to the same assistant, perfect for businesses where customers choose a specific team member or service (e.g. hair vs. nails in a salon).
• 🩺 Medical Appointments Made Easy – Patients can now book directly with the right doctor’s calendar, streamlining scheduling in clinics and practices.
• 🔄 Seamless Management – Assistants handle multiple calendars effortlessly, reducing back-and-forth and ensuring accurate bookings every time.

This update makes your AI assistants even more versatile, whether you’re running a salon, clinic, or service team.

🔀 Dualplex Assistant Mode (Beta)

We’re excited to introduce Dualplex Mode, our newest and most powerful assistant experience — combining the best of both worlds from Speech-to-Speech and Pipeline modes.

What’s new in Dualplex Mode:

• 🎤 Speech-to-Speech Intelligence Leverages the advanced STT + LLM stack from our Realtime models for natural, fluid conversations with ultra-low latency.
• 🔊 Premium Voice Output Uses ElevenLabs TTS, bringing high-quality voices and voice cloning capabilities directly into the experience.
• ⚡ Best of Both Modes Seamlessly blends the responsiveness of Speech-to-Speech with the flexibility and voice quality of Pipeline.

Why it matters:

Dualplex gives you the most realistic AI voice interactions yet, making conversations smoother, more human-like, and easier to integrate into production environments.

⚠️ Note: Dualplex is currently in beta. We’re gathering feedback and improving quickly — try it out now and let us know what you think!

🗣️ GPT-5 Realtime Speech-to-Speech

We’ve rolled out major improvements to enhance speed and conversation quality across the platform:

• 🗣️ GPT-5 Realtime in Speech-to-Speech Mode – Natural, fluid conversations with ultra-low latency for voice interactions.
• ⚡ GPT-5 Mini & Nano in Pipeline Mode – Optimized for minimal reasoning effort, delivering faster response times and smoother call flows.

🧠 Campaigns just got smarter

We’ve added powerful new controls to help you get more out of your outbound campaigns, avoid wasted calls, and make sure every lead gets the attention it deserves.

Here’s what’s new:

1. Campaign Call Status Options

You now have three new controls at the campaign level to fine-tune how your leads are called:

• Retry until goal completed Continue calling until the goal is achieved. Perfect for situations where a lead answers but doesn’t complete the desired action (e.g., no appointment booked). Calls will continue until the selected goal variable is marked as true.
• Retry on voicemail Continue calling until a human answers (skip voicemails). If a call hits voicemail, the lead won’t be marked as completed — the system will retry until someone picks up.
• Goal variable Variable that must be true to mark lead as completed. Select from your variables which one determines a “completed” lead. If the variable is false, the lead will be retried instead of being marked as completed.

2. Update Lead Data/Status from Automation Platform or API

You can now update lead details and statuses directly from your connected automation tools or API calls.

• Change lead status (e.g., “created” instead of “completed”) dynamically.
• Push updated information into your campaigns without manual edits.

3. Ringing Timeout in Assistant Settings

A new ringing timeout option in the Assistant settings lets you control how long to wait for an answer before moving on.

• For outbound campaigns, set it lower to avoid voicemail and save agent time.
• For campaigns with slower-answering customers, increase it to give them more time to pick up.

🚀 GPT-5 Integration

Experience the most advanced AI yet — faster, smarter, and more accurate responses now power your workflows. New paramters for adjusting reasoning effort will be added shortly. Right now the default value makes it very smart but slow.

📞 DTMF Input Tool

Easily send keypad inputs during outbound calls, enabling smooth navigation through IVR menus and automated workflows.

📋 Leads kanban board

Added Kanban view to the Leads section for a more visual workflow. You can now drag and drop leads between stages, quickly update statuses, and track progress at a glance.

📚 Knowledgebase search & improved management

Knowledgebases have just been improved. Now you can:

• Search within your knowledgebase documents
• Delete an individual document. This is useful when you want to update a small part of an info without having to replace the whole knowledgebase
• View all the ingested chunks in each document

📲 Make phone call action in automation platform

Added a new “Make Phone Call” action that instantly dials a single number as soon as the API endpoint is invoked—no campaign setup required. Perfect for one-off or time-critical calls, while campaigns remain the recommended solution for bulk calling lists. You can find this action in the automation platform or API.

👥 Secondary contacts on leads

You can now attach multiple backup contacts to every lead. When the primary contact doesn’t answer, calls automatically cascade to the next contact in the list. Add, reorder, or remove secondary contacts from the lead details panel for smoother outreach and higher connect rates.

📶 Stable SIP proxy server

Customers that used "Outbound SIP proxy server" for their outbound SIP trunk connections to come from a fixed IP, can now switch to the new stable server which offers better connection and stability as this feature was previously in beta. If you used this feature before the update, you will see a button to migrate to the new server from the SIP trunk interface. Make sure to allow in your PBX/VoIP provider the new IP which you can find in the same SIP trunk interface dialog from phone numbers page.

🌍 Multi-lingual assistants

Assistants can now speak multiple languages in the same conversation by setting up "secondary languages" in assistant settings first tab. They can switch language mid-call based on customer needs, providing seamless multilingual support.

🌐 Web widget - Voice AI for websites

Embed assistants on websites with the voice AI widget similar to chatbots but with voice. You can fully customize the widget appearance from assistant settings from the "Web widget" button: colors, texts, positions and more. Paste the generated script tag snippet on any website page and you're good to go.

🌟 Go High Level calendar integration

Fetch available slots and book appointments using GoHighLevel calendars

New Gemini 2.5-flash dialog model (beta)

Added newest speech to speech model performing best for quality and latency. Includes 30+ voices and works in multi-lingual as well.

---

🖥️ Application

Interface

• Added account level timezone setting
• Added minimum words interruption option in assistant settings
• Added volume slider for background ambiance audio

🚀 New 2.0 Voice Engine - Speak Like Never Before

The long awaited new engine has been deployed in the platform. It comes with improvements on overall conversation from response times, more emotion, better understanding to improved interruption handling and natural feel.

⚡ Real-time Speech-to-Speech Mode

Skip TTS round-trips and give callers instant, more natural replies. Experience unprecedented conversation flow with real-time processing.

🎭 Sentiment Aware Filler Audios

Select different filler audios based on customer reply sentiment for even faster responses. Your AI assistant now adapts its tone and pace based on customer emotions.

---

📚 Knowledge Management

Build a Living Knowledgebase in Minutes

• Point-and-scrape any website — Product docs, help centers, wikis, or Notion pages
• Drag-and-drop PDFs & files — We'll ingest the text automatically so your agents can use it in-call

---

🖥️ Application

Call directly from your browser (beta)

This feature is present in beta mode and you can test it from assistant settings. It will soon be extended as a widget that can be embedded on websites.

Multi number call transfer

You can now route calls to multiple destination phone numbers like different people or departments according to your flow.