New features
Recordings play in a media player docked at the bottom of every page. Recordings now open in one player that stays put while you move around the app. The Call Log has a new Play column, and the waveform under a call’s header hands the recording to the docked player. The player has Back 10 seconds and Forward 10 seconds, Playback speed from 0.75× to 2×, volume and the call’s sentiment. Expand player shows a transcript that scrolls with playback and a Speakers list: click a speaker to hear only their side of the call. The waveform is coloured by who was speaking, and clicking any transcript bubble jumps to that line. See Media player. Listen in on a live call, or take it over, from the player. Opening an in-progress call now shows it in the docked player with its live sentiment. The player’s Action menu holds Live listen, which joins you to the call muted so you can hear it without interrupting, Trigger handoff and Cancel call. Set the Handoff destination to This browser and Trigger handoff becomes Take over, which puts you on the call with your microphone live. You need to be listening first. See Live monitoring. Pin a step to a specific AI model, including your own provider keys. A new Models item in the step menu chooses which model writes that step’s reply. The default is included with your plan. You can also connect your team’s own SCX, OpenAI, Google AI or Anthropic key under Integrations, and that provider’s models then appear in the menu. A step pinned to a model that’s no longer connected shows a Model unavailable warning and runs on the default until you reconnect it. See Steps and Integrations. Intent Pathways can be made conditional, and can return the caller somewhere new. Each pathway has a new Only active when step: checks against call variables (such as is true, is empty or equals) that must all pass before the pathway is even considered. A new When that destination ends setting keeps the default Return to previous, or can send the caller to any main-flow step once the detour is over. See Intent Pathways. A new “Off” voicemail mode keeps the conversation going. The Voicemail setting now has four modes. Off turns voicemail detection off completely, so the call carries on as a normal conversation. Use it for scripts that must never be cut short by a mistaken voicemail detection. Hangup still detects the answering machine and ends the call. See Voicemail message. Start a new script from a template browser. The Create New Script dialog now has three options: Generate with AI, From Existing and New Script. Below them, a browser with Templates and Existing Scripts tabs lets you search, filter by industry and call type, and see each template’s thumbnail, description, call type and voice. Choosing a template and clicking Create makes a copy named after the template, using its voice. See Creating a script. Embed a call button on your own website. The newPOST /api/v1/create-web-session endpoint lets your server create a short-lived, single-use session so a visitor’s browser on your site can call one of your scripts. The script, contact and number are fixed when the session is created, and the call counts towards your team’s concurrent call limit. See Embedded web calls.
Set your team’s timezone. Settings → General has a new Team Timezone card. Scheduled calls and business hours for the team follow this timezone, and it defaults to Australia/Sydney. Only the team’s primary owner can change it. See Team settings.
Wait for whoever answers before opening an outbound call. The new Outbound Opening | Message Check tool goes before the first voice step of an outbound script. The call dials as normal, but the opening line isn’t spoken until the person who answered has said something, or the wait runs out. The steps after it can read that first utterance and route on who picked up: a person, a voicemail greeting or a call-screening service. See Outbound Opening | Message Check.
Code Steps can adjust turn timing for the rest of a call. Two new functions: idle_offset(seconds) lengthens or shortens how long the assistant waits in silence, and endpoint_offset(seconds) does the same for how long it waits after the caller stops speaking before replying. The last call wins, and 0 clears it. See Code Step functions.

