Product update · August 2026

    One agent, two languages. Plus smarter redials and a queue for follow-ups

    Here is what we shipped in August. Your AI voice agent now speaks two languages on one call. Campaigns redial from a fresh number. Calls that need a person land in one queue. And you can test a script on your own phone. Plus a date-and-time question type, a spoken-currency setting, and more.

    Vitālijs Silins-Ozers

    Vitālijs Silins-Ozerson LinkedIn

    CTO & co-founder · 31 August 2026

    major

    Your agent now speaks two languages

    Since July your agent could notice a borrower answering in another language and check first. Now the call simply continues in that language. Your whole script, in your own translations, including answers to common questions.

    The call script's Multilingual settings, simplified: Latvian as the flow language with Russian added, On request only selected as the trigger, the confirmation question written in Russian, and a Translation readiness line reading Ready to deploy. Beside it: One call script. Two languages.
    One script, two languages: the borrower asks, the agent confirms in their language, and the call goes on in your own words. Product representation with synthetic values.

    Set it up once in the script's new Multilingual tab. Pick the languages, what starts a switch, whether the agent confirms first, and what it does next. A readiness check shows what still needs translating. The script cannot go live until it is complete.

    • Two ways to start a switch. On request only: the borrower asks. Automatic detection: the agent hears them speaking the other language.
    • The agent asks the confirmation question in the new language. A yes proves the borrower can follow it.
    • After the switch the agent stays in that language to the end. Amounts and dates are read the way that language says them.
    • Translations stay with the script when you export or import it. Each translation view has a link you can share.

    What shipped

    • One Multilingual settings tab
    • On request only mode
    • Translated answers, spoken word for word
    • Confirmation in the new language
    • Translations survive export and import

    major

    Redial from a number they have not declined

    A campaign can now dial from several of your trunk's numbers, not one. Each redial prefers a number this borrower has not seen. A call screened once is not screened again on the same number.

    The campaign builder's Caller ID setting, simplified: Rotate across the trunk's numbers selected, three trunk numbers with the current pick highlighted, the mode Prefer a number they have not seen, and a Redial policy with a 15-minute wait before the first redial and 120 minutes before the second. Beside it: Redial from a number they haven't declined.
    Pick rotation and its mode under Caller ID, then give each redial its own wait. Product representation with synthetic values.

    You already set how many redials a campaign makes and how long it waits. What is new: which number the borrower sees, and a different wait before each redial. Try again in 15 minutes, then in two hours.

    • Three ways to pick the number: prefer one the person has not seen (default), keep one number per person, or pick at random.
    • Add or remove a number on the trunk and it applies at once, even to a running campaign.
    • Set a wait before each redial: up to ten waits, from 1 minute to 24 hours. The last wait repeats if more redials remain.
    • Both settings are in the campaign builder and on the campaign API.

    What shipped

    • Caller-ID rotation, three modes
    • Per-redial waits, builder and API
    • Trunk changes apply to live campaigns

    major

    One queue for calls that need a person

    A new Escalations tab on the Calls page lists every call that needs follow-up. Transfers to your team that did not connect, plus calls that match your own rules, such as a callback request or a dispute. Resolve each one when it is handled.

    The Escalations tab, simplified: four open calls with reasons Transfer failed, Callback requested, Dispute raised and Not interested as coloured badges, a callback time on one row, and a Resolve button per row. Beside it: Every call that needs a human, in one queue.
    Failed transfers and the calls that match your rules land in one queue, each reason in the name and colour you chose. Product representation with synthetic values.

    Each row shows who, when and why in plain words. It carries the callback time when the borrower gave one, and a link to the call. Resolved items move to their own view with who closed them and when.

    • Failed transfers are always in: no answer, rejected, connection error, not attempted, or the borrower hung up.
    • Add your own rules on collected fields: a field equals a value, or is set. Financial, card and identity-check fields are excluded.
    • Give each rule a name and a colour, so a callback request and a rejection look different at a glance.
    • Call history gets a Transfer failed filter for the list, the export and the metrics tiles.

    What shipped

    • Escalations tab, Open and Resolved
    • Your own rules per project
    • Named, colour-coded reasons
    • Transfer failed filter

    major

    Hear your script on a real phone

    Type your number in the script editor and your AI voice agent calls you. It runs the exact version you have open, over your own phone line. The conversation appears on your screen as you talk.

    Edit, ring, listen: a test call in 27 seconds. Conversation shown is illustrative and synthetic. Music: “Deliberate Thought” — Kevin MacLeod (incompetech.com), CC BY 4.0.

    Browser testing checks the logic. The phone is where you hear how the greeting lands and how an amount sounds. Nothing goes live, and there is no campaign to set up. Full story in the launch post, Test calls: your agent calls you first.

    • The call runs the draft you are editing, not the published version.
    • If the script needs details to run, such as a name or a due date, you enter them before dialing.
    • Test calls get a Test badge and stay out of your reports by default. The Test & practice filter brings them back.
    • Before a browser session, a headset check explains why laptop speakers make the agent mishear.

    What shipped

    • Phone test calls from the editor
    • Live transcript during the call
    • Test badge and filter
    • Headset check for browser tests

    Everything else that shipped

    Campaigns illustration

    Campaigns

    First attempts same-day only (API)Carrier dial format per SIP trunkCaller-ID format per SIP trunk
    Building call scripts illustration

    Building call scripts

    Date and time in one questionChoose the spoken currencyNames in the right grammatical caseFlow settings open over the diagramLetters and numbers as labels
    Calls and reporting illustration

    Calls and reporting

    Test & practice filterCallback time on escalation rowsEscalation reasons named and coloured
    Languages illustration

    Languages

    Shareable translation view linksTranslations kept in export and importAmounts and dates, in their languageConfirmation in the new language

    See it on your own flows

    These shipped for every workspace with outbound campaigns.