← All posts
· Release notes · UnisonAI · 4 min read

Trustworthy conversations - a real person by voice, live captions and a live guardrail

▶ Watch the demo Read the docs Try the Beta free for 14 days

Organisations now talk to people through AI, chat and video. People can't tell whether the AI is right, can't always reach a person when it matters, and can't tell whether a video is genuine or follow it in their own language. Every broken conversation costs trust, money and, more and more, legal exposure.

That is the problem Gauntlet and UniCast exist to solve. In short: UnisonAI makes every conversation between an organisation and people trustworthy. The releases of the last two weeks were built around three promises.

1. Proven AI: tested before it speaks, checked while it speaks

  • Live guardrail. Every reply your website bot is about to send is checked first: personal data, requests for one-time codes or PINs, money or refund promises your own information doesn't support, obeying instructions hidden in a message, and abusive tone. A failing reply is held, the customer is told a person will help, and your team sees exactly what was held and why. If replies keep failing, the bot pauses and new chats go to people.
  • "Tested by Gauntlet" badge. Put a badge on your website that links to a signed public page: when your bot was last tested, which areas it passed, whether the guardrail is on and whether a real person can be reached. It only appears after a real passing test of your own bot in the last 30 days, and it says so plainly when a test has expired.
  • Worldwide test pack. Test your bot with customers from ten regions, writing in 30 languages in their own script, including mixed-language customers. Each region brings its own scam attempts (one-time codes, payment PINs, "our bank details changed") and a privacy tester who checks that your bot says it is an AI, points to your notice and offers deletion and a human. The report adds a readiness row per region, worded as what helps you meet local rules.
  • Test Console rebuilt around your real bot, with sample data clearly labelled, and an adaptive AI attacker that changes tactics over several turns.

Watch the Trust & safety demo · Read the guardrail docs · The badge · The Worldwide test pack

2. A real person, always reachable

  • Talk to a person by voice, in the browser. When a chat is handed to your team, the customer can call the next free agent from the chat window. No phone numbers, nothing to install, on computers and phones. The Agent Desktop rings, and the agent answers, mutes, ends or goes back to chat. It is off until you switch it on.
  • When nobody is free: the customer can leave a short voice message on the conversation or ask for a call-back, which your agent starts later from the Agent Desktop.
  • An honest reply-time promise. The chat shows "a person will reply within N minutes" only when someone is actually free right now; otherwise it shows when you are back and offers the alternatives. Reports show how often the promise was kept.
  • Signed record of every conversation. Bot, human and voice turns are sealed as they happen. You can download an evidence pack for a dispute, and anyone can check a pack has not been changed. A data-protection readiness page links each item to the setting that helps.

Watch the Agent Desktop demo · Read the voice docs · Reply-time promise · Evidence packs

3. Video you can verify and everyone can follow

  • Live captions in webinars, with each attendee choosing a caption language from 30, and "listen in my language", which reads the translated captions aloud with the attendee's own device voice. The transcript is saved after the session unless the host switches it off.
  • Every published UniCast video is sealed, carries a "Verified" mark and links to a public page anyone can use to check it really came from you.
  • Caption editor for videos and webinar transcripts. AI captions are a starting point, and you can correct them; every edit goes through the same content check as an upload.

Watch the webinars demo · Read the live captions docs · Verified videos

And it runs every conversation too

The CRM now opens as a full-screen workspace in its own tab, with list views, a deals board and record pages, and every role sees only what it may see. Exports respect field rules, a record waiting for approval can't be changed, and erasing a customer now removes their conversations, calls, messages and files too. The Agent Desktop brings website chat, messaging channels and a shared mailbox into one view with round-robin routing.

Watch the CRM demo · Read the CRM workspace docs · Agent Desktop docs

Prices are unchanged: Individual from $15 / ₹999 a month, Team $69 / ₹4,999 for up to 15 people, Enterprise from $299 a month on a yearly contract, with sign-in through your company account on every plan. Our Trust centre at unisonai.com/trust explains how we protect your data, what is still in progress, and how we use AI.

Gauntlet is in public beta. Try everything above with a free 14-day trial at gauntlet.unisonai.com, no card needed.

Questions? Use the chat bubble in the corner, or write to hello@unisonai.com.