Bu sayfa senin dilinizde henüz mevcut değil - İngilizce sürümü gösteriyor.
Talk to your agents. The work shows up while you're still talking.
Tap Voice in any conversation and you're in a real spoken call. Ask a question and you get an answer out loud. Ask for a post, an image, an email or a document - and the agent reacts in its own words, keeps the conversation going, and the finished card lands on your screen a few seconds later.
You never wait in silence while something "processes". That's the whole point of how it's built.
The one thing voice will never do
Voice cannot publish and cannot send. Not by prompt-wishing - by design. An agent on a call prepares the draft and tells you to tap the button yourself. 17 irreversible actions (publishing to Instagram, Facebook, LinkedIn or X, sending or scheduling mail, posting to Slack, deleting a calendar event or a Trello card…) are confirmation-gated in the database, so nothing fires unattended. If one is ever triggered mid-call, a Yes/No card appears above the orb and you have to click it.
Two brains, so you never wait
Most voice assistants make you sit through the pause: you ask, it goes quiet, it "thinks", it comes back. Unyo splits the job in two.
| What it does | What you experience | |
|---|---|---|
| The voice | Talks. Only talks. No tools, nothing to wait on. | Instant, natural, 1–2 sentences at a time - like a call. |
| The worker | Runs the agent's real, full toolset in the background and produces the card. | The card appears on screen while you're still mid-sentence. |
The one that speaks never waits for the one that works. Moments after the agent answers you out loud, a "preparing" skeleton appears with a paced progress sound. When the real card is ready, you hear a notification and it drops into the conversation - the same editable card you'd get from typing.
The agent says something warm and human back - different every time, never a canned "got it, I'm on it" - and the post draft, caption, hashtags and image arrive on screen without either of you pausing.
Nine agents, each with their own voice
You can call Ashley (social), Maya (email and calendar), Alex (sales), Riley (product), Tyron (images and audiovisual), Sam (support), Lucy (content), Blake (data) and Ema (personal development). Each has its own voice, per language.
Steve, the website builder, has no voice mode. Building a site is a workshop, not a phone call - you do that by typing.
And the boundary you get in text is the boundary you get on a call. Maya is the only agent who can touch your inbox, on a call exactly as in chat. Calling Ashley does not hand her your mail.
Ask. Change your mind. Cancel.
Modify by voice. Edits are first-class, not an afterthought:
A fresh card comes back. "Change the background to orange", "redo it in a warmer style", "add my logo" - all of it reaches the agent as a real request, not as small talk.
Cancel by saying no. While a card is being prepared, say non, annule, laisse tomber, cancel, forget it or never mind - and it's genuinely killed. Not hidden: the row is deleted. The card never appears live and it isn't there when you reload either.
It only listens for a cancel when there's something to cancel
The cancel words are only checked during the window where a card is actually being generated, and only in a short turn. A stray "no" in the middle of a normal conversation can't wipe out your work.
It asks when it doesn't know. If the agent is missing a detail - "what's the email about?" - it asks you out loud, in its own voice, rather than guessing or going silent.
The transcript survives the refresh
Both sides of the call stream into the chat as live bubbles, word by word, while you speak.
Then reload the page. You'll see exactly what you heard: your spoken turns, the agent's spoken turns, and the card. No unspoken prose the agent never said out loud. No duplicated half-sentences from the speech recogniser refining itself mid-turn. Just the call.
This is not a transcript feature bolted on afterwards - every turn is written down as the call happens, deduplicated so each one is stored once, at its final wording.
It hears your business, not just your words
11 languages, plus auto-detect. Go to Settings → Preferences → Voice language and choose what Unyo listens for: English, French, Spanish, German, Italian, Portuguese, Turkish, Arabic, Chinese, Japanese or Russian - or leave it on auto-detect if you switch languages mid-call.
It knows your vocabulary. Every call rebuilds a recognition boost list from your live Neural Core: your company name, your product names, your contact names, plus "Unyo" and the agent names. So your brand isn't transcribed as three different misspellings. It's derived fresh on each call, which means renaming a product or deleting a contact just works - nothing to re-train, nothing cached.
Only names are ever used for this. Never a token, never an email address.
And it knows you. The agent on the call gets the same context as the agent in your chat: your first name, your company, your Neural Core, your contacts, your products. Mention a saved product by name in an image request and it uses that product's reference photo and your saved style preferences.
Auto-detect vs. picking a language
Auto-detect understands several languages but is less precise than locking one in. If you always call in the same language, pick it - you'll get better recognition. A few languages listen in multilingual mode regardless, so if a name is being misheard, adding it to your Neural Core is the fastest fix.
During a call
- Mute and unmute your mic at any time, with a distinct sound for each.
- End the call with the red hang-up button, or just press Escape.
- A live orb reacts to both your voice and the agent's, with an honest status line: Connecting… / Listening… / {agent} is thinking… / {agent} is speaking…
- Start from nothing. Open a fresh chat and hit Voice - the conversation is created for you and appears in your sidebar the moment you actually say something. Hang up without speaking and it cleans itself up. No graveyard of empty conversations.
- The attachments button is off during a call, and web research is off too - the agent answers you directly rather than going quiet to search.
What you can walk away with
Everything the card system does in text, spoken:
- Social post drafts - caption, hashtags, platforms, generated image.
- Images - generated or modified, product-aware when you name a saved product.
- Email drafts - from Maya, ready to review and send by tapping.
- Documents and captions.
All of it produced by the agent's real toolset. Not a summary of what it would do.
FAQ
Can an agent publish a post or send an email while I'm on the call?
No - and this is deliberate, not a limitation we're working around. The agent prepares the draft and tells you, in one short sentence, to tap the Publish or Send button on the card. Behind the prompt there's a real code gate: 17 irreversible actions are confirmation-gated by database configuration, so they can't fire without a human tap. If one ever is triggered, you get a Yes/No card above the orb.
Which agents can I call?
Nine of the ten: Ashley, Maya, Alex, Riley, Tyron, Sam, Lucy, Blake and Ema. Steve, the website builder, is text-only.
Does the agent on the call know what I told it in chat?
Yes. It's given the same context the text agent has - your name, your company, your Neural Core entries, your contacts and your products. And because the Neural Core is shared across all ten agents, anything saved in it is known to whichever agent you call next.
What happens if I say 'no' halfway through?
The card is thrown away - genuinely deleted, not just hidden. It won't appear live and it won't be there after a reload. The cancel words are only listened for while something is actually being generated.
Will my call be in my chat history?
Yes. Both sides of the conversation are written down as you speak, and reloading shows you exactly what you heard - your turns, the agent's turns, and the cards. Nothing the agent didn't say out loud gets added.
Can I change the language it listens for?
Settings → Preferences → Voice language. Auto-detect, or one of 11 languages: English, French, Spanish, German, Italian, Portuguese, Turkish, Arabic, Chinese, Japanese, Russian. Auto-detect is more flexible; picking one is more precise.
Does talking cost credits?
You're charged for the work an agent produces, not for the conversation. A voice request runs through exactly the same pricing as the same request typed: drafts, reads, searches and plans are free, and generation costs what it costs in text. If a piece of work fails, it's refunded automatically.
Can I ask it to look something up on the web during a call?
Web research is switched off during a call by design - the agent answers you conversationally rather than going silent to search. For research-backed work, use the chat.
Ready to try it? Open any agent, leave the composer empty, and tap Voice.
See also: The agents · The Neural Core · Security