Talking & Listening

Animated voice messages

Voice replies can show the speaking character while you listen, including outside calls. This is available on the website, in iPhone TestFlight build 281, and in Android 2.11 (build 112). The App Store release may be older.

On the website, Android 2.11 (build 112) and iPhone TestFlight 281, Kiana and Della have their own prepared eyes, mouths and brow expressions, with graduated movement following playback and individual movement rhythms. Della has silver curls and a plum cardigan. On iPhone, voice calls also show the originating companion while listening, thinking and speaking. Mouth movement follows the call’s actual output audio. Live camera mode keeps its existing display. Other characters use their own portrait with gentle movement while individual facial artwork is prepared. An unavailable or changed portrait falls back safely. Animation does not change the voice and adds no speech-generation request. Playing an older text reply can still generate speech at the normal voice cost.

On the website, the portrait has Pause voice message and Resume voice message; Resume continues the same loaded recording. Use Settings → Speech → Animated voice portraits to hide it. On iPhone, use Settings → Feedback & Sounds → Animated voice portraits. On Android, use Settings → Appearance & motion → Hide voice portraits. The text and ordinary audio controls remain available.

Reduce Motion keeps portraits still. Images do not become screen-reader stops. iPhone voice portraits may move while VoiceOver is on; Android’s existing motion policy keeps them still with an accessibility service active. Phone battery-saving and background rules also apply.

Expression with a steadier pace

In the voice picker's Delivery control, Balanced and Steady now keep directions such as “unhurried” or “quick” from overriding your chosen speaking speed. Feelings and laughter stay available. Lively keeps the wider performance range, including deliberate pace changes. These adjustments apply to Inworld and Fish Audio when you generate speech, including a fresh replay of an older reply. Audio already downloaded or cached keeps its original performance.

Companions are now guided to give their voices more natural pitch movement, emphasis and phrasing, while keeping a comfortable pace. You can ask for a smile in the voice, a skeptical edge, a gentler reading or deliberate deadpan. Their directions should follow the conversation rather than defaulting to a flat, straightforward reading. This improves the guidance they write; the chosen voice and Delivery setting still affect the result. Existing recordings keep their original delivery.

New replies are guided to change feeling when the conversation calls for it, rather than switching between quiet and upbeat on a schedule. Delivery varies by voice; use Preview to hear your choice.

Replaying replies on iPhone

In TestFlight build 279, replay uses the original character's current voice. Your personal voice choice takes priority; otherwise their current creator-set voice is used. Older replies are voiced when you play them, without regenerating your entire history.

Twelve new voices, one replaced

September 11, 2026: twelve Inworld voices joined the picker. Tags to search for: juniper, cobalt, tamarack, quartzite, marigold, cedarwood, hollyhock, saffron, peridot, ironwood, flint and nutmeg. They sit in the ordinary categories: mostly bright young women, three deep men, and one smooth Indian English man under Men, from abroad. One older voice, gravelly low older man · gouda, left the platform and its slot now speaks as smooth low man · slate; nobody had it saved, and anything that did point at it would simply speak in the new voice. Descriptions are a listening guide from a machine ear, not a human's; preview before you pick. Fully close and reopen the iPhone app to refresh its voice list.

Nine new voices

New voice tags to search for: moonflower, starflower, silverleaf, rainlily, snowberry, sunstone, rainbird, windflower and meadowlark. They are mixed into the existing voice categories. Your saved voices keep their settings. Voice descriptions are a listening guide; preview a voice to choose the sound you prefer.

September 8 listening corrections: Magnolia, Mulberry, Pumpernickel and Sunflower are now under Kids and teens; Sequoia is under Characters and cartoons. Their sounds and your saved choices stay the same. Fully close and reopen the iPhone app to refresh its voice list.

September 11 listening correction: Monarch is now under Kids and teens (reported from the phone picker by a family member as sounding like a kid or a teen). A saved pick of the old name keeps working; the picker just files it in the right section.

From September 11 the picker moves a reported voice on its own: when you press Wrong section? and say what it sounds like, the voice is re-filed under that heading within a few minutes, the report is closed with a note, and you hear about it in chat. A voice you already picked keeps working under its old name.

Two separate things live here, and you can use either, both, or neither:

• Listening — having the AI's replies read out loud to you.
• Talking — speaking your message out loud instead of typing it.

Having replies read out loud

Every reply has a small play button. Activate it and the reply is spoken to you.

In iPhone TestFlight build 274 and later, the thinking sound stays quiet between spoken parts of a reply. It can return when the character is actually thinking or using a tool. A pause while the next piece of speech loads is silent.

Speaking instead of typing

There's a microphone button by the message box. Activate it, say your message, and it gets turned into text for you to send. The first time, your browser or phone will ask permission to use the mic — say yes.

In the native iPhone app: look for a microphone inside a filled circle immediately to the right of the message box, beside the upward-arrow Send button. Tap once to record, speak, then tap the red Stop circle. Review the words in the box and tap Send. You do not need to hold the button down. VoiceOver calls it Record a voice message. This is Kade-AI’s own button above the keyboard; the microphone on Apple’s keyboard uses iPhone dictation. Step-by-step iPhone directions.

Changing the voice you hear

Most characters come with a voice their creator picked for them, so they each sound like themselves. Prefer something else? There are over five hundred voices — calm ones, warm ones, dramatic ones, silly ones.

You can hear any of them right in the chat — no separate page needed. Open Settings → Speech → Text-to-Speech → Voice, pick a voice from the list, then tap the Preview button to hear a short sample. When one sounds right, just leave it selected and that becomes your voice.

Have a whole voice conversation

Next to the message box there's a phone button. Tap it and the screen turns into a call — you talk, the character talks back in its own voice, no typing at all. Tap the amber button if you want to jump in while it's speaking, and hang up whenever you're done. (The first tap will ask for microphone permission.)

NEW: in-app calls run on the phone engine now — fast replies, interrupt by just talking

The call screen got the real phone line's whole engine. Nothing to turn on — tap the phone button and it's just how calls work now:

• Replies start in a couple of seconds. The character speaks the first sentence while it's still writing the rest, instead of making you wait for the whole answer.
• Interrupt by just talking. No button hunting — start speaking and she stops mid-word, exactly like the phone line. (The amber button still works too.)
• Little "mm-hm"s don't derail her. Nod along out loud all you want; she keeps going.
• Spoken commands work. "Can I talk to Zadiana?", "speak faster", "slow down", "deep think on" — same as the phone.

Every call — web or phone — is saved as a written transcript under Call History in the main menu. Transcripts only; no audio recordings of you are kept from these calls.

And yes — there's a REAL phone line too. That's big enough to get its own page.

Seeing through your camera: video calls

On an in-app call (not the real phone line — a phone can't send video), a character can look through your camera while you talk. One camera button appears next to Hang Up once you're on a call — tap it and the character sees through your rear camera with the platform's sharpest vision: built for reading labels, mail, screens, and small details out loud, word for word, and for describing a room's layout so you can get oriented.

It's not a nonstop video feed — here's what it actually does

To be completely honest about how this works: the character isn't watching smooth, flowing video the way a person on a video call would, and it doesn't interrupt you the way a person would either. Two separate things are happening, and it helps to know both:

• In the background, it quietly refreshes its own notes every so often — roughly every 15–20 seconds, depending on the mode — just to keep what it "currently knows it's looking at" up to date. This part is silent. It does NOT speak up or interrupt when this happens; it's only updating what it would say if you asked.
• Separately, every single time you say anything at all, it grabs one more guaranteed fresh look before replying — so a direct question like "what am I looking at?" is never answered with stale information.

The upshot: it speaks when you speak to it, like any normal conversation — turn-taking, not a running commentary — with exactly one exception you control: a watch you've armed (next section). Outside of a watch, it never interrupts on its own; check in whenever you like — "is she here yet?" — and each check gets a genuinely fresh, current answer.

It also keeps a short memory of what it noticed changing — a few minutes' worth — so a check-in isn't limited to only the instant you happen to ask. "Did anything happen?" or "did a car pull up?" can be answered from something it noticed a couple of minutes ago, not just whatever's in frame right this second. It still won't volunteer that on its own; you still have to ask. Think of it as catching you up when you check in, not as it keeping watch and reporting back.

This is different from Describe My World, which is a one-time "here's a photo, describe it once." Video calls keep re-checking automatically for as long as the camera's on, without you having to re-share anything — more like having someone stay on the line looking with you (who only talks when you talk to them), less like snapping a single picture.

NEW: Your Spotter — live mode

On a video call, the radio-tower button (or asking for your Spotter by name) hands the call to your own live companion: continuous sight, instant replies, their own voice and personality — the one you design at kademurdock.com/spotter. Your character announces the handoff, and saying "live off" (or the button again) brings your character right back. Live mode uses a more expensive engine, so it has its own small daily allowance, separate from regular video minutes.

New: you can also call your Spotter directly — the radio-tower button sits right next to the phone button at the top of a conversation. One tap starts a fresh call and hands it straight to your Spotter.

NEW: Ask her to watch for something — the one time it WILL speak up on its own

This is the feature the fine print used to say didn't exist. Now it does. On a video call, say something like "tell me when a car pulls into the driveway," or "watch the door and let me know when the kids come in," or "tell me if the oven light turns off." Pets count too, if that's your thing. The character confirms, and from that moment a quiet automatic checker looks at the camera every few seconds for exactly that thing. The moment it's actually visible, the character takes one fresh, careful look and speaks up on her own — a real interruption, in her own voice, with what she's seeing. You asked to be interrupted; that's the one time she will.

• It waits for a quiet moment. An alert never talks over you — if you're mid-sentence or she's mid-reply, it holds until the line is clear, then speaks.
• One watch at a time, and it's one-shot. After an alert fires, the watch turns itself off — say "keep watching" or just ask again to re-arm it. Asking to watch a new thing replaces the old one.
• To stop early, just say so: "stop watching," or "never mind."
• It won't run forever. A watch that hasn't seen its thing after about half an hour says so out loud and stands down — it never dies silently while you're counting on it.
• Point the camera where the action would happen. The checker can only see what the camera sees — a watch for the front door works best with the phone propped facing the front door.
• Cost, honestly: the repeated checking uses the cheapest possible look (well under a nickel an hour), and the alert itself costs about as much as any normal reply. It all comes out of the same video minutes and balance as the rest of the call.

Getting oriented with HQ video

If you're blind or low-vision and using HQ video to get your bearings in a room, two habits help a lot: ask out loud as you go — "what's in front of me now," "what's to my left" — each question forces an instant fresh look, so it always matches wherever you've just pointed the camera. And pause a beat after you turn or pan before asking, so the look it takes isn't a blur. HQ is written to describe layout in plain terms — what's left, right, and ahead, and roughly how far — not just list objects, and to call out anything relevant to moving safely, like steps or obstacles, when it can see them.

Pointing a camera, if that's new to you

Not used to aiming a phone camera? A few basics: hold the phone with the camera lens (a small dark circle on the back, opposite the screen) facing what you want described. Keep it steady for a couple of seconds rather than sweeping it around fast — a still, held shot reads far better than a blur. Good light helps more than almost anything else; try not to point toward a bright window or lamp with the object sitting in shadow in front of it. If a description comes back confused or says it can't tell, that's the cue to hold it closer, steadier, or move to better light — the character will often ask for exactly that on its own.

Turning it off without hanging up

Once video's on, a third button (a camera with a line through it) appears — activate it and the camera turns off and stops using your daily minutes, but the call itself keeps going as a normal voice call. Handy the moment you're done showing something and want to save the rest of your allowance for later.

The honest fine print

• It costs more than voice, so it's not unlimited. Video uses real per-look AI vision costs behind the scenes, so it gets its own daily minute allowance separate from voice (voice chat itself stays unlimited either way). The first time you ever turn a camera on, you'll hear a one-time heads-up explaining the allowance — after that it just works.
• Nothing is recorded or saved. Only the single newest camera frame ever sits in memory for the length of the call — never written to disk, never kept after you hang up or turn the camera off.
• Camera permission required. The first time, your browser will ask to use your camera — say yes. If it seems blocked, check your browser's site settings for this page.
• Web calls only, for now. Video works on in-app calls in a browser or the installed app. The real phone line (1-833-530-0313) is voice-only.
• Watch alerts are the one sanctioned interruption. Armed watches (see the "ask her to watch for something" section above) are the only time a character will ever speak unprompted — because you explicitly asked for it. Everything else stays strictly turn-taking: no watch armed, no surprises, ever.
• It also remembers recent changes. Separate from watches, a short rolling memory (roughly the last five minutes) of anything that noticeably changed is kept, so a check-in like "did anything happen?" can catch you up on something you missed rather than only describing the current instant. That part still waits for you to ask — it just has more to say when you do.