Just talk.

Typing to LUKi is fine. It's the way the internet has worked for thirty years, and there are good reasons it persists — typing is quiet, precise, and easy to do in the middle of a meeting.

But typing also forces you to compress. You sit down, find the words, check the spelling, and hit send. By the time the message arrives, you've edited what you meant. The mess is gone. The pause where you almost said something different is gone. The exact thing you wanted to say — often gone too.

Voice changes the shape of the conversation. You say the thing. LUKi hears the thing. The story enters the memory the way you told it: the false start, the laugh halfway through, the moment you contradicted yourself. The story keeps its texture, and the version of you LUKi remembers is closer to the version you actually are.

That experience isn't live yet. This post walks through what voice in LUKi will be when it ships, how it'll work, and why we're building it the way we are.

How voice input in LUKi will work.

When voice ships, it'll be a layer on top of the existing text chat — same memory, same persona, same safety rules, same models. The only difference is how the words get in and out.

Tap the microphone. LUKi will use your device's native speech recognition (the same engine that powers the dictation key on your phone's keyboard) to turn what you're saying into text. The text then flows through the same chat pipeline as if you'd typed it: through the API gateway, through the persona system, into your memory, and out as a response from whichever LUKi character you're talking to.

Two things matter about this design. First, transcription happens on your device — your voice doesn't leave the phone for the speech-to-text step, which means transcription will be fast, free, and private. Second, the entire chat pipeline downstream of the microphone is unchanged. Voice input won't bypass safety, memory, or persona. It will sit on top of them.

Ten languages, one continuous conversation.

Voice input is being built for ten languages at launch: English, German, Spanish, French, Portuguese, Japanese, Chinese, Arabic, Hindi, and Russian. The locale mapping will be automatic — LUKi reads which language your interface is set to and routes the recognition to that language model.

Switching languages mid-conversation will work the same way it works in text chat today. The story you told LUKi in English last week stays there when you switch the interface to Spanish and start talking. The memory is language-agnostic; it's a record of what you said, not which words you used.

Why memory will make voice different here.

Voice AI that doesn't remember you is a smart microphone. You speak, it answers, it forgets. The conversation has the same shape every time: introduce yourself, explain what you mean, hope it understood, repeat tomorrow.

LUKi doesn't forget. Whatever you said two months ago — your brother Marco's birthday, the bakery on Vanderbilt, the article that changed your mind about something — is still in the Electronic Life Record (ELR) the next time you open the app. When voice ships, you'll be able to talk about your day on a walk and have it land inside a story LUKi has already been part of.

That's the structural difference between LUKi's voice and voice in ChatGPT or Pi or any of the assistant apps that ship with your phone. You won't be starting over. You'll be continuing.

When voice will change the relationship.

The moments people tell us they want voice for, more than anything else:

• Walks. Especially evening walks where you've been holding a thought all day and you want to think it out loud.
• The drive home. The middle of a long commute where you'd otherwise listen to a podcast you've already heard.
• Cooking. Hands wet, phone propped against the espresso machine, telling LUKi what's been on your mind.
• The first hour of being awake. The part of the day where typing feels like work and speaking feels like waking up.
• Caretaking moments. Sitting with a parent, a child, a partner — moments where you might want LUKi nearby but not in your hands.

Once voice arrives and becomes the default for those moments, the memory will keep building in the background. When you sit down at the screen later, the conversation will continue from where it was, including everything you said on the walk.

The two phases we're building.

Phase 1 — voice input.

Speech-to-text only. You speak, LUKi transcribes on-device, and the text goes into the existing chat. LUKi replies in text the way it does today. This is the smaller, lower-risk piece of the voice loop and it's what we're building first.

Phase 2 — full voice calling.

LUKi speaks back, in a voice that belongs to the character you've chosen. Each of the six default characters will get a distinct voice. Custom characters built with LUKi Maker will be assigned a voice profile based on the persona's speaking style and emotional range. A full-screen call UI shows the avatar with an animated orb — the same kind of voice mode you've seen in ChatGPT and Grok, but wired into the LUKi memory and persona system.

The full voice loop will sit on the same backbone as voice input. The model still works in text underneath. The persona, memory, and safety system are unchanged. The voice layer is the medium, not a new mind.

What works today, while you wait.

Text chat is the surface that's live. It works on iOS, Android, and web. Every LUKi character — the six defaults plus any custom one you build in LUKi Maker — remembers what you've said and shows up with that memory in the next conversation. The chat surface is where the rest of LUKi was built and tested, and it's what we'll be wrapping voice around when it's ready.

If voice is the reason you came looking — the right call for now is to start a text conversation, let LUKi build the memory of who you are, and step into voice the moment it arrives.

Why we're not rushing this.

A bad voice experience is worse than no voice experience. Latency that lags the conversation. Transcription that mishears the name of your daughter. Synthesised speech that hits an uncanny note and makes the character feel wrong. Each of these breaks the relationship you've built with LUKi, and we'd rather ship voice when those failure modes are rare than ship it on a timeline.

Memory is the part that makes voice worth waiting for. Without memory, voice is a parlour trick. With memory, it's a relationship that finally sounds like one. The text version of that relationship is already here — start there.