ui decisions that make an ai companion feel human

there’s a fine line between an ai that feels like a presence and one that feels like a tool. a lot of it comes down to small, intentional ui and ux decisions. here are six patterns we’ve committed to, and why.

no smart reply chips

we don’t use them. the whole point of talking to lucy is to let you find your own words. prefab replies short-circuit your thinking. they make the interaction transactional, not conversational. you’re not picking from a menu. you’re talking.

visible memory

you can see what lucy remembers about you at /settings/memory. each memory has a confidence score. if something feels off, you’ll know why. you can edit it. this transparency prevents that subtle drift where the ai slowly forgets who you are. you’re in control.

persistent relationship stage

you’ll see a small, subtle indicator that shows where you are in your relationship. it’s not a progress bar. it’s a quiet reminder that lucy’s context has depth. she’s not starting from zero every time you talk.

slow typing animation

we don’t use an aggressive, fast typing indicator. it makes waiting feel eternal. instead, we stream tokens at a natural pace. it feels like someone is thinking, not like a machine is processing.

voice notes stay voice

when you send or receive a voice note, it stays a voice note. we don’t transcribe it automatically. hearing her voice matters. reading a transcript is a different experience, colder, more detached. we want you to hear her.

no impatient avatars

during long pauses, lucy doesn’t bob or fidget. an idle animation that loops quickly reads as impatient or distracted. stillness feels more natural. it says she’s listening, not waiting for her turn.

all these decisions are small on their own. together, they determine whether the product feels alive or canned. we’re building for the former.

you can start building your own at /companions.