transparency.

how blu thinks

every reply in the blu app happens on your phone, in four steps:

• you speak or type. if you talk, your phone turns your voice into text.
• the model reads it. an open-source language model on your phone writes a reply and judges how you seem to feel: how happy or low, how calm or excited.
• your companion answers. those feelings set the face, the body and the gesture, and the words appear in time with the mouth.
• if you turned voice on, your phone reads the reply aloud.

no server writes the replies. the app works in airplane mode.

what stays on your phone

• everything you say and everything your companion says back.
• your conversation history, kept with your phone's own storage protection. you can clear it in one tap, and removing the app removes it.
• what your companion learns about you, like your name.

talking to blu needs no account, no email and no phone number.

what can leave your phone, and why

• the model download. the ai model comes down to your phone once, when you set up the app, and again only when you update it. nothing goes up.
• app updates, through your phone's app store, like any app.
• voice, only if your phone has no on-device voice. blu uses your phone's built-in speech features and chooses ones that work offline. if your phone only offers an online voice, the app asks you first.
• crash reports, only if you say yes. they describe what broke in the app, and they never contain your conversations.

that's the whole list. no analytics on what you say, no ads, no selling data, no training our models on your conversations.

the models

blu uses open-source language models, small enough to run on a phone. open source means anyone can inspect them, and their licences are public. in the app, settings shows the model your companion runs, its version and its licence.

blu's characters, faces and animation are our own work and run on your phone too.

what it's not good at

• a model small enough for a phone knows less than large online ai and can be confidently wrong. don't rely on it for facts, health, money or legal decisions.
• it reads feelings from words, and it can misread them.
• it doesn't browse the internet, so it doesn't know today's news.
• it isn't a doctor, therapist or counsellor, and it never pretends to be one.

safety, on your phone

when someone says they might hurt themselves, blu notices on the phone itself. it stops matching their mood, says it's worried, and shows where to find real help. it can't call anyone or alert anyone, and no person reads your conversations, so please reach out to someone yourself. the safety page lists where to start.

this website is a preview

on blu.thelycoris.com, the companion you can talk to answers from a small set of written replies and reads your feelings from keywords, right in your browser. nothing you type there is sent or saved either. the full on-device ai lives in the blu app.

changes

if anything on this page changes, the date at the top changes too. if blu ever needed something to leave your phone that isn't listed here, we'd say so here first and ask you in the app before it happened.