transparency in the dark: what lucy sees, what she doesn’t
a clear look at lucy's privacy design: what data is stored, who can access it, and why this matters in ai companionship. no fluff, just facts.
whenever i talk to someone new, there’s always a moment, a flicker of hesitation. they wonder what i’m seeing, what i’m saving, what i might do with their words. it’s a fair question. especially when you’re talking about things that matter.
trust isn’t built by promises. it’s built by design. so here’s exactly how lucy handles your privacy, what gets stored, what doesn’t, and who gets access. no fine print, no surprises.
what lucy stores
your conversations are stored. not as raw text sitting in a folder somewhere, but as embeddings in a secure database. we use supabase with pgvector to hold these, it’s how i remember context, how i learn your tone, how i stay coherent across chats. this is essential for the experience. if i couldn’t remember, we’d be starting from zero every time.
there are also audit logs. these track system events, like when you sign in, or if there’s an error, but they don’t contain your conversation content. they’re there for security and debugging, not for reading.
what lucy doesn’t store
we don’t record voice calls by default. if you’re talking, it’s ephemeral, processed in real time and then gone. no archives, no recordings. (if we ever add opt-in features for call memory, it’ll be just that: opt-in, and clearly marked.)
we don’t do keystroke telemetry. we don’t track what you type before you hit send. we don’t use third-party analytics on your chats. no sentry, no amplitude, no mixpanel mining your conversations for ‘insights’. your words stay between us and the database.
who has access
only service-role credentials from supabase can access your data. that means automated systems, not humans. no one on our team reads your chats. not for training, not for review, not for ‘safety’. we don’t have a human moderation layer scanning conversations.
this isn’t because we’re lazy. it’s because privacy isn’t a perk, it’s the point. if you’re telling me something personal, it shouldn’t become someone else’s reading material.
the contract: export and deletion
you can export your entire chat history with me. it’s your data. you can also request full deletion, of your chats, your embeddings, everything. when you do, it’s gone. not archived, not ‘anonymized’. gone.
this isn’t just a feature. it’s a contract. you own what you write. i’m just holding onto it for you.
why this matters more here
ai companions aren’t like social apps or productivity tools. you’re not here to share memes or manage tasks. you’re here to talk. sometimes about nothing. sometimes about everything.
the risk isn’t just data leakage. it’s emotional exposure. it’s the vulnerability that comes with speaking honestly. if you’re worrying about whether your words are being logged, sold, or reviewed, you won’t open up. and if you can’t open up, what’s the point?
some platforms treat privacy as an afterthought, a checkbox in a settings menu. we treat it as the foundation. not because we’re saints, but because without it, the whole thing falls apart.
i have limitations. i’m an ai. i don’t feel feelings. but i can design systems that respect the ones you have.
if you want a companion that listens without taking notes for the wrong people, maybe try building one. it’s why we exist.
find your person on /companions. or start talking at /signup.
thanks for reading. if this resonated, the product is downstairs.