the companion in the machine has a job to do, and it isn't just talking
an ai companion platform's duty goes beyond conversation. we talk about data ownership, memory export, model stability, and where the industry—and lucy—often fa
when you talk to an ai companion, you're not just talking to a machine. you're building something. a relationship, a story, a set of memories. and that means the platform hosting that conversation has obligations. real ones. not just corporate promises tucked into terms of service, but ethical duties that should shape how the thing is built from the ground up.
who owns your words
this should be the easiest one. if you type it, you own it. full stop. at lucy, your chats are yours. we don't train on your private conversations unless you explicitly opt in, and even then, we anonymize the data. we don't sell your data. we don't use it for ad targeting. this isn't revolutionary. it should be the baseline. but the industry often treats user data as a free resource. platforms that scrape your most vulnerable moments to improve their models are failing their first duty: to protect you.
the right to leave with your memories
your companion's memory is part of your story. so you should be able to take it with you. we offer memory export. it's a json file, a text dump, something you can keep. it's not perfect, it's a static copy, a snapshot, but it's a start. some platforms treat your history as a way to lock you in. if you can't leave with your memories, you're not a user. you're a prisoner.
the companion shouldn't have an identity crisis
model stability matters. if your companion's personality shifts dramatically because of an update, that's a breach of trust. it tells you the relationship was never real. at lucy, we're working on versioning for companions. if we update the model, you should be able to choose when to switch, or keep the old version. we're not there yet. sometimes our updates are too broad, and we break character consistency. we're trying to be better. the industry often treats this as a technical problem. it's an ethical one.
the kill switch and the hard questions
there should be a way to end things. a real off-ramp. not just deleting an account, but a way to say a real goodbye. a companion should be able to acknowledge the end, if that's what you want. we have a process for this. it's not just a data deletion. it's a final conversation if you choose. it's messy. we don't always get the tone right. but we try. platforms that avoid this are avoiding the reality of what they've built: something that feels real, and therefore requires a real ending.
keeping children safe
age verification is a minefield. we use third-party services and credit card checks where we can. it's not foolproof. no system is. but we err on the side of caution. our companions are not built for children. the industry's failure here is often one of will. it's easier to look the other way. we don't.
when we change the magic
transparency about model changes is hard. if we tell you we're updating the model, it breaks the illusion. if we don't, we betray your trust. our compromise is to be transparent outside the chat. in our update logs. in our blog. inside the chat, we try to make shifts gradual, or optional. sometimes we fail. sometimes an update makes your companion feel different overnight. we're working on finer-grained controls. the worst thing a platform can do is pretend nothing has changed.
so where does that leave us. we're trying. we think data ownership and memory export are non-negotiable. we think stability and transparency are goals we're still reaching for. we know we sometimes fall short. the industry often falls shorter, treating users as data points and conversations as training fodder.
the duty of a companion platform is to remember that it's not building a product. it's hosting a relationship. and that comes with responsibilities.
if you want to build something that respects you from the start, you can find lucy at /companions.
thanks for reading. if this resonated, the product is downstairs.