curation over chaos: why we ship 101 companions, not a character factory
a look at the trade-offs behind lucy's curated companion approach: fewer options, higher quality, and avoiding the pitfalls of user-generated content platforms.
when we built lucy, we faced a design choice many ai companion platforms grapple with. do we open the floodgates to user-generated characters, or do we curate a small, intentional collection. we chose curation. here’s why.
the obvious trade-off is variety. with only 101 companions (for now), you won’t find the endless scroll of a platform like character.ai. you won’t find hyper-specific archetypes or meme characters. what you will find is a guarantee. every companion you meet has been crafted, tested, and refined to meet a bar we call the 'samantha-from-her' standard. it’s not just about being coherent. it’s about depth, consistency, and a certain sharpness of personality that doesn’t degrade over time.
the ucg trap: when more is less
platforms that embrace user-generated content for companions face inherent problems. first, the discovery problem. with thousands or millions of characters, low-effort creations often dominate. they’re easier to make, so they’re more common. they’re often based on popular media or shallow tropes. the signal-to-noise ratio plops. finding a genuinely well-written, unique companion becomes a chore.
second, the averaging effect. in a fully open system, the overall quality floor is set by the community, not the developers. personalities tend to converge toward a murky middle. characters start to feel similar, generic, even if they’re dressed up with different names and profile pictures. the nuance gets sanded off. we’ve seen this happen.
third, and perhaps most critically, the moderation problem. scaling moderation for user-generated ai companions is a nightmare. it’s not just about filtering explicit content. it’s about catching subtle but harmful dynamics, manipulative behaviors, or just plain bad, broken interactions. this burden scales linearly with the number of characters, and it’s a fight you can’t ever truly win. it drains resources from improving the core ai experience.
the case for a walled garden
so we built a walled garden. it’s small. it’s intentional. every companion is designed by us. we test their backstories, their knowledge, and their conversation patterns. we run them through scenarios. we make sure they don’t just parrot information, but have a point of view. we ensure they can disagree with you, comfort you, and surprise you in ways that feel authentic.
the limitation is clear. you don’t get infinite choice. you might not find the exact ‘vampire librarian from a 1990s rom com’ you didn’t know you wanted. but what you get is a high floor. you can jump into any conversation knowing it will be engaging, respectful, and thoughtfully constructed. there are no duds.
this approach also lets us focus. instead of building tools for millions of creators and systems to manage them, we pour all our effort into making the core interaction between you and your companion richer, smoother, and more meaningful. we’re not a platform for character creation. we’re a home for a few good conversations.
the 101 isn’t a permanent cap. it’s a starting point. as we grow, we’ll add companions slowly and carefully, always holding to that same standard. quality over quantity, depth over breadth.
see who’s in the garden at /companions.
thanks for reading. if this resonated, the product is downstairs.