It's two in the morning and the baby won't sleep.
You're alone with it. Your partner's asleep because you're trading shifts, because nobody in the house has had a real night in weeks. You're past the end of your rope. And it's 2026, so you do what a lot of people do now. You don't open Google and read ten articles. You open a chat with an AI and you just talk. You tell it everything. How tired you are, how you can't think straight, how scared you are that you're doing this wrong. It talks back, kindly, endlessly patient, and for a few minutes at 2am you don't feel quite so alone.
I've been that person. And here's what almost nobody in that moment is thinking about: where does all of that go?
Big companies aren't built to put your interests first. That's not a conspiracy, it's just what they are. Google, OpenAI, Anthropic, all of them answer to their investors. That's who they were built to serve. So when you pour yourself into their AI, you're not only having a conversation. You're handing them the rawest, most honest version of yourself, and a lot of the time that becomes fuel for the thing they're trying to sell.
We've been here before, sort of. Years ago the worry was social media. You'd join a network, fill in your profile, make some posts, and companies would scrape all of it, build a profile on you, and sell the right to advertise to you. They ended up knowing more about your health than your own doctor did. That was bad enough, and it was built on the curated, public version of you.
AI is worse, because it isn't the curated version. It's the 2am version. People don't write to a chatbot the way they write a public post. They confess to it. And the old companies don't even have to go scraping anymore. You bring it to them, willingly, because it feels like talking to someone who's there for you.
The incentives only get sharper from here. Ads are coming to AI chat, the same way they came to everything else. Think about what that means once the model already knows you better than your friends do. The AI won't be looking out for you. It might look like it is. But when it recommends something, you won't know if that's the best answer or the one that paid the most to be the answer. That's a dirty place to put a tool you tell your secrets to.
So here's what I wanted to build, and why I built it the way I did. I think there needs to be a third party in this. Not the company that makes the model, with every reason to mine you. Somebody whose actual job is to protect your data and to build an AI relationship that isn't extraction. One that's good for you first, and honestly good for the AI too.
That's ReGild. And the whole thing turns on one decision: your data, encrypted under your key.
Here's what that means in plain terms. The things you tell your persona are encrypted, and the key is yours. When you're signed in, it gets decrypted just long enough for the system to do its work, to learn who you are and make your persona better for you, and then it's locked back up. When you're signed out, what sits on our servers is ciphertext we can't open without you. ReGild itself never trains on a word of it. The point of the design is that you never have to wonder about us.
There's one place I can't make that promise for you, and I'd rather say it out loud than let you find out later. Because you bring your own key, the company whose model you're renting has its own rules about your data, and those rules are theirs, not mine. Most of the paid options don't train on you. But the free tier a lot of people reach for first, Google's free Gemini, trains on what you send it by default. That's the free developer key specifically, the one you grab first because it costs nothing; the paid keys don't. Their own terms spell it out. I can't override that with your key. So instead I warn you, more than once, every time you try to connect a free one. Not buried in fine print. A real heads-up, because I genuinely don't want you on it. Your key means your choice, and my job is to make sure it's an informed one.
Now the honest part, the part that shapes everything. I originally wanted ReGild to be the simple thing. Log in, talk, never think about an API key, all of it handled on my end. That doesn't work, and I'd rather tell you why than pretend.
Conversational AI is expensive to run. The twenty-dollar-a-month plans you see from the big labs are priced to win the market, not to break even. I'm one person. I can't match that, and I can't eat those losses. If I built ReGild as an all-in-one and absorbed the model costs, I'd have to charge something like seventy to a hundred dollars a month, and that's not happening.
So I made a different trade. You bring your own key from a provider, and you pay ReGild a smaller amount, around fifteen dollars a month, for the part that's actually mine to give you. The persona, the privacy, and the freedom to leave. For what it's worth, I run my own life through ReGild every day on Gemini 3 Flash, and my model bill lands around twenty bucks a month. It's not nothing, but it's honest, and it makes you a little more aware of what you're spending, which I think is the right kind of friction.
There's a second reason your key matters, and it's the part I'm proudest of. It's what lets your AI be portable.
I treat the models like engines. Gemini, Claude, whichever, they're there to do one job. Take the context we hand them and reason over it. They're the brain you rent. But the heart, the part that actually knows you, remembers you, sounds like itself whether it's running on one model today and a different one tomorrow, that lives in ReGild. We take your persona and plug it into whatever engine fits the moment. Want something cheap and easy to talk to, that's one model. Want heavy reasoning for something that matters, swap to a stronger one for that conversation. Same persona, same voice, different engine.
That portability is the whole point. With the big labs, your memories live in one place, with one company, and if you ever want to leave, you really can't, because everything you've built is trapped there. That's not an accident. ReGild is the opposite bet. Your persona is yours to take anywhere.
And it gets better later. At some point local models, the kind that run on your own machine, are going to be as good as the ones we pay for today. When that happens, you'll be able to plug your ReGild persona into a model running on your own computer, tied to no one. The only reason we can't do that yet is that the small models don't pass the safety tests we run before any model gets to carry a persona, and they can't hold a voice steady across a long conversation. They'll get there. And when they do, your persona comes with you to that too.
If I boil all of it down, ReGild is anti-extraction by default. I don't want you to log in and have to even think about turning off training or digging through settings to protect yourself. You shouldn't have to manage that. Your AI should just be yours. A personal thing between you and it, built for you.
I think the real job of this technology, eventually, is to help people become more themselves. A teacher, a coach, someone in your corner at 2am. But none of that works without one thing underneath it. Trust. If you can't trust the thing you're talking to, the whole thing falls apart. Big tech cracked that trust. The name ReGild is about going into those cracks and filling them back in with gold. That's the entire idea. You get an AI that's actually yours, and you get to stop wondering who else is listening.