Why this exists

Every tool for meeting people was built around a category. Dating apps for romance. LinkedIn for colleagues. Meetup for groups. Bumble BFF for friendships, as if friendships needed their own lane. The result is that you manage five accounts, five identities, and five different sets of rules about who gets to see what about you.

Connection doesn’t work that way. A friendship can turn into a business partnership. A mentor relationship becomes a collaboration. The best people in your life probably don’t fit into a single category, and the moment you met them didn’t announce itself as one type of meeting rather than another.

MeetPeeps is a single interface for all of it. You tell one concierge what you’re looking for — not by filling out a profile, but by talking — and it handles the matching, the context, and the introduction. One place. No switching.

The concierge is a matchmaker, not a companion. Its job is to understand what you’re looking for, find someone worth meeting, and make a warm introduction — then step back so the two of you can take it from there.

The relationship that matters is the one you build with the person you meet. Everything the concierge does — the questions, the matching, the introduction — is in service of that first real conversation and whatever grows out of it.

Privacy is enforced in tiers, not by policy. When a match is proposed, neither person sees the other’s name, photo, employer, or contact information. What they see is a description — enough to decide whether this person sounds interesting, without enough to search for them on the internet. Contact details only move when both people have spoken, and only if both want them to.

This is true regardless of the type of connection. Romantic intent is explicitly in scope, and it gets the same protection as every other category, not less. The stakes of a bad introduction are highest in the romantic context, so that’s the standard we build to.

We run the AI ourselves. Most products like this are a thin layer over someone else’s model — what you type is shipped to a third-party AI company and processed on their servers, under their policies. The things you tell a matchmaker are exactly the things that should never make that trip. So the model that reads your words runs on hardware we operate. When you talk to the concierge, the reply you read is generated on our infrastructure — no outside AI company generates it.

The rest of the stack — hosting, database, text messages — uses vendors the way any service does, and the privacy policy lists each one and exactly what it receives. One entry belongs in this paragraph, not in fine print: to improve the model, we have small anonymized samples of concierge conversations graded offline by an outside AI company — never as part of a live conversation, never tied to your identity. The model you talk to stays ours. No AI company sits between you and the person you’re trying to meet.

The first call is a short audio conversation — no video, no text exchange beforehand. You have a real conversation with someone you don’t know, and at the end you both decide whether to go further. This is intentional. A conversation is harder to fake than a profile. It’s also faster: you learn in fifteen minutes what a message thread takes weeks to reveal.

We started in San Francisco because it’s a city full of people who moved here to build something and ended up more isolated than they expected. That’s a solvable problem. Most of what makes it hard is infrastructure, not desire.