I was trying to send one decent message. The kind that responds to what someone actually said, notices the strange little detail in their profile, and doesn’t sound like it was produced by a committee of LinkedIn influencers. A message that is confident without performing confidence. Playful without becoming a clown. Interested without turning into an interview. That should be easy. It wasn’t. Dating apps give you a handful of photos, a prompt, and a conversation that can disappear into the scroll. I kept moving between the profile, the chat, my own half-written thoughts, and the growing suspicion that I was either overthinking everything or not thinking enough. The problem wasn’t a lack of words. It was friction. What did she actually say? What had I already asked? Was there a thread worth following? Was I replying to the person in front of me, or to some imaginary version assembled from three prompts and a blurry photo? I tried the obvious thing. Ask an AI. But generic advice was exactly that — generic. “Be authentic.” “Ask an open-ended question.” “Reference her interests.” Wonderful. The machine had discovered dating. I wanted something grounded in the actual conversation. Something that could say: this idea came from that message, this profile answer, and this thing you wrote about yourself. Or: we don’t know enough yet. Don’t invent a personality. Don’t turn one sentence into a psychological profile. Then the privacy problem arrived. I didn’t want an invisible cloud service quietly collecting intimate conversations and building a permanent little dossier of my romantic failures. I didn’t want a bot logging into Bumble or Hinge, swiping, sending messages, or impersonating me. That would remove the hardest part of dating while preserving all the awkwardness. I wanted a thinking tool. Something local. Something visible. Something that showed its work and left the decision with me. That became Match Box. The first idea was simple: read only the visible screen, with my approval. Mirroir could act as a small bridge to the screen. Swift could store the observations. SwiftData could keep the approved history on the Mac. Apple’s Foundation Models could turn that context into structured suggestions — openers, callbacks, follow-ups, closers, bonding ideas, even a kind roast when the moment deserved one. Not a black box. A matchbox. A small place where a conversation might find a spark. The more I built, the more the product changed. It stopped being “AI writes dating messages” and became something closer to a conversation desk. A place to review what happened, understand what I actually wrote, notice where a conversation stalled, and decide what to do next. Machiss is the name now (English:Match Box). The product still has rough edges. The Mac app still needs polishing. The model still needs to know when it doesn’t know. The capture flow must remain read-only. No swipes. No likes. No scraping. No sending. That boundary is the point. Machiss doesn’t promise to make someone fall for you. It can’t manufacture chemistry. It can’t replace attention.It can reduce the friction between noticing something and saying something real. That was enough of a reason to build it.

Machiss

Conversational dating, grounded in stories and conversation

Conversational dating, grounded in stories and conversation