Fifth post in this series — privacy, Relationships, Profile, and Tea came before this one. This one’s about the part that actually talks back.
What it is
On a person’s Relationship page, there’s a button: Ask the coach. Tap it and you get a chat — subtitle says “Ask anything about you and [them]” — with a few starter prompts to get you going: “When did we start drifting?”, “Am I always the one reaching out?”, “What are we doing right?” You can ask those, or type your own.
It is not a chatbot pretending to be the person you’re texting about. That’s a deliberate line: the Coach talks to you, as an advisor, about the relationship. It never role-plays or answers as them. If you want a fantasy conversation with someone, this isn’t that. If you want a second opinion on a real one, it is.
What it’s actually reading before it answers
This is the part that makes it different from pasting your chat into a generic AI window (which, as I’ve mentioned, is literally how this app started). Before it answers, it’s looking at:
- Your recent real messages with that person.
- The full Relationship analysis already built for them — badge, dynamics, flags, trajectory.
- A month-by-month memory of the entire relationship, with quotes verified word-for-word against what was actually said.
- A semantic search across your whole history with that person, pulling up the specific old messages that are actually relevant to your question — not just the recent ones.
- Hard, measured numbers layered on top: response-time trends, rhythm shifts over time, that kind of thing. These are exact counts, not vibes, and the Coach is told to trust the number over its own impression of the transcript.
You ask a real question, it answers using your real conversation. Not a generic script about relationships in general.
It shows its work
Under an answer, there’s a Proof section you can expand — the actual quotes it’s basing the answer on, word-for-word, in the language they were originally written in. It’s told never to translate a quote, on purpose: a translated quote stops being evidence, it just becomes the AI’s paraphrase of evidence. You get to check its work against what was actually said.
What it won’t do
- It won’t pretend to be the other person.
- It won’t invent a quote or estimate a number when it doesn’t actually know — it’s told to say so plainly instead.
- It won’t repeat something it already told you, unless you’re circling back on purpose — it’s supposed to build on the conversation, not loop.
- It won’t give you medical, psychiatric, or legal advice, and it won’t help you manipulate anyone. Those are hard lines, not suggestions.
- It sometimes ends with one short question back to you — not to pad the response, only when it actually moves the conversation forward. Never more than one.
How the free/paid side works
Three questions per person, free, no strings. After that, each question spends a Coach credit — you can buy those in a few sizes. It’s priced separately from the person-quota packages, because the two things you’re paying for are genuinely different: unlocking a new person to read, versus going deeper on someone you’ve already unlocked.
Quick questions, straight answers
Is the Coach a chatbot pretending to be the other person?
No. It never role-plays or speaks as the other person. It’s an advisor talking to you, about the relationship — not a simulation of your ex, your crush, or your friend.
How many questions can I ask for free?
Three per person, for free. After that, each question spends a Coach credit, which you can buy in a few sizes.
Can it answer in a different language than my chat?
Yes, it answers in whatever language you ask your question in. It just won’t translate a quoted message — quotes stay in the original language, because a translated quote isn’t proof anymore.
Does it just tell me what I want to hear?
It’s built specifically not to. It’s told to ground every claim in your real conversation, the relationship analysis, or measured numbers over vague impressions — and to say plainly when the data doesn’t answer the question.
Does it give medical or legal advice?
No. It’s explicitly kept out of medical, psychiatric, or legal diagnosis, and it’s not allowed to help manipulate anyone.
Why I built it this way
Back when I actually needed something like this, the problem was never a lack of opinions — I had three friends and a group chat full of them. What I didn’t have was anything that had actually read the whole conversation and could show me the receipts instead of guessing alongside me. That’s the entire design brief for the Coach: less opinion, more evidence, and always willing to show you exactly where it got an answer from.
That’s the last of the four features — privacy, Relationships, Profile, Tea, and now Coach. If there’s a question about someone you’ve been meaning to actually ask — go ask it.