# Whose Serve

> Read a text thread with someone you are dating or just matched with as a rally on a clay court - who opens, who asks, who writes more, who double-texts, and whose turn it is - then get three reply drafts (light, warm, direct) that fit what you want next. Free line calls flag asks for money, gift cards or codes, a push to move off the app, talk of an investment, and pressure or secrecy.

URL: https://whose-serve.skillsafe.ai/
API tutorial: https://whose-serve.skillsafe.ai/api.html

## Free, in the browser
- Reading the thread: one message per line as "Name: message"; times and dates in front ("[10/18/26, 7:02 PM] Sam: ...") are understood; a line without a name joins the message above. Two people only. A .txt export can be dropped in.
- Rallies: a new rally starts after a blank line, a date line, or a gap of 6 hours or more between stamped messages; the first message of each rally is its serve. Two or more messages in a row inside one rally are a double hit; a message with a question mark is a question.
- The court: their messages across the net, yours on this side, left to right in order; a ball sits deeper the longer the message; arcs over the net are returns, dashed loops are double hits; serves are ringed, questions marked, line calls crossed.
- The carry share: the average of your share of the words, of the questions (when anyone asked) and of the serves (when there are two or more rallies). A rough rule of thumb - an estimate of effort on the page, not of anyone's feelings.
- The verdict rule, checked in order: wait (3 or more messages from you since their last), you-carry (you carry 65% or more), they-carry (35% or less), even (everything else).
- Line calls: plain word patterns over their messages only - a hit is a reason to look twice, not proof.
- Exports: Markdown, JSON and a rally sheet (.csv, one row per message); kept reads (in the browser, or in your account when signed in).

## Metered (gpt-terra), one task: read
The model gets every count from the page and does no arithmetic of its own. It returns the verdict by the rule above, 2-5 hedged reads that each quote one message word for word, three reply drafts (light, warm, direct) aimed at the goal you pick (keep it going, suggest meeting up, slow things down, clear the air, end it kindly), up to three things to avoid, one question to ask yourself, and a care note whenever the page made a line call. It never diagnoses or labels anyone, names no app, uses no he/she, and adds no plan, place, day or backstory the thread does not state. The page re-checks every reply: the verdict against its own rule, every quote against its message, the three drafts, the care note, and every number, capitalised name and field name against the input.

## Sources
Built on the SkillSafe skill @erichowens/partner-text-coach - https://skillsafe.ai/skill/@erichowens/partner-text-coach - (a communication coach for partner and relationship texts: emotional subtext, thoughtful replies, de-escalation), narrowed to one question: whose turn is it, and what is a good next shot.
The trend note on the page is from Wikimedia pageviews of the English Wikipedia article "Cuffing season": weekly views rose from 1,034 (September 1-7, 2026) to 3,597 (October 6-12, 2026).