Conversation
Signals
AI readingRead against the rubric, which you can audit and edit.
Drafts
Written to the style rules, which you can audit and edit.
If you send that
model estimateThreads
Your outcomes
measuredLog whether each message got a reply and this becomes real data about what works for you — which is the only sample that matters.
Model
Your data
Everything lives in this browser's localStorage. Your key is sent only to the provider you chose. There is no backend and no analytics.
Readings are cached, so analysing the same thread twice gives the same answer instead of a re-worded one. Clearing the cache makes the next Analyse call the model again.
What this won't do
- Write to someone who said no, asked for space, or stopped replying.
- Manufacture jealousy, urgency or scarcity as leverage.
- Anything involving a minor.
- Impersonate anyone.
- Model "how women think" — it reads one conversation, not a gender.