100 ways to say the things that matter — in 30 seconds.

No. It uses Claude under the hood — but the value isn’t the generation. It’s the five questions we ask per moment, in the order we ask them, with the hard question (different for each of the 25 moments) that no chatbot would think to ask. Plus the inference layer that detects your generation, your region, your relationship-history, and produces output that sounds like you. Strip those out and you’d have a worse chatbot. Add them back and you have a product.

Pick “not really” on step 3. We’ll write you a repair message instead of an apology — words that acknowledge the impact without confessing to something you didn’t do. Most apology tools can’t do this. It’s the case we’re proudest of.

Yes. We bake banned phrases (“I’m sorry you feel that way,” “sorry if,” “sorry but,” and six more) directly into the system prompt, so they never appear in your output. Every option must do three jobs at once: name the harm, take responsibility, close the loop. If a draft fails any of those, the model rewrites it.

Because most people only need this a handful of times a year. A subscription would be a worse fit for the moment — you’d feel obligated to use it. One-time payment matches one-time use. If we’re wrong and you do use it constantly, $9.99 once is already a deal.

We store the metadata of generations (which moment, which tone, which relationship category) to enforce the free-tier cap and improve outputs. We do not store your context sentence or the AI’s generated apologies after the session ends. Your draft isn’t training data and never will be.

Regenerate. Free, unlimited, no penalty. Most users get the one they send within 1–2 generations.

Yes — within 7 days, no questions. Reply to your Stripe receipt and we’ll refund. We’d rather lose $9.99 than have you carry a grudge about it.

/no (declines & boundaries), /thank-you, /condolences, /goodbye, /space, and /money. If you bought all-access, every one of those is already unlocked for you the day it ships.