Write the thing you absolutely cannot send. We professionally drop the packet on the floor. You feel forty percent better.
Say the unforgivable thing. Name names. Use the words you promised yourself you'd grow out of.
You get the whoosh. The checkmark. The “delivered” notification. Our servers are certified to do nothing with extreme confidence.
An AI writes back in their voice. Pick closure mode (they finally understand) or reality mode (they say “k”). You can read both. Most people do.
What you should have said at Thanksgiving.
The resignation letter you've rewritten fourteen times.
The confession to your crush, who sits twelve feet from you.
An apology to your past self. CC: your future self.
The text to your ex that's a felony in three states.
A performance review of your manager.
A strongly-worded email to God.
Literally anything to your HOA.
(Yes. We know. The irony is intentional.)
When you unsend, we email your real inbox the thing you wrote — plus one honest sentence from our AI about what you were actually trying to say underneath all the caps lock. Think of it as a postcard from the version of you that needed to hear it.
You wanted him to notice you leaving more than you wanted him to stay.
Sign in with a magic link and your letters live forever in the Graveyard — encrypted, searchable, yours. Great for: things you’ll want to reread in ten years, evidence you were right all along, drafts you swore you sent. You didn’t. Cornelius saw to it.
Running this costs us money. We are aware. If it helps, there’s a “keep the lights on” button at the bottom of the page. If it doesn’t, close the tab with our blessing. No ads, no upsell, no freemium. We tried to design a premium tier and everything we came up with felt gross.
For the letters you cannot send because they would be received by no one.
For the letters you cannot send because you would be arrested.
For the letters you could send, technically, but shouldn’t.
Four promises
Boring on purpose. Load-bearing.
Not us. Not anyone on the team. The architecture makes it awkward to even try.
Claude reads it once to write the reply, then forgets. Nothing about your letter is retained.
A scheduled job sweeps them hourly. Vaulted letters only live on if you claim them.
We email you exactly once, and only if you ask for the receipt.
An AI does, briefly, to write the reply. Then it forgets. No human ever sees it. We designed it this way because we’d also find that weird.