Voice
learned from your own writingA skill that calibrates to how you write from your own past work — never smoothed toward the mean, never let the machine flatten it.
№ 03 · writing provenance · pre-launch
prove which words are yours
A thinking-and-writing platform where AI is disclosed, not hidden and not banned. Every paragraph you write carries a mark — human, AI, or edited — signed at the time you wrote it, so the split is provable, not just claimed.
In one line: worlds shows the human/AI split in what you write, instead of trying to detect it or ban it — and makes that split something anyone can check, not just trust.
A skill that calibrates to how you write from your own past work — never smoothed toward the mean, never let the machine flatten it.
Everything you bring in gets marked — your own words, transcribed, research material, AI output. Only your own writing ever trains your voice.
Each paragraph is signed and entered into an append-only log. The spec is open, the verifier is open-source, and nobody has to take our word for it.
It will never detect AI, and it will never ban it. It shows the split instead — that refusal is the whole point.
Brain dumps, journaling, source material — the platform learns how you think before it ever helps you draft.
AI drafts in your calibrated voice. Every unit of the result is tracked as verbatim-you, paraphrase-you, cited, or AI-net-new.
Each paragraph is cryptographically signed and entered into an append-only log — no blockchain, a Merkle log, the same pattern that already secures the web's TLS certificates.
The Provenance Sidecar Specification is open (CC BY 4.0) and the reference verifier is open-source (MIT) — anyone can check a signed manuscript offline. No worlds account required.
memoirists, journalists, Substack book-builders who want AI to sharpen a voice, not replace it — and a way to prove it.
a pre-publication check on an acquired manuscript that doesn't rely on anyone's benefit of the doubt.
the spec and verifier are public — reader-side tools, classroom checks, other provenance pipelines can build on the same log.
R&D active, pre-launch. The open spec and the reference verifier are real and public today — the verifier passes its full test suite against the canonical fixtures. The writing platform itself is still forming: voice calibration is built and waiting on a small author cohort, and the disclosure engine that tells verbatim-you from AI-net-new is merged but not yet validated against real writing samples. No launch date set.
worlds isn't open for everyone yet. Leave your email and we'll write when the room does — no lists, no noise, just word when there's something to try.
Draft copy — pricing and launch details publish once they're locked.