You are reading more machine-written words than you think.

Not obviously — the good ones don't announce themselves. They sit in the same fonts, the same paragraphs, the same confident voice a person would use. Somewhere on the web right now a model is finishing a sentence a human will sign their name to, and neither you nor the signer will be able to point, later, to the seam where one ended and the other began.

That seam matters. It is the difference between a writer used a tool and a tool wore a writer's face. We have spent two years building machines that blur it, and almost no time building anything that keeps it honest.

inkstone keeps it honest.

What it is

inkstone is a publishing platform — a headless CMS — with one idea built into its spine: every post carries a verifiable record of who actually wrote it. Not a vague "AI-assisted" badge bolted on after the fact. A mark, pressed into the page, that a reader can check.

When you write a post in inkstone, every revision records its authorship — handwritten, collaborative, derived, and so on — along with the people and the models that touched it. A post written by your own hand gets a wax seal. A translation produced by an AI says so plainly, naming the original it came from and the model that made it. Nothing machine-made passes quietly as yours, because the system was designed so that it can't.

Don't trust it — check it

Here is the part that makes inkstone different from a promise on a marketing page.

Every revision is SHA-256 hash-chained to the one before it. The hash is computed from the post's own content and authorship — a fixed, versioned projection of exactly the fields that constitute the work. Change a published post's history and the chain breaks, visibly, for anyone who looks.

And anyone can look. The proof travels with the content: a reader, an archivist, a fact-checker can recompute the hash from a post's public data and confirm the history is intact — without trusting inkstone, or even talking to it. Provenance you take on faith is marketing. Provenance you can recompute is infrastructure. inkstone ships the second kind.

Write with AI — out loud

The honest answer to "should writers use AI?" is of course they will. inkstone's bet is that the problem was never the tool; it was the silence around it.

So inkstone makes AI a first-class, fully-labeled collaborator. Write in one language and its editorial agent drafts the translation — in your voice, following your glossary — and drops it into a review queue for you to approve. Approve it, and it publishes as a derived work, its lineage recorded: translated from your original, by this model, reviewed by you. The reader sees all of it.

The agents are yours to shape. Their personalities — the translator's restraint, the editor's bite — are plain text files you edit. Bring your own model, your own keys, your own voice. The platform carries the provenance; you carry the taste.

Honest, open, and yours

Why now

The web is filling with confident, frictionless, unaccountable text. The reflex has been to build detectors — to catch the machines after the fact, in an arms race nobody wins. inkstone takes the other path: not detection, but disclosure you can verify. Stop asking "was this written by an AI?" and start being able to prove the answer.

A blog that can show its work. A platform that would rather be honest than seamless.

Sealed by hand. Verified by math.

inkstone is open source. Read the source, or run your own: github.com/nch3ng/inkstone. This post was drafted collaboratively — a human outline and edit, an AI first draft — and is labeled exactly that, because that is the whole point.