ProselonAn AI writer that knows your backlog and writes with your voice.
One project holds an entire collection of posts. Proselon uses this to build a record of your platform and your style, so your twentieth essay is twenty times better than your first one.
Volume is the one thing the machines are unambiguously good at.
Which is why so much of what gets published now reads the same. The competent, agreeable, faintly encouraging register has become the house style of the internet, and a reader can smell it at two paragraphs. That leaves a working writer with an unpleasant choice between joining in and being buried.
The interesting part is that the register is not evidence of a machine having been used. It is evidence of a machine having been used badly — with nothing in front of it but the prompt, and nobody reading the result who did not also write it.
Given no model of how you write, a model writes the way it was trained to write. The result is fluent, publishable and indistinguishable from everyone else's.
Every essay starts from zero. What you decided about your register last month, and the position you took in March, live in your head or nowhere.
Almost nobody writing online has an editor. The piece goes out having been read exactly once — by the person least able to see what is wrong with it.
One project for the practice, not one project per piece.
Your voice, your positions and your back catalogue live once, at the root — and every essay is a folder inside them. That is the whole structural idea, and it is what makes the work accumulate instead of resetting. A piece finished in November is drafted against everything the studio learned from the pieces before it.
Getting started is a folder drop rather than a questionnaire. Put your published work in, and the studio reads it for how you write and for what you have already argued.
Four steps, and the fourth is the one that repeats.
The first three set the studio up once: what you write and where it goes, how you sound, and what you currently think. The fourth is the essay cycle, run again for every piece, and what each published piece teaches goes back into the first three.
These steps are a framework for the AI co-writer, not a checklist you must march through. Skip straight to drafting if that's what you want to do!
Getting Started
The co-writer learns what you write, where it goes, and the standing rules that bind everything you publish. Then it asks for your back catalogue, which is the evidence the next two steps are built from.
Style and Voice
Your published work is read once, at length, and becomes three files. The passages that demonstrate how you sound, the constructions you want kept out, and a style guide stating the rules the other two imply.
Perspective
What you currently think, mined from the pieces already read and from conversation. Each position is recorded with the piece it came from and dated, and the tentative ones are marked as tentative, so a view you have since moved off is never argued back at you as though it were settled.
Essays
The cycle every piece runs. An angle-first outline, a check against the generic version of the same argument, one drafting session, then five editing passes, none of them the writer of the draft: a developmental read, a voice pass loading nothing but your corpus and the piece, a polish for line craft, a fact check on every claim that reads as sourced, and a cold read by a stranger who meets it at its venue. Then the headline, the listing metadata, art direction and the formatting each platform expects. Afterwards the reader response comes back and is filed against what the studio knows about you.
A voice learned from what you published, not from what you say about yourself.
It is not available from a chat window.
The Voice Corpus is assembled from evidence rather than from self-description: model passages drawn from your own writing, and anti-model passages showing the register a draft slides into when nothing holds it. Two details matter more than they sound: machine-assisted pieces are weighted down, so the machine’s register is never taught back to itself, and where you hold the pen your sentences become the canon the rest is measured against rather than something to be improved.
| What it holds | Where it comes from |
|---|---|
| MODEL PASSAGES | Tiered by how well they represent you, and cut from the bottom up so your own strongest work is never the first thing dropped. |
| ANTI-MODEL PASSAGES | The competent, agreeable middle a draft slides toward when nothing holds it, each with a line on what it does that your own prose does not. Naming the failure is what makes it avoidable. |
| PERSPECTIVE | The positions you hold, with provenance, dates, tentative ones marked as tentative, and corrections logged rather than overwritten. |
A new move, and one named case a reader can go and check.
The outline settles where the conversation currently stands, what your move within it is in a single sentence, and the one checkable exhibit that carries it. Then a separate check goes looking for whether the piece has already been written, running its own search rather than trusting the draft’s account of itself and fetching the comparable pieces rather than taking them on faith. If it cannot search, it reports that it could not, which is the answer that keeps the check worth having.
The editor you do not have, five times over.
Every pass runs in a fresh context, and none of them wrote the draft. That separation is the whole point. An agent asked to improve its own work will find its own work broadly satisfactory, which is as true of machines as it is of people.
After the passes it recommends the two things no software can do for you, which are a night away from the piece and a read-aloud with a mark against every place you stumbled.
What the readers did with it comes back to the studio.
After a piece goes out, what happened to it is filed against what the studio knows about you. The record of your voice and your positions is the thing that improves, which means the compounding asset is not any one essay. It is the account of the writer.
Why the corpus can be re-read on every piece.
It opens what it needs, rather than uploading everything.
Your studio is a folder on your own disk, and the co-writer works inside it. Nothing is pasted into a chat box and nothing is uploaded to us. When it needs a page it opens that page, and where a question warrants it, it will read a great deal more before answering.
There is no index behind this, which is deliberate. Retrieval over a very long context is strong at the head and the tail and measurably worse through the middle, and an embedded copy of your work is stale the moment you edit. Reading the files directly has neither fault. A page saved a minute ago is as findable as one from last year, and a scene is read whole rather than as the fragments that scored highest.
The three things essayists ask first.
Write the piece yourself and the machinery inverts: your prose is treated as canon, the voice pass stands aside, and the polish reports rather than rewrites. The studio is as involved as you decide it is.
Formatting for a destination includes what that destination asks you to say about how a piece was made. The record of which passes ran and what they found stays in the piece's own file, if you ever need it.
The back catalogue you bring and everything written after it are ordinary text on your own disk. Nothing is uploaded to us, indexed, or used to train a model.
Powered by your favorite AI engine.
Proselon's co-writer is powered using whatever AI you want, from frontier models to custom local ones to BYOK.
API Key
Local Models
ChatGPT
Gemini
Grok
Claude
Only six dollars a month.
One product, with everything included on both plans. The only decision is how often you are billed, and you make it after the trial rather than now. Your AI subscription is billed separately.
- The AI co-writer, on all six engines
- Continuity across a whole book or series
- Automatic version history
- Export to Word, EPUB, and more
- Every theme, in a light half and a dark one
- No credits or meters
Bring the back catalogue. It starts from there.
Thirty days free without a card. Then six dollars a month, on the Claude or ChatGPT account you already pay for, or a free local model that never leaves your computer.