YourVoiceCraft
The science of your writing voice

See what makes
your writing yours.

An AI that writes in your voice — and actually knows what you’ve written.
Anyone can prompt ChatGPT into a passable imitation. It can’t quote your March essay, catch the draft that contradicts it, or keep your unpublished writing off Big AI’s servers. A private model trained on your archive does all three.

cites your own posts catches self-contradiction never leaves our hardware
The tour

A Brief View

The heart of it, in under three minutes — the founder, live on the real product. Prefer a list? →

In Depth

The Full Tour

Every tool, shown live on the real product — nine minutes, no cuts around the rough edges.

The honest comparison

How is this different from ChatGPT or Claude?

They are magnificent generalists. This is a specialist.

YourVoiceCraft ChatGPT Claude
A model trained on your writing alone
Shows which of your posts an answer drew on
Admits it when you’ve never written about something
Flags a draft that contradicts what you’ve published
Scores every paragraph against your voice
Keeps dated checkpoints of your voice you can talk to
Your archive never touches a third-party AI
Chat about your writing with a model that has read all of it
Convene an automated dialog about your writing
A general-purpose assistant for any topic
Frontier-scale reasoning

Different tools for different jobs. They know everything — this knows you.

Free

Your free Voice Check-up

Paste your writing and get a measured report of what makes it distinctively yours — sentence rhythm, punctuation, signature vocabulary, where your voice breaks — plus a copy-paste prompt that makes ChatGPT or Claude write like you. See a sample ↓

Paste your writing — 1,000+ words works best
Analyzed on our own server — never sent to any third-party AI service.
Live now

Chat with your voice.

Once a model has learned how you write, something strange becomes possible: a conversation with yourself. It answers in your cadence, grounded in what you’ve written and what it knows about you. Ask it what you think. Argue with it.

what do I actually believe about first drafts?
YOUR VOICE
That they’re supposed to be bad. You’ve said it a dozen ways — the draft’s job is to exist, not to impress.
what’s the news?
YOUR VOICE
[this morning’s headlines, the weather, the markets — read back in your own cadence]
It grows with you

Feed your voice.

A trained voice isn’t static. Everything you do here feeds it — five ways in, one voice out.

Upload
Essays, posts, your Substack — what you’ve already written
Pin a fact
Mid-conversation — it treats your notes as true
Upload a document
Discuss the piece you’re working on right now
Be interviewed
It asks; your answers become training material
Grow
Every conversation makes it more precisely you
It doesn’t just write like you — it accumulates you.
Sample report

Here’s a sample Voice Check-up

This is an example, not your report — yours is built from your own writing. It shows what the free Check-up delivers: the stylometry that fingerprints a voice — sentence rhythm, rare-word rate, punctuation signature, and the words you reach for.

Sample · example dataVoice distinctiveness 82/100
Style fingerprint — you vs a typical writer
RhythmVocabularyRare wordsPunctuationFormalityVariety
This voice┄┄ typical writer
Avg sentence
17.4 words
tight, varied rhythm
Rare-word rate
6.1%
richer than most
Reading grade
8.9
plain, not dumbed-down
Punctuation signature
Em-dash
a signature move
Semicolon
used sparingly
Parenthetical
wry asides
Signature vocabulary
reckonunderneathplainlya certainratherthe thing is
Why our approach

Private. Efficient. Yours.

Three things a shared frontier model can’t give you — and the whole reason we built our own.

Private by design
Your documents run on our own hardware — never sent to ChatGPT, Claude, or any third party. Your drafts, your ideas, and your finished voice never leave our servers. And this is the hosted edition — for teams that need everything inside their own walls, we build truly local versions that run on hardware you control. Everything is encrypted in transit and at rest, and if you want us to keep nothing at all, Private mode discards your prompts and history. How we handle your writing →
Efficient, not metered
A small model built for one job runs for pennies — so you get one flat membership with no per-word API meter ticking. Rewrite, draft, and repurpose as much as you like.
Trained on you
Not a prompt describing your style — a model that has absorbed it. That’s what powers the gauges, the X-ray, and drafting that actually sounds like you.
“Frontier models are better at almost everything — except sounding like you and keeping your writing private.
Which is to say: the entire job.”
Who it’s for

For people whose voice is the point.

Newsletter & Substack writers
Keep your voice as you scale your output.
Essayists & bloggers
Draft and polish without sounding like everyone else.
Ghostwriters & content teams
Match a client’s voice, on demand.
Founders who write
Sound like you — not like a press release.
Academics & researchers
Plain, precise, unmistakably yours.
Anyone sick of AI sludge
Get the last mile a prompt can’t reach.
See the difference

The same line — anyone’s AI, or yours.

A generic model rephrases your words. A model trained on your writing rewrites them the way you would — your rhythm, your habits, your turns of phrase. All on our own fine-tuned 7B, on our GPU. No ChatGPT, no Claude, nothing sent to a third party.

You write
“The meeting ran long and everyone was tired.”
A generic AI
“The meeting extended beyond its scheduled time, and attendees reported feeling fatigued.”
A model trained on you
“We blew clean past the hour, and by the end you could feel the whole room sag.”
Try the demo →
Prompt vs. model

The prompt gets you close.
The model gets you there.

The free prompt is a great start — but it only describes your voice to someone else’s AI. A model trained on your writing becomes it. That last mile is the whole point — and the reason to own a model, not just a prompt.

How closely the AI matches your voice
65The prompt — close
93Your model — there
The free prompt
Paste it into ChatGPT or Claude. It roughly doubles how well the AI catches your voice — but it drifts over a long piece, can’t draft from a blank page, and your words ride through a third-party server.
Your own model
We retrain the model’s own weights on your writing — it doesn’t follow a description of your style, it has genuinely read and absorbed your corpus. Every sentence you upload nudges the network until your rhythm, diction, and habits are baked into the model itself — part of how it writes, not an instruction it’s handed. That’s why it holds your voice across a whole piece and can draft from a blank page, not just rewrite. Runs on our hardware. Never a frontier API.
Same sentence, two ways
You write: “The meeting ran long and everyone was tired.”
GENERIC AI, TOLD TO “SOUND DISTINCTIVE”
Purple, over-worked — the sound of every AI.
A MODEL TRAINED ON THE WRITER’S OWN CORPUS
Plain, specific, human — an actual voice.
Real, unedited output from our engine — same input, one model given a style prompt, the other trained on the writer’s corpus.

That demo you just tried? That’s a model wearing a voice, not a prompt describing one. Frontier models work from a description of your voice — your private model works from having absorbed it.

Why small & local beats big & borrowed

Wait — why not just use ChatGPT?

Because for this one job — sounding like you — a small model you own beats a giant model everyone shares. Frontier models are more capable in general; they’re the wrong tool here, not the weaker one.

Frontier model  vs  your voice
Sounds like you
Frontier
45
Yours
93
Keeps your writing private
Frontier
leaves
Yours
stays
Flat monthly price — no meter
Frontier
per-word
Yours
unmetered
It’s your voice, not a costume
Every model has a default voice — the flat, generic-professional register it reaches for when nothing stops it. A prompt only asks it to sound like you; that default is still underneath, and it resurfaces over a long piece. Yours was trained on your actual writing, so your voice becomes the default — the habits a prompt can’t name are baked into the weights, and it doesn’t drift back.
Your documents never leave
Every ChatGPT rewrite sends your unpublished drafts, ideas, and voice through someone else’s servers, logged and maybe used to train the next model. Ours runs on hardware we control — nothing is sent to a third party. For a writer, your voice is the asset; don’t lend it out.
It won’t flatten you
Frontier models are tuned toward one smooth, agreeable “assistant” register — the exact sameness you’re trying to escape. A model trained only on you has no pull toward the mean, and no ChatGPT tells. And it stays put: it won’t change under you when the next version ships.
Once your voice is trained

Your voice, put to work.

A trained voice isn’t just for rewriting. It becomes a measuring stick — and a compass. Every tool below is private, running on our own models.

Ideal readerSoccer momUrban DINKr/writingNewsletter fans
Targeting: whom does this target?

Paste a draft and a private 14B model works out who’d most want to read it — the ideal reader, standard marketing segments (Soccer mom, Urban DINK…), the exact communities to reach, and a pitch-ready hook for each. Alongside it, three engine gauges score the piece against your own voice — content, style, and distinctiveness.

Try it →
Appeal to Soccer moms
lessmore
Nudge who it’s for

Same page, one step further: name an audience to appeal more to — or less to — and it rewrites the draft to shift its reach, then re-writes the result in your own trained voice. Not the generic AI register: your rhythm, your habits. Changes from your original come back highlighted, so you see exactly what moved.

Try it →
A96% you
M91% you
Crawl & Index

We search the public web on your topics, read each result, and keep only the pages that match both your subject matter and your voice — articles that sound like you, not SEO sludge. Only topic keywords ever leave the building; the matching runs against your own corpus, on our models.

Try it →
88
61
34
Voice X-ray

Paste a draft and see inside it: every paragraph scored against your voice, color-coded — catch the tired stretch, the pasted-in AI passage, the editor’s flattening. One click re-writes a drifting paragraph the way you’d say it, and re-scores it to prove it.

Try it →
your samples your prompts & chats
your voice, truer
Adaptive: it keeps learning

Your model starts from your writing samples — then keeps getting truer as you use it. It learns not just from your samples but from your prompts and chats: what you ask it to write, the facts you feed it, the project you’re in. Mention cocker spaniels today, and next week “my favorite breed” already knows. Private to your voice — never pooled with anyone else’s.

See it →
sparse autoencoderscovered
grokkingcovered
reward hackinga gap ↑
Overview: your beat, mapped — and what it’s missing

We read your entire body of work, map your topics, and compare your beat against similar writers across the web. What do they cover that you never have? Who do they cite that you never mention? Every gap comes with a concrete piece to write — and one click hands it to your voice model to draft.

Try it →
Compared to you
87
Content
86
Style
Every piece scored against your own voice — content (is it your subject?) and style (is it how you write?). Measured, not guessed.
The X-ray, demonstrated
We trained the engine on 22,000 words of Mark Twain, then scanned four passages it had never seen. Real, unedited scores:
75
Twain — a passage the engine never saw
to my stateroom, he says: “Here—come in here.” And in he come, and Bill after him. But before they got in I was up in the upper berth, cornered, and sorry I come. Then they stood there, with their hands on the ledge of the berth, and talked. …
86
Twain — another unseen passage
somehow. I was the only one that could swim, so I made a dash for it, and Miss Hooker she said if I didn’t strike help sooner, come here and hunt up her uncle, and he’d fix the thing. I made the land about a mile below, and …
45
An AI told to “write like Twain”
Well, I reckon there ain’t nothing in this whole wide world quite so fine as a lazy river on a summer afternoon. A body can just float along, watching them fluffy clouds drift by, thinking about all manner of things and nothing at all. …
29
Generic AI prose aboutTwain
In today’s rapidly evolving world, it is more important than ever to embrace the timeless wisdom of classic literature. Mark Twain’s enduring masterpieces continue to resonate with readers of all ages …
The imitation says “I reckon” and “ain’t” — surface costume. The engine reads the rhythm underneath and isn’t fooled. Your voice model runs this same scan on your writing — showing you, paragraph by paragraph, where a draft drifts from how you actually sound.
Why we built this

I study how language models represent writing — and I got tired of watching them flatten every voice into the same smooth, agreeable register.

Prompts can describe your style, but they don’t change what the model is. So we did the harder thing: train a small model on your actual writing, on hardware we control, until your rhythm and habits are part of how it writes — no frontier API, no sending your words to anyone. The science of your voice, put to work for you.

Arshavir Blackwell, PhD · founder · I write about how models work at Inside the Black Box · agency & brand consulting
Questions

The honest answers.

Will it actually sound like me?
The free Check-up shows you, measured, before you commit. Membership trains a model on your own writing — not a prompt describing your style, a model that has absorbed it — so it holds your voice across a whole piece, not just a sentence.
Is my writing safe?
Yes. Everything runs on our own hardware. Your drafts, ideas, and voice are never sent to ChatGPT, Claude, or any third party, and we never train on your writing for anyone but you. And if our hardware still isn’t private enough: this site is the hosted edition — we also build truly local versions that run entirely on infrastructure you control.
How much writing do I need?
The Check-up works on a few hundred words; 1,000+ is better. For a trained model, more of your real writing means a truer voice — and you can keep adding to it any time.
What if it’s not for me?
The Check-up is free, no card required. Membership is month-to-month — cancel whenever, no lock-in.
Which model is this?
A small model, fine-tuned per writer, running on our GPUs — not a wrapper around a frontier API. That’s the whole point: for sounding like you, a model you own beats a giant everyone shares.
Plans

Pick your tier

Every plan runs private on our own hardware — your documents never leave our servers. Use it as much as you like; a generous fair-use policy keeps it that way.

Writer
$29/mo
One private voice + the full writing studio.
  • 1 private voice model
  • Voice Check-up & Voice X-ray
  • Rewrite, draft, continue & transform
  • Overview: your topic map & gaps
  • Monthly retraining
  • Unlimited use (fair-use)
Choose Writer →
Most popular
Pro
$59/mo
Three voices + the growth tools.
  • 3 private voice models
  • Everything in Writer, plus:
  • Repurpose: one piece → many Notes
  • Targeting & voice-scoped Search
  • Voice Mixer (experimental)
  • Weekly retraining
Choose Pro →
Studio
$129/mo
Ten voices, everything unlocked.
  • 10 private voice models
  • Everything in Pro, plus:
  • Train on your entire archive
  • On-demand retraining
  • Priority training queue
Choose Studio →

Pick a plan to create your account below — redeem a beta coupon for free, or add your card. No coupon yet? Leave your email and we’ll send one.

Get started

Ready to sound like you?

Your voiceprint — no two writers alike
Beta · invite only

Pick your tier, then start free with a beta coupon or add your card — a private model trained on your writing: yours, on our hardware, never a frontier API.

Enter a beta coupon to start free now. No coupon yet? Leave your email and we’ll send an invite (card payments open at launch). Your documents never leave our servers.
Already a member? Log in

Analyzing your voice…

 Feedback
© 2026 Arvoinen.AI LLC · YourVoiceCraft · the science of your writing voice · all rights reserved. The engine, prompts, and models are proprietary.
Your documents never leave our servers.
Terms · Privacy · Fair Use · Documentation · Features · Contact · Consulting · The Weeds
v0.1.1646