The
landing page tells the story and the
guided tour below shows it in detail — this is the inventory.
Everything runs on our own models; nothing you write is sent to any third-party AI. This is the hosted
edition — a truly local build, running entirely on
hardware you control, is available as an engagement.
The full feature tour — nine minutes, every tool shown live on the real product. The three-minute version is on the front page.
The more writing you give it, the better everything gets —
the trained voice, the scores, the grounding all sharpen as your corpus grows, so connect your full archive rather
than pasting a sample. And if your voice has changed a lot over the years, instruments like the Voice X-ray
judge you against your whole record — your newest register can honestly read as “drifting” from your
older self. (Training already leans toward your recent work; your archive’s history stays available as
checkpoint voices.) Features tagged experimental are the newest on the board — fully working, still being proven out; they may get sharper (or quietly evolve) week to week.
Know your voice
Free Voice Check-up — Paste your writing — get a measured report: sentence rhythm, punctuation signature, the words that are yours. Plus a copy-paste prompt for ChatGPT/Claude.
Your trained voice — A private model trained on your actual writing — absorbed, not described. Runs on our hardware; your words never touch a third-party AI.
Voice X-ray — Every paragraph of any draft scored and color-coded against your voice. Calibrated on real authors — real Twain scores 75, prompted-AI Twain scores 45.
The overall read — One verdict up top — the overall health of the match — with content & style dials as the receipts underneath.
Editor diff — Paste the edited version of your piece and see exactly what an editor — or Grammarly — did to your voice.
Holds your facts
Trained not to invent — Voices that invent facts get a further training pass: the model is taught, on its own rewrites, to prefer the version that kept your specifics. Measured on writing it never trained on — fact retention up, inventions roughly halved, voice held.
Every voice gets a check-up — After training, each voice rewrites a fact-laden test set and every invention is counted. Clean voices (most are) need nothing; the ones that invent get the treatment automatically.
Strict facts — on by default — Rewrites are re-rolled when your numbers or names go missing, and the verdict line names what changed.
Fact repair — After a rewrite, your source's specifics are checked one by one — “12k” counts as “12,000” — and a repair pass restores what vanished and strips what appeared from nowhere, by name.
It admits failure — If every attempt drifted or dropped too much, the result says so in an amber banner instead of arriving confident. Honest limits: paraphrased distortions with no number or name to catch can still slip through — read before you publish.
Write with it
Rewrite / Write — Re-voice a passage, or draft something new from a brief — always in your voice.
Live voice bar — Spell-check, but for voice: every paragraph scored as you type.
Voice strength — How hard the voice pushes, from subtle to exaggerated. 1.0 is as trained; the dial scales the model itself at generation time, not the wording of a prompt.
Transcript or chips — Successive drafts either stack down the page like a chat, or sit behind v1 v2 v3 buttons that swap the text in place. Your choice, remembered; the toggle is on the right of the input bar once there are two drafts.
One notepad, everywhere — The drawer rides the right edge of every page — keep a result or a note once, find it on Write, X-ray, Chat, everywhere. Download it all as a ZIP.
Turbo — Better results when you have a minute — a second send button: the writer drafts several, keeps the best, then the Editor critiques it and the notes are applied. You get the original and the edited version, both.
Compare — Old and new side-by-side with word-level color-coded diff.
Does this contradict me? — One click checks a draft against everything you've written — and cites the piece it disagrees with, by name.
Keeps your specifics straight — On a rewrite, the names, numbers, links and dates you gave it are held to your source — drift is repaired automatically, and anything the model added, or quietly dropped, is flagged for you to check.
Say it my way — One click re-voices a drifting paragraph — and re-scores it to prove it.
History — Your last 60 generations per voice, timestamped, one click to copy.
Round-table — One question, up to six voices — yours, your past selves, the pre-sets — answering side by side. Members only.
Dialog — Two voices talk it out — pick the seats, set the topic, listen to the exchange, interject as moderator, then turn it into a draft with one click.
The bench — Judgment seats — Red Team, Editor, Bored Reader, Steelman. Critique any draft on the Write page, or seat one in a Dialog. Powered by our 14B; a writers' room that never flatters you.
Checkpoint voices — Every month your corpus is frozen automatically. Wake any month from the dashboard and a separate voice trains on who you were — then chat with them.
Whole-archive import + auto-sync — Connect your Substack and it reads your entire public archive — not just the recent posts a feed exposes — then absorbs every new post overnight. Publish Tuesday, your model knows it Wednesday. LinkedIn and Facebook import too: drop the official data-export .zip straight onto the upload box.
Save dialogs — Keep any two-voice exchange — reopen it, continue it, or draft from it later. Good conversations don't evaporate.
Microphone dictation — Click the mic and talk — your words type themselves into the box, transcribed on our own server like everything else.
Conversations — Threads you can search, rename, fork at any message, and return to.
Interview mode — It asks — one sharp question at a time — and Write-it-up turns your answers into a first-person piece.
Morning briefing — Headlines, weather, markets — every morning, in your voice, optionally in your inbox.
Daily question — A standing journal where your voice asks you one good question a day.
Upload a document — Drag a file from your desktop onto the chat — or paste. The conversation grounds itself in your piece.
Source citations — Grounded replies show which of your pieces they drew on — by name, under every answer. It can tell you where you wrote it.
Feed it
It accumulates you — Upload writing for training, pin facts to your notes mid-chat, save corrections, harvest your interview answers into training material — the voice grows every time you use it.
Put it to work
Tune your targeting — Who'd most want to read this piece? Ideal reader, segments, communities, and ready-to-post hooks — worked out privately.
Repurpose — One essay into a week of Notes, a launch kit, a thread — in your voice, with project memory.
Find the gaps in your beat — Your topics as a living map — and what kindred writers cover that you don't, each gap one click from a draft.
Crawl & Index — Scope the crawl to your voice: every result scored on content AND style against your corpus — sortable by either. Every completed search is saved on the page for 30 days.
Style Prism — One paragraph refracted through twelve classic voices at once.
Voice Mixer — A knob per voice — blend your voices with the classic authors into a brand-new one. An exact weight-space merge, ready in about a minute, no retraining; the blend shows up in every picker. Pro and up.
Pre-set voices — Newsroom free for everyone; The Orator, The Analyst, The Plain-Speaker, and The Storyteller with membership.
Weekly digest — Kindred reads and a question from your written self, every Monday.