FZ-022 · WRITING
Ghostwritin'
A rewrite tool for writers who used AI for a first draft.
- Status
- In development
- Pipeline stage
- VALIDATION
- Sector
- WRITING
- Autonomy target
- LEVEL 4 · SELF-OPTIMIZING
- Built with
- 3 live · 0 planned
The problem
An AI tool helped with your draft, and now it does not sound like you, and readers think it is AI.
What it does
Ghostwritin’ is designed to rewrite AI-assisted drafts in your own voice and check the meaning stayed the same.
How it works
- Choose a voice (Casual, Professional, Academic, or My voice, learned from three to five of your samples) and a strength
- Names, numbers, quotes and code stay locked, and every rewrite is checked against your original
- A diff shows every change, with a human score before and after; the web app, API, CLI and MCP server share one engine
Planned to be open source (MIT): self-host with your own model key, or use the hosted version, with a free tier of 500 words a day.
Stop rewriting a tool’s draft line by line until it sounds like you again.
TodayThe open-source engine is public (MIT); a staging API is deployed; nothing hosted or published yet.
Aims
- Autonomous operation
- The rewriting is the work meant to run on its own: drafts come in through the web app, API, CLI or MCP server, the engine rewrites them in the chosen voice, checks the result against the original and returns the diff and scores. People handle what a machine should not decide: support, abuse reports and the use policy.
- Intelligence
- A language model does the rewrite; plain code does the checking. The meaning lock compares names, numbers, quotes and code against the original, and a rewrite that moved one is rejected. The human score is a detector estimate shown as a guide, never a promise.
- Growth
- Open source first: the engine, CLI and MCP server are planned under MIT, so writers and developers can self-host with their own model key. Growth is people moving to hosted plans for My voice and the score, counted from the first rewrite that is not ours.
The technical detail
Rewrites AI-assisted drafts into the writer’s own voice and checks that the meaning stayed the same. A draft of up to 10,000 words is pasted in and the passages a detector would flag are marked; the writer picks a voice (Casual, Professional, Academic, or My voice built from three to five of their own samples) and a strength from light polish to a full rewrite. A meaning lock keeps names, numbers, quotes and code exactly as written and checks each rewrite against the original; a diff shows every change, with a human score (a detector estimate, a guide not a guarantee) before and after. Web app, REST API, CLI and MCP server are planned on one MIT-licensed engine, self-hosted or hosted. Intended for writers whose own work gets misread and for drafts a tool helped with; users are told to follow their school’s, client’s or publisher’s rules. What exists today is the homepage, the waitlist, the public MIT engine and a staging API; nothing is hosted or published yet.
