A lot of people write with AI now. The problem is that most of it sounds like AI, em dashes and all.
That isn’t because the models can’t write. They can. It’s that writing something which genuinely sounds like you takes more than a clever prompt. It needs to understand what you care about, who you’re writing for, what is actually true about you, how you put sentences together, and how you react when a draft gets it wrong. None of that fits in a prompt box.
So I built Writers Room.
A studio, not a prompt
Writers Room is an open-source writing studio that runs inside Claude Code. It helps you develop rough ideas into researched, edited drafts that are worth publishing, rather than the plausible-sounding filler you get when you ask a chatbot to “write a post about X.”
The core of it is a local knowledge vault: the context a serious writing partner should already have before it types a word.
- who you are writing for
- the themes and ideas you keep returning to
- canonical facts about you and your work
- writing samples you have actually published
- an auditable specification of how you write
That vault is what lets the studio guess two different things at once: what you might say, and how you would say it. Get either one wrong and the draft reads like someone doing an impression of you.
The part I actually care about
Uploading context is table stakes. The part I care about happens after the draft.
When I edit the studio’s final output, Writers Room captures the difference between what it wrote and what I actually published. Those differences become evidence-backed rules it can apply next time. It doesn’t only learn from what I feed it up front. It learns from my pushback, which is the part of my judgment I could never have written down in advance.
That’s the bet the whole thing rests on. Voice isn’t a style you can describe once and lock in. It’s a running set of small corrections, and the useful system is the one that remembers them.
How it runs
The whole thing runs on local Markdown files, with a kanban board for moving ideas through the writing process. There is no hosted backend. Nothing publishes without my explicit approval. The project is MIT licensed, so you can read every line and run your own.
It’s still alpha. It has sharp edges. But it’s far enough along that I’m using it to develop and write my own work, including this piece.
Today I’m releasing it publicly: github.com/dhalarewich/writers-room.
If you use Claude and lean on AI to assist your writing, give it a try. I want to hear where it works, where it breaks, and what feels more complicated than it should.