← All work
MARKETPLACES & CONSUMERIN DEVELOPMENTQuireA writing studio you talk to
Marketplaces & consumerIn development

Quire

A chat-first writing studio for novelists: you talk to it, and it drafts, remembers and checks your book.

25
tools the assistant can use on a book
4
effort levels, each with its own models and context budget
10
slash commands for the common writing jobs

The problem

Writing a long novel with a general chat assistant falls apart. It forgets the characters, contradicts earlier chapters and drifts from the author's voice, so the author keeps pasting in context and hunting for continuity errors by hand.

What it does

Quire is a chat window where the book itself is the thing being worked on. The assistant can read and edit the actual chapters, characters and timeline, not a pasted excerpt, and slash commands cover the common jobs: continue, rewrite, outline, draft, summarise, check and narrate.

In the background it keeps a bible of characters, places and plot threads up to date. After each chapter is saved, it checks the chapter against that bible and lists any contradictions, quoting the line. Highlighting a passage lets the author shorten, expand or rewrite it and review a word-level comparison before accepting.

Around the writing are a chapter read aloud by a self-hosted voice, generated character portraits and covers, import from Markdown, text or Word files, export, and an optional public reading page for each book. Each chat has an effort level that sets which models are used and how much of the manuscript goes with every request.

Key features

  • Chat that works on the book

    The assistant has 25 tools for reading, editing and creating chapters, bible entries and timeline items, so nothing has to be copied in and out.

  • A bible that keeps itself

    Characters, places and plot threads are pulled from the chapters automatically, so the assistant stays consistent with what's already written.

  • Automatic continuity checks

    Every saved chapter is checked against the bible, and each contradiction comes with the quoted line and a suggested fix.

  • Highlight to rewrite

    Select a passage, review a word-level comparison, then accept or reject, and every chapter change keeps a version to restore.

  • One drafting pipeline

    Chapter-length writing runs through plan, draft, check, critique and revise, and it aims at the length of the author's recent chapters instead of drifting.

  • Covers spelled right

    Cover art is generated without lettering, and the title and author are set by the app in real fonts, so a name is never misspelled.

Where it stands

A private, invite-only studio in active development, built for a single author. There's no public sign-up or pricing yet, and the draft-quality evaluation so far covers only two books.

Under the hood: for the technically-minded

How it's built

Quire started from the studio's own document-publishing app, keeping its sign-in, storage, job queue and deploy setup. It is one server-rendered React Router app with a single SQLite database, running in one container on the studio's own server.

Every AI call goes through one client and a single table that maps each task and effort level to a model, a reasoning budget and fallbacks. That table was chosen from benchmarks run on a real manuscript, not from model rankings. Chat turns run on the server and keep going if the browser drops, and the page re-attaches when it comes back.

The design is deliberately quiet: near-monochrome graphite with one reserved indigo accent that appears only while the assistant is working. Each product decision is logged with the reason behind it.

The hard problems

  • Keeping long drafts consistent

    Drafts drifted in length and voice. All writing now goes through one entry point that feeds the manuscript in within a token budget, sets the length target from the author's recent chapters, applies second-pass edits exactly in code, and runs a model-free scan for machine-sounding prose.

  • Measuring quality honestly

    Judged quality turned out to track draft length, so a longer draft looked better whether it was or not. The evaluation now compares drafts at equal length.

  • Unreliable cover text

    Image models misspell names. The fix generates artwork with no text and typesets the title and author with the app's own fonts.

Built with

  • React Router 8
  • TypeScript
  • SQLite
  • OpenRouter
  • Tiptap

Building something in this space?

Work with us →