User Guide — Cothyra

Cothyra is a workspace for writing and reading books with an AI assistant. Conversations happen next to a real document, and what you keep at the end is the book — not a pile of chat transcripts. This guide covers the essentials, roughly in the order you'll meet them.

1. Getting started

Open app.cothyra.com and create an account — with an email and password, with a six-digit code we email you, or with Google or GitHub.
New accounts receive a one-time grant of 500 free credits, and a Welcome Guide book appears in your library to walk you through the main features hands-on.
Cothyra runs in any modern browser. On a desktop you get a three-pane layout — contents, chapter, and chat side by side. On a phone or tablet the same panels sit behind a tab bar at the bottom of the screen.

2. Books, chapters, and notes

The library is your home screen. Create a book, and it gets a table of contents, chapters, and a free-form notes area for material that isn't part of the manuscript — research dumps, outlines, verification reports.
Book titles, file names, and links work in any language and script.
Everything saves continuously and syncs across your devices; your work is also cached in the browser, so reading and light editing keep working through a flaky connection. Edits made on two devices at once are merged automatically — if two changes genuinely collide, a banner asks you which side to keep. Nothing is silently discarded.

3. Writing: the editor

Chapters are Markdown underneath, but you rarely look at raw syntax: the document is always fully rendered, and the block your caret enters opens for editing in place — no split-screen preview to keep in sync. On top of that, the editor goes well beyond standard Markdown:

Block tools: type / at the start of a line for an insert menu (headings, tables, code, math, references, and more), drag any block to rearrange it, and pop open the outline rail at the right edge to jump between headings.
Select some text and a floating toolbar offers formatting — plus AI actions such as "Explain", which hands the selection to the assistant.
Math is first-class: KaTeX inline $…$ and display $$…$$, multi-line equations, and extra macros for analysis notation. Math also works inside tables, footnotes, and diagram labels.
Tables are edited in place: click a cell to type, drag the row and column handles to reorder, and set alignment from the handle menu.
Scholarly apparatus built in: Pandoc-style citations ([@key]) render in author–year, bracket, or numeric style from your book's reference list, and footnotes show their text in a hover tooltip.
Figures: paste or drop images, drag to resize, click to zoom. Pen sketches can be drawn straight onto the page with a stylus — pressure-sensitive, palm-rejecting — and are stored as crisp, printable vector graphics.
Editable vector diagrams: draw boxes-and-arrows figures on a diagram canvas and reopen them later to keep editing — nothing is ever flattened to pixels, and labels can contain math.
Live figures: mermaid fences render diagrams from text, and JavaScript render fences draw charts with bundled plotting libraries — entirely offline, no external services involved.
Code blocks with syntax highlighting and one-click copy, task lists, and emoji shortcodes like :smile:.
Every change — yours and the assistant's — lands in a single shared edit history per chapter. Undo steps back through both, and the history survives reloads, so an AI edit is never something you just have to live with.
Long manuscripts stay fast: rendering is virtualized, so even very large documents scroll and search smoothly, and your reading position is remembered per document, down to the line — on any of your devices.

4. The AI assistant

The chat panel sits beside the chapter and the assistant reads what you're reading, so "tighten the second section" or "is this derivation right?" needs no copy-pasting.
The assistant can edit the chapter directly. Its edits are ordinary entries in the shared history — review them, build on them, or undo them like your own.
Pick your model per conversation: models from Anthropic, OpenAI, Google, xAI, DeepSeek, and others are available under one account, at one price list, with no separate subscriptions. Reasoning models show their thinking as they work.
The assistant can research on the web — search, fetch pages, and query scholarly sources such as arXiv, OpenAlex, Wikipedia, and Crossref — and cite what it found into your references.
Actions with consequences (overwriting a file, running a tool you've marked as sensitive) ask for your approval before running. You can branch a conversation from any earlier message to explore an alternative without losing the original thread.

5. Voice reading

The headphones button reads the current chapter aloud — including math, spoken as words rather than skipped.
Three narration styles: Verbatim (the text as written), Faithful (lightly smoothed for the ear), and Summary (the short version). Several voices and engines are available, including a free voice; playback speed is adjustable.
Playback continues while you browse other chapters or tabs, and you can ask a question by voice mid-listen — the assistant's answer is read back to you in the flow.
Listen responsibly: keep your attention on the road, the trail, or the machinery first. See the safe-listening note in our Terms.

6. Export, print, and sharing

Export a chapter or a whole book to PDF, Word (.docx), or LaTeX, or print directly. Diagrams and math carry over.
Your writing is portable by construction: chapters are plain Markdown files, and your full account data is downloadable as a zip at any time (see section 8).
Share a book as a read-only public page with a link — no account needed to read it; unshare whenever you like.
Connect GitHub (Settings → Connections) and the assistant can push your book's source files to a repository you choose. Connections are scoped and revocable.

7. Credits and billing

Cothyra charges for usage, not seats: 1 credit = $0.01, spent per AI request at the per-model rates listed in Settings → Billing → Pricing.
Your 500 free signup credits work on every model except a few frontier models marked with a lock — those require purchased credits. Purchased credits never expire.
Buy credit packs by card via "Add credits" — checkout is handled by Paddle, our merchant of record; prices are shown in USD and tax is added at checkout where applicable.
The Usage page shows exactly where credits went: spend, token counts, and cache-hit rates per day, per model, and per book.
Unused purchased credits are refundable within 1 year — request a refund from Billing → Payments, or see the Refund Policy linked below.

8. Your data and account

Download everything at any time: account menu → "Download my data" gives you a zip of your profile, billing history, feedback, and every file in your library.
Delete your account in-app (account menu → "Delete account"); the Privacy Policy describes exactly what is removed and what accounting law requires us to keep.
When you use the assistant, your prompt and the chapter you're working on are sent to the provider of the model you selected. Don't put credentials, card numbers, medical records, or other people's sensitive personal data in your documents — see the Privacy Policy and the sensitive-data section of the Terms.

9. Getting help

Found a bug, or something confusing? The feedback form in the account menu goes straight to us, and you can always email [email protected].