Welcome to Typetoe — a calm place to write long things and keep track of them.
On an empty workspace you have two ways in:
If the book already exists somewhere else, you do not have to retype it. Use Import in the Library toolbar and drop in Markdown or plain text files — one becomes one document, and several at once become several, in the order you give them. From Scrivener, Ulysses or Word, export to Markdown first; that is the format every one of them can produce and the one that keeps your structure intact.
Headings inside a file stay headings. Nothing is reformatted behind your back.
You can collapse either side panel to give the editor more room; drag the edges to resize.
Everything autosaves as you type — there's no Save button. The small bar at the top shows when a change is being saved.
A project holds a tree of documents and folders. Create them from the Library toolbar, drag to reorder or nest, and rename inline. Folders group chapters, parts, notes — whatever structure fits your work.
Every document carries a status you choose — Draft, Canon, or Material — and a scan state it simply happens to be in. The first is a decision, the second an observation, and the tree shows both.
Each document in the tree carries a small round dot for its status:
A document whose excerpt begins with a small document icon has been scanned by Typetoe, and the text has not changed since. On the web that mark is the only scan state shown in the tree — a document never scanned, or rewritten since, shows nothing. In a project of fourteen chapters what you want at a glance is which ones are already done, and the desk is always on screen to tell you the rest.
On iPhone and iPad the row says more. The desk there is a tap away rather than beside you, so each document carries a small square for its scan state next to the status dot: teal scanned, amber out of date, pale never scanned. Same colours, same meanings as below.
The Editorial Desk names all four in words, under the status, each with a small square in the same colours:
"Out of date" is decided by comparing the text itself, not by dates. Opening a chapter, changing a comma and undoing it does not make the scan stale — only real changes to the prose do.
Write in a clean, formatted editor. Paste from anywhere — Typetoe keeps the structure. There's a markup view if you prefer to see the raw formatting, and a customiser for fonts, size, spacing, and paragraph style (in the editor's settings).
Footnotes: insert one from the right-click menu (Insert footnote). Once inserted, edit its text at the foot of the page in the Hybrid and Markup views, or in Formatted view right-click the footnote marker and choose Edit footnote.
Typetoe quietly keeps automatic versions as you make meaningful changes, so you can look back. You can also save a version by hand at a moment worth remembering.
Set a writing goal on a document or the whole project — a word target, optionally on a schedule — and track progress as you write.
You never lose work. That one is true on every plan and needs nothing switched on: when the line goes, your text is written to the device and goes up by itself the moment there is a connection again. Close the laptop, shut the app, land — it is still there.
While something is waiting you will see a small amber note saying so. It goes out on its own when everything has arrived; there is nothing to press and nothing to confirm.
Pro adds the rest of it. With no connection you can still begin a chapter, rename one, move it into a folder, throw one away — the whole shape of the project stays workable, and every change is sent in the order you made it when the line returns. On the phone, Pro also brings your chapters onto the device while you have a signal, so a project you have never opened offline is still readable at thirty thousand feet.
What does not work without a connection is anything that asks a model to think: the Scan, the chat, source research. Those are server calls and no amount of local storage changes it. Write on the train, scan when you land.
If a subscription lapses, nothing you have written goes anywhere: what is on the device stays readable and exportable. What stops is starting new work offline.
A Scan is Typetoe reading one of your documents, once, and keeping what it found.
Most AI tools re-read your text from scratch on every question. Typetoe reads it deliberately, when you ask, and stores the result — so the next question starts from something it already knows rather than from the chapter all over again.
Reading a document fills in three things:
Everything except the recap is kept. It is what the Editorial Desk shows, and it is what the assistant draws on when you ask about the book rather than about the paragraph in front of you.
After you have written or pasted something substantial, and again when a document has changed enough that what Typetoe knows about it has gone stale.
There is no need to scan as you type. A half-written chapter produces a reading of a half-written chapter.
Every document carries one of three marks, in the tree and on the Desk:
The mark is worked out by comparing the text against the text that was read, so it is exact: it is not a timestamp, and editing a comma will not make a document look stale.
Asking to re-read a document that has not changed will prompt you first. It would return the same reading and spend a scan to do it, so Typetoe asks whether you meant it rather than quietly obeying. Sometimes you do mean it — a second opinion is a reasonable thing to want.
Scans come out of an allowance, because each one is a real reading by a real model.
| per week | per year | |
|---|---|---|
| Free | 10 | 150 |
| Pro | 200 | 5,000 |
The year is not the week multiplied out: it is a separate ceiling, so an account that scans hard every week eventually meets it. In practice 150 covers writing a novel of thirty or forty chapters and revising every one of them three or four times.
The reading is not a report you read once. It becomes the project's memory: the synopsis, the people, the places. When you later ask "does this scene contradict anything?", that memory is what the question is answered against — which is also why an unscanned project gets vaguer answers.
See Editorial Desk for where the results live, and The Brief for the other half of what Typetoe knows: not what you have written, but where you meant to go.
Every project has one document called the Brief. It is the only place in Typetoe where you say what the book is meant to be.
That distinction is the whole point. Everything else Typetoe knows about your project is deduced from what you have already written — the synopses, the characters, the places, all of it produced by reading the manuscript. The Brief is the one thing no amount of reading can produce, because it is not in the text yet.
Write it for yourself, in whatever shape helps you think:
Notes are fine. Repetition is fine. Half-finished sentences are fine. It is working material, not a pitch — and specifically not a synopsis for a publisher, which is a summary of what exists. The Brief is the opposite: what you intend.
The Brief carries a status like any document.
Set it to canon when the direction has stopped moving. Leave it draft while you are still deciding; nothing is lost by doing so.
For rules the assistant must obey — tone, hard limits, canon facts stated as law — use the Charter instead. The Brief explains where you are going; the Charter constrains how Typetoe behaves on the way. See Editorial Desk.
The Brief is the first thing the assistant is given, on every question you ask. It is what stops "what should happen next?" from being answered by a machine that knows only the last paragraph.
Because it leads every question, Typetoe does not send it word for word. It reads your Brief once and keeps a condensed version — who exists, where, what the book is about — and rebuilds it when you change the Brief. Your own phrasing survives: distinctive wording is quoted rather than reworded, in the language you wrote it in, because a synonym would lose exactly what you meant.
You do not have to do anything for this. It happens after you save, and the first question you ask after a rewrite simply uses the Brief itself.
Because the honest translations were all wrong. In Italian, Spanish, French and German the natural word came out meaning summary — and a summary is what the Scan produces from what you have already written. Naming this document after that would promise the opposite of what it is for.
A name that promises nothing is better than one that promises the wrong thing. The empty page says what the name does not.
The right-hand desk is Typetoe's memory of your project. It fills up as you work.
Ask Typetoe to read a document once, from Read Document on the desk or the Scan button in the editor toolbar. It saves what it found — a synopsis, the characters and locations — so it does not have to rediscover them on every question.
The desk tells you where each document stands, on the line under its status: Scanned, Scan out of date, Never scanned, or nothing at all when the document is empty. See Writing for the full legend of the marks, and The Scan for what a reading produces and what it costs.
Continuity notes, decisions, and things to verify — kept per project. Typetoe can suggest checklist items while you work; you decide whether to keep them.
As documents are read, the people and places in them are collected here, each with a status you can confirm (to verify, canon, unused).
Rules, not direction: where the book is going belongs in the Brief (see The Brief), while the Charter governs how Typetoe behaves on the way.
The Charter is your set of rules for the project — tone, canon facts, dos and don'ts. Accepted Charter rules are treated as law: they're injected into every AI reply, so the assistant stays inside the world you've defined.
Save references you've checked. When the assistant verifies a real-world fact, it can offer to save the source here.
If you already pay for Claude or ChatGPT, you can point it at your Typetoe project and work there. It reads your chapters, your characters and locations, and the rules you set in the Charter; it can suggest research, flag things to check, revise a passage, or write one on your instructions.
There is no API key to set up and nothing extra to pay: the reading happens inside the subscription you already have, at its limits. And you never paste your manuscript into a chat and paste the answer back — which is the part that quietly ruins a novel, one lost revision at a time.
The alternative — your own provider key in AI settings — puts the assistant inside Typetoe instead, and you pay the provider per use. See AI and keys.
A Typetoe account with at least one project in it, and a browser. Claude's iPhone app has no screen for adding a connector from an address, which is what Typetoe is, so add it in Safari or Chrome instead.
The connector is added inside Claude, not in Typetoe. Typetoe's part is only the address:
https://app.typetoe.com/mcp
Open Claude in your browser, open the sidebar, and click your account at the foot of it. Go to Settings, then the Connectors tab, then Add, and in the menu that opens, Add custom connector.
Give it any name you like — Typetoe does — and paste the address exactly.
The settings underneath are filled in by Claude itself: it reads the server and works out how to connect. Leave them as you find them, scroll to the bottom, and press Add. If your screen shows something different from what you expected, leave it as Claude set it and press Add anyway.
Adding it is not the same as connecting it. The connector now exists in your list, unlinked: click Connect, sign in to Typetoe, and authorise the access. That screen lists exactly what Claude will be able to reach.
Open a new chat and ask for something that needs your project. What projects do I have on Typetoe? is the one to start with — Claude will ask permission, once or for good, one tool at a time.
It connects to your account, not to one project, so it is worth naming the project you mean inside the conversation.
It works on the text and the desk. It cannot move documents, rename them, empty the Trash, or change your plan — the shape of the project stays yours.
Your plan applies here exactly as it does in the app: the same project limit, the same storage limit. An assistant cannot be used to get around them.
Both from the same list in Claude. Disconnect breaks the link and leaves the connector where it is, ready to reconnect. Remove, in the three-dot menu, takes it away entirely.
The same address works. ChatGPT keeps custom connectors in its developer settings, and asks you for a name, a description and a logo, which Claude reads from Typetoe instead.
Almost every failure is the same one: it was added while signed out of Typetoe, or to the wrong account. Sign in at app.typetoe.com first, then add the connector — and if it is already there and misbehaving, remove it and add it again rather than retrying.
There are four different things here, and they cost different amounts. Worth two minutes.
The chat you meet without setting anything up is Typetoe's guide, and Typetoe pays for it. Ask it how the Scan works, what the Charter is for, how to import a manuscript, how to connect Claude — it answers those. It comes with 200,000 tokens for the lifetime of an account, which do not expire — they get used up. Enough to get a feel for the assistant and the Scan, deliberately not enough to write a book with.
It is not an editorial assistant, and it will tell you so rather than pretend. For that, read on.
Free on every plan, no key needed. Ask Typetoe to read a document once and it keeps the essentials — what kind of text it is, a short synopsis, who and where it mentions — so nothing has to be rediscovered later. Re-reading text that has not changed is free and does not count.
There is a weekly allowance, larger on Pro, and a ceiling on the year. The week rolls rather than resetting on a fixed day: your allowance comes back one scan at a time as the oldest ages past seven days. Bring your own key and the allowance disappears entirely — at that point it is your money being spent, not ours.
A scan is quick and it collects facts. An advanced reading is the other thing entirely: a stronger model reads a whole chapter and tells you what works, what does not, and what the pages are promising the reader. Typetoe pays for it, on its own account — your key is neither used nor needed here.
Two come with a new account. Pro adds two a month, and those are an allowance: they stop accruing at a ceiling, and they lapse if the subscription ends.
When the plan's are gone, subscribers can buy more, in packs of five or ten. Bought readings behave differently from the included ones in the way that matters: they never expire, they survive cancelling, and they are spent last — the plan's own go first, so topping up early never wastes what you already had.
The counter sits on the Editorial Desk beside the reading, in Subscription, and in the message you get when there are none left. Top up is in all three places.
Two ways, and the first costs nothing extra.
Connect Claude or ChatGPT. If you already pay for either, point it at your project and work there — no key, no second subscription. See Claude and ChatGPT for how.
Or bring your own provider key, added once in AI settings, which puts the full assistant inside Typetoe itself. You pay the provider for what you use, top it up with whatever suits you, and Typetoe shows you what each request cost.
You can point Typetoe at OpenAI, Anthropic, Google Gemini, DeepSeek, Groq, xAI or Moonshot and choose a model; Google Gemini has a free tier, so this can start at no cost. For searching the web from inside Typetoe there are Tavily, Brave, Exa and Perplexity, added separately in AI settings and needing a key of their own.
There is no rule about which to choose, and any rule would go stale: at the time of writing Anthropic's Claude reads subtext best, ChatGPT is the one long brainstormers tend to prefer, and Gemini is the most current on anything recent. Adding two or three keys and switching between them costs nothing until you use them.
On the device you typed them into — the browser's own storage on the web, the system keychain on iPhone. Typetoe never writes them to its database, so there is no store of keys here to break into. They are sent with each request, because the request is made from our server on your behalf; they are not kept afterwards.
The side effect is the one you would expect: a key typed on the laptop is not on the phone, and you paste it again on each device you write from.
Every provider's console works much alike. Each asks you to make a project or workspace to hold the key, and offers it an expiry date if you want one.
Copy the key the moment it appears. Most providers show it once and never again. If you lose it, nothing is broken: disable that one, delete it, and make another.
Anthropic may offer you identity federation before it offers you a key. That is for servers and CI pipelines, which have an identity of their own to borrow; a key pasted into a browser has not. Choose Continue with an API key.
The free plan keeps 2 active projects, has a generous storage limit, and includes the Scan and the guide. The Trash is Pro: on the free plan your projects stay put, though you can always rename them or clear their contents by hand.
Pro lifts the project limit, unlocks the Trash, raises the Scan allowance, adds two advanced readings a month, and adds working offline (see Writing). It is available now, monthly or yearly — the current prices are on typetoe.com and on the Pro card inside the app.