Privacy
Last updated 2026-08-31 · software v0.2.1
EnglishDictationMaker is local-first desktop software. Your word lists, templates, history and settings live on your own computer. On launch it makes one reachability check against our server to decide whether the online features are available. Beyond that it uploads none of your content - word lists, templates and history never leave. Everything else online happens because you asked for it.
What stays on your computer
- The word library: the built-in entries, and anything you type in or import
- Templates and worksheet settings
- Generated history and the current draft
- Your quiz record: how many times you got each word right or wrong, and when — the mistake book is built from it. The questions and the answers you typed are not kept; remove words one by one, or clear it, on the Learn page. When the AI helper is set to tier B or C, the words you missed most recently go with your questions - see the AI helper section below
- AI assistant conversations (the 50 most recent)
- AI character cards you imported: names, profiles and pictures
- Custom layouts you saved: a name and the layout (columns, row heights, the fill-in fields in the header and so on). When one is made from a Word worksheet, only the layout is kept - not the words or names on it
- Interface preferences
- Records of redeemed books
- Trial usage and licence credentials
All of it sits in the system application-data folder and is never uploaded on its own. Deleting that folder, or uninstalling, removes all of it.
When it goes online
| Action | When it happens | What is sent |
|---|---|---|
| Opening the app | Every launch, when the device is online | One /api/health reachability check; if it succeeds, one public-IP lookup used to pick the interface language. None of your content |
| Reading announcements | When you open the announcements page | Fetch only; none of your content |
| Checking templates and updates | When you press that button | Fetch only; none of your content |
| Syncing the online library | When you ask for it | Fetch only; none of your content |
| Contributing to the shared library | Only if you tick it in settings, and only when you start it | Entries you typed in yourself, up to 100 at a time |
| Trial and activation | When you activate, or when the licence is checked | A device identifier and the licence code |
| The AI helper | When you send a message in it | See the section below |
Everything above except the last row goes to api.ychmmmt.com.
Contributing to the shared library
Off by default. Once you tick it, the only thing uploaded is entries you typed in or imported yourself. Built-in words and the contents of redeemed books are never uploaded - that material is licensed.
Uploaded entries become part of a public shared library, so do not put personal information in an entry or a tag.
The AI helper (bring your own key)
This is the only feature that sends your content to a third party. Read this section before deciding to use it.
The AI helper is not a service we run. You supply your own provider address, model name and API key, and the software sends your messages straight to the provider you chose. Our servers are not involved and never see any of it.
You can list up to five providers in the order you want, and each question goes to the first one on the list that is set up. A separate switch, "when a provider fails, try the next one", is off by default. With it on, if a provider fails before it starts answering (rate limit, no balance, unreachable and so on), the same conversation, with whatever your setting below attaches, goes to the next provider on your list, and each provider is tried at most once. So with it on, a question can reach more than one provider, and the one that failed may already have received it. These are only the providers you listed yourself; the software has no fallback provider of its own. The line under a reply saying who answered is shown only on your computer; it is not kept in the conversation history or sent anywhere.
How much goes out is your choice - each model is added with one of two settings:
- Only what you type (the default): just the words you enter in the chat box - plus the profile of a character you imported, if you chose one (see below).
- Let the AI see everything: the word list from the current worksheet is attached as well, together with the words you missed most recently in your mistake book. At most 20 of those go, each as its spelling and how many times you got it right and wrong, with no times and none of the answers you typed.
Beyond those two there is a third setting, C - everything, off by default. The two features below appear only when it is chosen, and choosing it sends nothing by itself: each has three choices - Always, Only when I click and Never - and both start at Never.
One is "where you are". It goes to the provider you chose, so the helper can say things like "it is evening where you are". With Always, the software reads this computer's time zone (for example Asia/Shanghai) each time you ask and sends it with the question; with Only when I click, only what you typed, or read once with Use my system timezone, is sent; with Never, neither your location nor your time zone is sent.
That field is city or timezone level only. The software reads no precise coordinates, asks for no system location permission, and gives you no reason to run it as an administrator. You can clear it at any time in settings.
With tier C selected, questions also carry your date, local time and weekday, so the helper can say things like "it is late"; the time zone goes with them only when "where you are" is set to Always. Tiers A and B send no time information at all.
Tier C attaches the word list and the mistake book words just as tier B does. Its questions also carry the current worksheet's layout settings (font size, columns, layout and so on - never the title or header note) and your defaults for new worksheets, so the helper can propose changes to the worksheet, save settings presets and change your defaults. Each one is shown to you as before and after, takes effect only when you click Apply, and can be undone. It cannot change the model setup, your key, privacy settings or licensing, and it cannot delete your data.
There is also a "now playing" feature. You type a title, artist and album, and the helper talks about the song and marks the textbook words it used. Only those three fields are sent, and they go to the same provider you configured yourself.
The textbook marking happens on your own computer; your word library is not sent. That is why every word marked comes from the textbook you chose rather than from the model's guess. By default you type those three fields yourself. The other tier C choice is "Reading what is playing from the media player" (Windows only for now): with Always, the software asks Windows for the title, artist and album of what is playing as often as you choose and puts them in the three fields - When I open the AI assistant reads once each time you open the AI assistant, and Every N minutes reads every N minutes while the AI assistant is open and its window is in front; with Only when I click, it reads once when you press Read from player; with Never, the player is not read. Whatever is read is sent only after you check it and press Tell me about it. Nothing is read while the window is in the background, and nothing you listen to is recorded.
You can import character cards into the AI assistant so it helps you in a different voice. Cards are stored only on your computer. Once you choose a character, every question also carries its condensed profile (about 1,200 characters) to the provider you picked, at every level - it is content you chose and imported yourself. The picture stays on your computer and is never sent. Anything in a card that asks the AI to hide that it is an AI, ignore its rules, or act as your romantic partner has no effect: the software's own rules come before the character, and a card cannot change them.
There is also Generate a layout from a photo: take a picture of a dictation sheet, pick it and press Send, and the app sends that photo to the provider you chose, asking it to describe the layout. It works at every level, because you pick the photo and press Send yourself. Before it goes out, the photo is processed on your computer: turned upright, re-encoded (which removes the location, camera model and time the photo was taken), cropped to the paper and shrunk. What is sent is the processed copy, never the original. The photo is used once and is not saved or written to the log; the layout is kept only if you press Save. Dictation sheets often carry a student's name and class, so take a look before you send. It does not go through our server; it goes only to the provider you chose.
Each provider's API key is held as its own entry in the operating system credential store (Windows Credential Manager, macOS Keychain) rather than in a settings file, and removing a provider removes its key. The interface side of the software cannot read a key back; it can only learn whether one has been set. Requests are issued by the local program, and the provider address is kept on that side too - so even a tampered interface has no way to send your key somewhere else.
What the provider then does with what you send is governed by their privacy policy, not by us.
Feedback
The report the feedback form builds is copied to your clipboard or handed to your mail client. It does not pass through our servers. You can read the whole thing before it goes anywhere, and the diagnostic part is already redacted.
Your IP address is not fetched automatically - only when you press that button.
Logs
Logs stay on your machine, for your own troubleshooting. They never record an API key.
About this website
ychmmmt.com sets no cookies and loads no third-party analytics.
The visit count above the footer comes from the server access log, which the web server writes anyway. The page reads one static file and sends nothing anywhere in order to be counted. The figure counts page requests only, with crawlers and asset requests excluded.
Contact
If you have a question, or find that this page does not match what the software actually does, write to ych120911@163.com.