User Manual

1. Create an account

Enter your email address on the registration screen and we email you a URL to set your password (if you were invited, the invitation URL works the same way). Open it, set a password, tick the box to agree to the Terms of Service and the Privacy Policy, and the account is active. Every feature is yours to try for 14 days from registration. No credit card is required. For the plans after the trial, see 23. Plan.

• The emailed URL and invitation URLs expire after 7 days. If yours has, apply again from the registration screen, or contact us.
• Passwords must be at least 10 characters.
• If you forget it, "Forgot your password?" on the sign-in screen emails you a reset URL. The link works once and expires in two hours.
• Five failed sign-ins in a row lock the account temporarily. Wait, or contact support.

After your first sign-in there are two short setup screens — which mode to work in (see 5.), then your display name and language pair (JA→EN / EN→JA). All of them can be changed later in Settings.

2. How TerpPrep thinks — assignments and your glossary library

When you create an assignment you pick a client, and that client's own glossaries plus the shared ones it has been given are merged for reference (which shared ones is decided once, on the client's screen). Terms extracted from the material that already exist in your library are marked ✓ and your earlier translation is shown first. The more you use it, the less the next job costs you — that is the mechanism at the centre of TerpPrep.

3. Your first assignment

1. Press "New assignment" on the dashboard.
2. Enter a name, the date, and the mode (simultaneous / consecutive).
3. Choose a client / product (for a new client, type the name and it is created on the spot).
4. There is no glossary choice here any more: which glossaries the work reads is set once on the client's screen. An assignment with no client reads every shared glossary.
5. "Create" opens the assignment's own page (since 2026-08-20). You can book the assignment now and put the material in when it arrives; when you have it, "+ Add input" opens the input screen, where you put your material in and press "Next" (see 6.).
6. On the output screen, tick what you want (glossary, draft translation, document summary, background brief) and press "Run". Everything you ticked is produced in one background job (a few minutes). You do not have to wait — you can go back to the assignment page.
7. Check and correct the glossary (see 8.), then export to Excel/CSV or generate the live sheet (see 12.). More material can be added at any time from "Add input" on the assignment page.

4. Where things are

5. The three modes: Omakase, Custom and Flex

You can choose one of three ways from putting material in to running it (Settings → "Flow mode", or click the mode pill in the header). You can switch at any time, and switching does not touch the assignments and glossaries you have already made.

This manual describes the Custom screens. In Omakase mode many of the options below simply are not shown. In Flex mode the same options sit inside each output's panel (see "Flex mode" below).

Flex mode

In Flex mode, "+ Add input" on the assignment page opens the "Inputs and outputs" page. Inputs are on the left, outputs on the right; you connect them with arrows and press "Run".

1. Add inputs — under "Inputs" on the left, add a file, pasted text, a web page (URL), a captured photo or an assignment memo. Compression and anonymisation work just as on Custom mode's input screen (see 6.).
2. Connect them — drag the ● at the right edge of an input onto the output you want (Glossary, Document summary, Draft translation (layout kept), Draft translation (bilingual), Background brief). If dragging is awkward, the buttons on each input connect it too. One input can feed several outputs, and several inputs can feed one. Click an arrow to select it, then "Remove this arrow" to take it off.
3. Set each output — the number of terms, the brief's mode and "How to read the connected inputs" (as material / as an agenda), the draft translation's direction and format, and so on, are chosen inside each output's panel.
4. Run — the page does not move. You can wait on it and open each result with "Open" as it finishes. An output with no arrow is not run.

6. Putting material in

Files chosen in several goes add up. Remove one with the ✕ at the right of the list ("Clear file selection" starts over).

When the file is too large — shrink it first

Choosing "📁 Files" shows a "Shrink before uploading" toggle (on by default). Press "📉 Shrink" and, inside your browser only, embedded videos are removed from PowerPoint files and images are scaled down. Nothing is sent anywhere to do this.

• The current total and the limit are shown. If you are over it, pressing "Next" does not send anything — it asks you to shrink first, rather than failing after a long upload.
• Three strengths — Standard (1280px long edge, recommended) / Stronger (960px) / Video only (images untouched).
• "⬇ Save the shrunk file" downloads it, which is also a way to make a deck small enough to email.
• Your original file is never modified — only the copy that is uploaded.

Does losing image quality matter? TerpPrep only reads text from your material. Videos and images are used by none of the glossary, brief, draft translation or summary — so removing videos and scaling images down does not change what is generated.

This works on .pptx / .docx / .xlsx (PDF and text are passed through untouched). Video removal is PowerPoint only. Files that do not get smaller are used as they were.

If you would rather not upload at all

There are two options — anonymisation (redact, then send) and No-AI mode (never send). See 21..

7. No material yet — start from the agenda

When the deck will not arrive until the night before, pasting the agenda is enough to start preparing.

1. On the input screen choose Text and paste the agenda (English or Japanese). If it arrived as a Word file, upload the file instead — that works too.
2. On the output screen choose the Agenda course (pasted text selects it for you). Tick Background brief and Glossary: the AI writes the background to the agenda, and the glossary is built from that brief (the arrows on screen show this order).
3. The brief panel also offers "Supplement with a web search" (opt-in, off by default), for recent information the model does not have.

Writing your own prompt

Choosing "Custom prompt" as the brief mode reveals a text box. It opens empty — what you write there is the instruction only, and the material you uploaded travels with it as what the instruction applies to. "Draft likely questions and answers from this deck", "organise the arguments by speaker", and so on.

• It works on the document course as well as the agenda course. Only the brief reads your instruction; the glossary, summary and draft translation are still built from the material itself.
• A line reading "＝＝ここからはプロンプト" in a pasted text makes everything after it the prompt, whatever mode is selected (the output screen says so when it happens).
• The other-language version can be produced too (see 9.): the reply comes back in whatever language the prompt asked for, and the opposite side is then written.

8. Finishing the glossary

The assignment's glossary screen puts every term from every referenced glossary into one table. What is generated is a first draft. What makes it valuable to the next job is the version a professional has corrected.

Where a term belongs, and "in library"

The belongs to column carries a badge for the origin: client shared glossary this assignment only / new. A tick in the 📚 column means the term is already in your library.

Narrow columns use icons as headers — 📚 in library, # occurrences in the material, ★ importance, 🔗 group, 🗑 delete. Hover a header for its full name.

The Source column names which document a term came from. Blank means the term was not found in the material but suggested by the AI.

Display and sorting

• Click a header to sort. Each click cycles original order → ascending → descending → original order — you never have to reload to get back.
• "Columns" hides the ones you do not use, and columns can be dragged into a different order (remembered in this browser). "Reset order" restores the default.
• The same menu's "＋ Add a column…" creates a column of your own (up to five) — a pronunciation note, a client's preferred rendering, a pre-assignment checklist: the things one Notes column ends up cramming together. Columns belong to your account, so the same columns appear on assignments and in the library alike. The ✕ beside a column's name removes it (values you typed are kept — the column is only hidden, and recreating one with the same name does not bring them back).
• Column widths follow the content. Long descriptions and notes wrap.
• Hovering a term shows it in full, and where the term has a reading (furigana), the reading is shown underneath it. Readings are added from the glossary editor and corrected on the term's own page.
• The filters above the table narrow by where a term belongs.

Editing

• Terms, descriptions, Notes and your own columns are all edited right in the table (the ✎ in the header marks them). What you type is saved the moment you type it. Belongs to and importance are dropdowns.
• Correcting a library-registered term here reaches the library copy and every assignment using that term, with a revision recorded. The ↗ beside the term opens the term page (revision history lives there).
• Notes is yours to write in. What you type is saved the moment you type it, and for a term already in the library it reaches the library copy too. Your own columns behave the same way. Exports (Excel/CSV) include your columns at the end (re-importing that file does not bring their values back in).
• The whole table is reachable from the keyboard, in the two modes a spreadsheet uses. The arrow keys move a selection outline from cell to cell — nothing is being typed yet. Enter starts editing that cell, where the arrows move within the text. Enter again commits and hands the outline back (that is when it saves); Esc discards the edit and returns to the outline. Tab/Shift+Tab go right/left, wrapping to the next row at the end. Clicking a cell with the mouse still starts editing straight away.
• The editing box keeps the cell's shape: text that wrapped over several lines does not collapse into one the moment you start editing it. Use Shift+Enter for a line break inside a cell (Enter on its own commits). With a Japanese IME, the Enter and Esc that confirm or abandon a conversion do not end the edit — the next Enter, after the conversion is settled, is the one that commits.
• Changing where a term belongs registers it in the library then and there. Deciding where it belongs and saving it are the same action — there is no separate button. The row turns blue for the moment it takes. If the blue does not clear, the request failed: reload and choose again.
• Terms set to importance High are the ones printed on the live sheet (with the default range setting).

Working on several rows at once

Select rows with the checkboxes at the left (Shift+click for a range). The toolbar above the table then offers:

• The "belongs to" dropdown — choosing a destination performs the move and the library registration at once (there is no confirm button). Glossaries the AI merely suggested appear as "(new shared glossary)" near the bottom and are created when chosen. The very last entry, "＋ New shared glossary…", lets you type any name.
• 🔗 Group — binds the selected terms together. A group does not come apart when the table is sorted, and it registers and prints as one block. Use it for pairs you want to read side by side, or a full name and its abbreviation. "Ungroup" undoes it.
• 🖍 Highlighter (yellow, blue, red) — marks terms, and the marks print on the live sheet. The meaning of each colour is yours to decide. "Clear highlight" removes it.
• "Delete" (top right) — deletes the selected rows.

9. Background brief and document summary

These two are a pair: they cover the job from the outside and from the inside.

The brief ends with "Points for the interpreter" — not general background but the surprising fact attached to a proper noun (an ownership relationship, a recent change, something particular to that company).

Specific facts should be verified before you work. There is a cap on how many briefs one assignment keeps (viewing and editing are always available).

Saving and printing

Next to the language switch are buttons to print and to save as PDF, DOCX or text (the same row on both screens). Each saves the version currently displayed — main language, other language, or bilingual. The PDF is a laid-out A4 with proper headings and lists, which reads better on paper in the booth.

The other language, and bilingual view

Both can be produced in the assignment's other language as well. Tick "Also produce the … version" in the panel on the output screen. Afterwards the switch at the top moves freely between main language / other language / bilingual (bilingual places the paragraphs side by side).

• Bilingual is a way of displaying, not a separate generation — deciding later that you want it costs nothing. Tick "Show bilingual by default" to open that way.
• Only the main-language version is editable. Once you edit it the other language still holds the pre-edit text, and the screen says so.
• The other-language version has the same deletion date as the original.
• Custom-prompt briefs can do this too. The reply arrives in the language the prompt asked for, and the opposite side is written from it. The language the prompt chose is never overwritten.

10. Draft translation

A full draft translation of the material, using your glossary's renderings consistently. Two output formats:

• The direction (en→ja / ja→en) is yours to choose. It can also be run on its own from the assignment page.
• PDF and text sources cannot produce a translated file, so they always come out as a bilingual script.
• You can also add translations to glossary terms only. Choose "Glossary terms only" under "What to translate" for the translated file: the text stays as it is, and each glossary term gets its translation in brackets after it (e.g. this is a pen （ペン）). PPTX / DOCX / Excel only. No AI is used — the glossary's own translations are written in. If the material contains none of the glossary's terms, nothing is made.
• In translated files, layout can shift where the translation runs longer than the source. In Excel only string cells are translated; formulas and numbers are untouched.
• Draft translations are deleted 7 days after they are generated. Download what you need to keep. "Delete now" at the top of the screen removes them without waiting out the seven days.
• You are told before you reach a limit (2026-10-29). On the Free plan the assignment screen shows both "N left on this assignment · N left this month", and once 2 or fewer are left this month it turns amber, says so and gives the reset date (the Flex screen says it too). High quality warns the same way once 3 or fewer are left.

A bilingual script can be read on screen from "Preview" without downloading (translated files have no preview — the source is not stored, so there is nothing to show alongside).

Practising simultaneous with a draft translation

In the preview, read-aloud reads only the source column, straight through, and stops at the end of it. The translation is not read until you press play again.

1. Under ⚙ choose "Break after each line" (none / 1, 2, 5, 10 seconds).
2. Start, and interpret as you listen. The break is your time to finish the line just heard.
3. When it stops at the end of the source, press play to hear the translation column and compare.

11. Mind maps and "Ask the AI"

Mind maps

"Create a mind map" under a brief or summary draws its content as a single diagram. Rather than turning bullets into boxes, it compresses each node down to the words you would actually say in the booth. It prints on A4 and saves as SVG or PNG.

• Made only when you press the button. "Redraw" gives a different cut of the same content.
• Label language is selectable (Japanese / English / bilingual). Bilingual puts two lines in each node — the word to read on top, the word to say underneath. Labels are written by the AI, so changing the language means redrawing (unlike the body text, which merely switches).
• You can say how to draw it (optional): "more of the figures", "group by theme", "smaller". Nothing outside the source is added — instructions change emphasis, not content.
• It is drawn from the brief or summary that already exists; your material is not sent again. Its deletion date matches theirs.

Ask the AI

Brief and summary screens carry "💬 Ask the AI about this" for the question that comes up while reading.

• You tick which material to use (brief / summary / draft translation / glossary). What is sent is your question, the material you ticked, and the earlier turns of this conversation — nothing else.
• The conversation is not saved — it disappears when you close the page. To keep an answer, press "Add this answer to the material" and it is written into the brief or summary.
• There is a cap on questions per assignment (normal use does not reach it).

12. The live sheet

The High-importance terms, the readings of proper nouns and the numbers, condensed onto A4 to carry in. It works best generated after you have finished editing the glossary. Open it from "🖨 Live sheet" on the glossary screen.

• The order you sorted the glossary into is carried over. Sort by "belongs to" and the sheet comes out grouped by shelf. The sort selector then reads "Glossary order" — choosing a different order and pressing "Update" returns to the sheet's own setting.
• Descriptions can be shown as none / one language / both languages. Terms hold a description per language, so the description written in the language you are about to speak can go on the paper.
• Highlights and groups print exactly as they look on screen. Groups get a rule down the left and are never split across columns.
• Range (High only / all), columns (1–4), orientation, text size and which language goes on which side are all selectable. "Save as PDF" produces the same layout as the screen.

13. Looking a term up from any screen

"How did I translate this last time?" is a question that arrives while you are reading something else. "🔍 Look up a term" at the top right answers it where you are — no trip to the library page and back.

• Put the window where you like. Drag it by the title bar; it opens in the same place next time (⤢ returns it to the start).
• ⧉ opens it as a separate browser window (2026-08-20) — for when what you are reading is Zoom or a PDF and TerpPrep is not on screen at all. Opened from an assignment, it stays fixed to that assignment. The window takes the height of what is in it. The same ⧉ is on "❓ How do I use this?" as well.
• There are two checkboxes beside the search box:
  • ✨ AI — asks the AI for terms that are not in your library. It asks once, the moment you tick it, and never while you are still typing. To ask again after changing the word, press Enter. The answer is labelled "AI suggestion" and kept apart from the library results — it is a guess, not a rendering you have settled on.
  • 📂 This assignment — searches inside this job (glossary, brief, summary, draft translation), for "where did I read that?"
• Results are ordered library → this assignment → AI suggestion: in order of how sure they are.

14. Read aloud

For reading through while travelling or with your hands full. Start from "Read aloud" on the brief, summary and draft-translation screens.

• Click a paragraph to start from there. Where you stopped is remembered, so play resumes from it.
• While playing, ⏮⏭ move by paragraph and ⚙ changes the speed (0.75×–2.0×).
• The voice follows the language of the text. English proper nouns inside Japanese are still pronounced properly. Voices are chosen per language in Settings.

15. Practice mode

"Practise these terms" on the glossary screen turns the glossary into flashcards. No AI is involved — it runs entirely on what is already on screen.

• Which terms — all / High importance only / the ones you marked "again".
• Direction — JA→EN / EN→JA / random per card.
• Order (random or by importance) and how many (20 / 50 / all).
• Timer (none / 3s / 5s …) — for practising how fast the rendering comes.
• Turn on read-aloud to hear the cards as well as see them.

Each library glossary page starts the same thing, for revising a field outside any particular job.

16. Closing an assignment — sorting new terms

When a job is over, sort its new terms into the library. This is the single most valuable habit in TerpPrep.

Sorting is the dropdown in the "belongs to" column (the AI's suggestion is marked ★). Choosing it registers the term immediately. For several terms at once, tick the rows and use the "belongs to" dropdown in the toolbar.

Delete the rows that are not worth keeping. The trick is not to keep everything — junk in the library lowers the quality of every future suggestion.

What the assignment screens say (2026-08-20)

• The assignment list shows, for each job, the client (or "No client") and the glossaries it reads. Same colours as the library: violet = client-only, sky = shared, and 「内」 marks one that sits inside another.
• A job with nothing in it yet carries an 「資料待ち」 (awaiting documents) mark. It goes as soon as material is added or anything is generated.
• The assignment page shows the same thing in two titled columns. Clicking the client opens the library with that client highlighted.
• An assignment inside a project shows the project's name above its own, as a tree.
• Each detail page now names itself — Assignment details / Client details / Project details — because one word can be an assignment, a project and a client at once.

Two rules about names (2026-08-20)

• Project names must be unique. A project is the heading in the assignment list, the tree above an assignment's name, and the prefix its assignments are named after; two with one name makes all three ambiguous. Assignment names may repeat — a 定例会議 under two projects is two meetings.
• Creating an assignment inside a project offers a name ("Project name assignment 1") in the box. Leave it blank and that is the name; the number is the next free one.

17. Your glossary library

• The Library tab is a tree - each client with the glossaries it owns and the shared ones it has been given, and the shared glossaries together at the bottom - and edits any of them directly. It behaves like the assignment editor. There is no save button — notes, moves and deletions take effect as you make them.
• The "belongs to" dropdown moves a term to another glossary (any glossary in the same language pair). If the destination already has that term, that one term is skipped and you are told.
• Terms are added either by sorting them from an assignment or through Import. To rewrite a term, translation or description, use the term page — the change reaches every assignment using it.
• "See all" under the search box opens one flat list across every glossary, editable in place.
• A glossary emptied by moves and deletions can be removed with "Delete glossary" (with confirmation).
• Which shared glossaries a client reads (ServiceNow → IT, SW) is decided on the client's screen — or by dragging a shared glossary onto that client's branch in the tree, and dragging it back out to remove it.
• A glossary can hold others (two levels, no grandchildren). Drag one onto another in the tree to put it inside; drag it out to the branch to take it back out. An assignment that reads the parent reads its children too.
• The + beside a client's name makes a glossary only that client can read; the + on the Shared branch makes a shared one. A client's own glossaries carry a Client only chip.
• A glossary's scope (client / shared) and its parent are changed on its Properties page.
• Tick glossaries in the tree to delete them together, or to merge them (choose which one survives). Glossaries with different scopes cannot be merged.
• The dashboard search bar searches every glossary at once.

The client screen

A client is the record of who the work is for. The list of clients is on the dashboard (there is no tab), and the name opens the screen. New ones are made with + New client just above the list.

• Glossaries only this client reads — made here too.
• Shared glossaries this client reads — added and removed from a list. The order matters: the one added first wins when two of them translate a term differently.
• The redaction dictionary — words added here are reused on this client's next assignment.
• This client's work — its assignments and projects, as a tree.

Client names are stored encrypted.

All terms — the sorting desk

Search all glossaries at once on the dashboard opens one table across every glossary you own.

• Word search looks at the term columns and the reading; full text also looks at the descriptions, the note, the glossary's name and your own columns. Matches are highlighted. Case and character width do not matter.
• It combines with the chips above (client / shared / high importance / AI-filled).
• This screen is a sorting desk: to stay fast over thousands of rows it does not wrap text and does not edit it. Re-filing and bulk selection happen here; rewriting a term happens in that glossary's editor or on the term's own page.

18. Import and export

Bringing in glossaries you already have

Import at the top right of the library screen takes the term lists you have built over the years (Excel / CSV / text). Drag and drop works too.

You choose one of three ways to read the file. All three read the same terms; what differs is what happens to the columns, and whether the blanks get filled.

• Fit the file to my glossary (no AI): only the columns the glossary knows — term, rendering, description, note, importance, source, and any columns you added yourself. Anything else in the file is left behind. No API call, so no cap on how incomplete the file is; blank cells stay blank.
• Fit my glossary to the file (no AI): the same, plus every heading the glossary has no column for becomes a column of your own — for bringing a client's spreadsheet across whole. You can have five such columns, so if there is not enough room the headings that did not fit are named on the result screen rather than dropped in silence. Delete a column you no longer need in the editor's column chooser and import again.
  A column made this way is provisional, and belongs to this import alone. Registering the terms to your library is what makes it a column of your account — until then it appears in no other glossary, and a later "fit the file to my glossary" import treats it as not existing. Columns you never register are discarded when you register that import.
• Fit the file to my glossary, and fill the blanks with AI: the default. Where fields are missing — a list of renderings with no descriptions, say — the AI fills them in. Anything the AI guessed is marked, so you know where to look.

• What you import does not go straight into the library. It lands in a staging glossary first; only the terms you assign a destination to enter the library. This is what stops an AI guess becoming an asset without a human reading it.
• Columns are matched by header name, not by position. There are caps on rows per file and on how many terms the AI completes (the two no-AI modes have no completion cap).
• The more you import, the more the No-AI mode in 21. can find.

Getting data out

• An assignment's glossary exports to Excel / CSV.
• Each library glossary exports to Excel / CSV too (top right of the glossary screen), and the whole library exports to CSV at any time (Settings → Data & privacy). The data is yours.
• To edit in Excel and bring it back, write into the downloaded file and import that file as it is. As long as the Source column is still there, every term returns to the glossary it came from. Adding a notes sheet of your own is fine — only the "Glossary" sheet is read.
• Any other workbook with several sheets is read one sheet at a time: each sheet's heading row is recognised on its own, and the column order may differ from sheet to sheet (a sheet with no heading row is read as a continuation of the one before). The sheet name becomes the default "belongs to" of its terms - except for names nobody chose, such as "Sheet1", or a workbook with only one sheet; a Source column, if there is one, wins.

Creating assignments from Google Calendar or Outlook (2026-09-06; Outlook 2026-10-29)

The events already in your Google Calendar or Outlook calendar can become assignments. TerpPrep reads the calendar only when you open the import screen, and only the events you tick become assignments. Nothing syncs in the background and nothing is created on its own.

Setup, once — Google Calendar

1. In Google Calendar on a desktop browser: the gear at the top right → Settings → pick the calendar in the list on the left → "Integrate calendar" → copy the "Secret address in iCal format".
   ⚠️ Anyone holding this address can read that calendar. Treat it like a password. If you think it has leaked, "Reset" on the same screen invalidates the old address.
2. In TerpPrep Settings → Calendar (Google / Outlook), paste the address and Save. Once saved, only its last few characters are shown.

Setup, once — Outlook

1. In Outlook on the web: the gear at the top right → Calendar → "Shared calendars" → "Publish a calendar". Choose the calendar and "Can view all details", press Publish, and copy the ICS link.
   Published as "Can view when I'm busy", the events have no titles. ⚠️ Anyone holding this link can read the calendar too. To stop, press Unpublish on the same screen and the old link stops working. In a company's Microsoft 365, the administrator may have turned publishing off.
2. In TerpPrep Settings → Calendar (Google / Outlook), paste the link and Save.

Only an address starting with https://calendar.google.com/, https://outlook.office365.com/ or https://outlook.live.com/ is accepted.

Up to five calendars can be set (work, an agency's shared one, and so on). Give each a name and the import screen shows which calendar every event came from. If one calendar cannot be read, the events of the others are still listed.

Each time you want assignments

1. Assignment list → Calendar (the month view) → "📥 Import from your calendar" at the top right.
2. Upcoming events are listed by date. Tick the ones you want and press "Create assignments from the selected events".
3. The event's title becomes the assignment name and its date the assignment date. Client, language pair, mode and the rest are set on the assignment screen afterwards.

Good to know

• Only the title and the date are read — not attendees, description, attachments or location. Events you did not tick are not stored at all.
• The same event is never imported twice. An event already imported shows as "Imported" and cannot be ticked. Delete that assignment and it can be imported again.
• This is not a sync. Changing the date or title in Google afterwards does not change the assignment (edit the assignment instead), and TerpPrep never writes anything back to Google.
• Repeating events (a weekly stand-up, say) are listed once, on their first date, marked "↻". Later occurrences cannot be imported yet.
• Past events are not shown. Timed events land on their date in Japan time.
• Your plan's assignment limit applies as usual. Once it is reached, the remaining events are not created and the count is shown.
• To stop, press Remove under Settings → Calendar (Google / Outlook). The address is deleted; assignments already imported stay. The address is also deleted when you close your account.

19. Notebook — keeping what you looked up

Writing

• Start from “+ New note”. Leave the title empty and the first line of the text becomes the title.
• The text can be written in Markdown (# heading, - list, **bold**). “Preview” shows how it will read.
• Nothing is saved until you press “Save”. Leaving with unsaved text asks first.

Clipping a web page

1. Put the page's address in the box above the list and press “🔗 Clip page”.
2. The text of the page is taken out and placed in the note form. Nothing has been saved at this point.
3. Check what was taken, cut what you do not need (menus, advertising copy), and press “Save”. The address it came from and the date are kept with it.

• Pictures and layout are not kept. Only the words are.
• A page behind a login, or one drawn entirely by JavaScript, cannot be read. Select its text, copy it, and paste it into a note instead.

Hooking a note to a client or an assignment

• Choose a client or an assignment for a note and it also appears on that client's page and on the assignment's page. Choosing an assignment hooks the note to that assignment's client as well.
• Deleting the assignment does not delete the note — only the hook goes. What you learned about a speaker stays for the next job.
• The assignment's own “Notes” field (the line or two you write when creating it) is a different thing: it belongs to the assignment and is deleted with it.

Finding and tidying

• Search covers titles and the full text. It ignores case and the difference between full-width and half-width characters. You can narrow by client and by kind (notes / clips).
• 📌 Pinned notes come first in the list.
• Archiving takes a note off the list and out of the search. It is under “Archived” and can be restored at any time.

Exporting and deleting

• “⬇ .md” on a note exports that one; “Export all (.zip)” on the list exports every note, archived ones included, as Markdown files.
• Deleting cannot be undone. Export what you need first.

Limits and plan

• One note holds up to 100,000 characters, and one account up to 3,000 notes (20 million characters in all). Anything over is refused rather than cut short.
• Writing, editing and clipping need the Basic plan (the trial counts). Without a plan you can still read, search, export and delete the notes you have.

20. What is sent, and what is deleted when

• Your data is never used to train AI models.
• Data is never shared between users.
• What you uploaded and when it was deleted is visible to you in the audit log (Settings → Data & privacy).
• The glossary library can be exported in full at any time.

The full policy is on the data-handling policy page. It is public — no sign-in required — so you can show it to a client or an agency as it is.

If you want to check before every send

Settings → Data & privacy → "Confirm before sending to the AI" shows, before each run, which files are about to be sent, how large they are and where they go, and waits for you to press Send.

On the same screen, "Default new draft translations and summaries to delete-on-download" makes them disappear when you download them rather than after seven days (still changeable per run).

For ones already generated, "Delete now" on the draft translation and summary screens removes them from the server on the spot instead of waiting out the seven days (it cannot be undone — download what you still need first).

21. Using it without uploading — anonymisation and No-AI mode

1. Anonymise before uploading

TerpPrep already deletes the source material the moment generation finishes, deletes the outputs (draft translation, document summary, document-course brief) after 7 days, and keeps only the glossary, encrypted (→ 20.). Even so, some assignments cannot have a client name pass through a server even once. Anonymisation is the layer in front of all that: the name is gone before anything reaches the server.

How it works

1. You build a redaction dictionary (the list of words to hide), per client.
2. When you choose a document, those words are replaced inside your browser with placeholders such as [COMPANY-1].
3. You review the result in a preview, then press “Next” to send it. The unredacted file never leaves your computer.
4. The AI is told that [COMPANY-1] is a placeholder and must be kept as it is. The placeholders stay in the outputs (→ “What the outputs look like” below).

Step 1 — Build a redaction dictionary (in the library)

• In the library tree, use the “+” to the right of the client's name to create a 🛡 redaction dictionary (→ 17.). Always give it a name — it is a different thing from the client glossary, and the same name for both is confusing. A client can have any number of them.
• One line = one word: “word to redact”, “type”, “replace with”. A replacement such as [COMPANY-1] or [PERSON-2] is proposed automatically, but you can write your own (“Company A”, “Mr X”).
• Matching ignores case (servicenow and SERVICENOW both match). Variant spellings (Service Now / サービスナウ / SN) must be registered as separate words; they are not expanded automatically.
• The four quick patterns (any number, any email, any phone number, any URL) always sit at the top of the list, switched off. Tick one to enable it; it is then saved to the dictionary.
• CSV import and export are available, in the same format as the free Document Anonymiser app (term, replacement, type, isRegex, label), so the two can be used interchangeably.
• A redaction dictionary is applied automatically to that client's assignments (every dictionary of that client, together). An assignment with no client gets no saved dictionary — only words typed on the spot are used, for that run alone.

Step 2 — When putting material in: “Anonymise”, then “Next”

1. On the input screen (or the Flex preparation panel), choose your files or text and turn on “Anonymise before uploading” (off by default). The panel shows the names of the dictionaries in use and their words. You can add words here too, but for this run only — they are not saved to a dictionary (dictionaries themselves are edited in the library).
2. Press “🛡 Anonymise”. The replacement runs inside your browser and the preview shows the number of hits by type and the redacted content. Nothing has been sent at this point.
   At the same time, the images, video and embedded objects (a pasted Excel sheet, for instance) inside the material, and the file's preview thumbnail, are emptied (the count is shown). Redaction works by replacing text, so it cannot reach information inside an image, such as a screenshot of a customer list. Generation only ever uses the text, so the outputs are unchanged.
3. Check the content.
   • Pasted text is shown in full, as it is. For files, “Show text” shows you the content (PowerPoint and Word do not let you see inside directly, so please do check).
   • If you get the warning that certain words are still inside the file, do not leave it. Those words remain somewhere redaction cannot reach (sheet names, formulas, file properties and the like). Fix them in the original file, or use a different input.
   • “⬇ Save redacted file” downloads the redacted file to your machine. You can use that file for something else entirely without uploading it to TerpPrep.
4. If it all looks right, press “Next”. Only the redacted data is uploaded, and you move on to the output screen. “Next” cannot be pressed until you have pressed “Anonymise”. If you change the words or the material, the preview is discarded, so content you have not reviewed is never sent.

Step 3 — After sending

• “Where your material is” at the top right (your machine / server / AI) shows where the material is right now, with a “redacted” mark.
• The processing history (Settings ▸ Data & privacy ▸ View processing history) records when, what, and whether it was redacted.

How matching works — how a word is found

Underneath, anonymisation is search-and-replace on text. But it is not a naive replace; it follows these rules.

1. Each word becomes a search pattern. A registered word is turned into a regular expression (a text pattern) internally.

• Punctuation (. ( + and so on) is matched literally, not as regular-expression syntax.
• Case is ignored. Leading and trailing spaces are ignored; an empty word is skipped.

2. You can register regular expressions. Set the isRegex column of the CSV to true and the word is used as a regular expression, as written (no boundaries are added).

A broken regular expression (an unclosed (, say) is skipped on its own and the rest carries on. One typo must not leave the whole document unredacted.

3. Quick patterns. “Numbers, emails, phones, URLs” are ready-made regular expressions (the same as Document Anonymiser's).

4. One pass, left to right, longest match wins. Every word's pattern is applied to the original text at once, and the matches are filled in from left to right. Where candidates overlap, the longer one wins at the same position; at equal length, registration order decides. Replacing word by word instead would let a later rule hit inside a placeholder an earlier rule put in — “any number” would catch the “1” in [COMPANY-1] and turn it into [COMPANY-[NUMBER]], breaking the very placeholder scheme the AI is told to preserve. A single pass never re-reads its own output. Longest-match is what lets “any phone” take 03-1234-5678 whole (with “any number” first it would be chopped into [NUMBER]-[NUMBER]-[NUMBER]).

5. Words split by formatting are joined first (PowerPoint / Word / Excel). Office files split a string into separate fragments every time the formatting changes mid-sentence. Make Service bold and Now plain and the file holds them in two separate boxes, where ServiceNow would never match. Anonymisation joins the fragments of each paragraph before matching and writes the redacted text back into the first box. The formatting that survives is the first fragment's.

6. How placeholders are numbered. When you register a word, a numbered replacement is proposed per type: [COMPANY-1], [COMPANY-2], [PERSON-1]… (the highest number of that type, plus one). The square brackets keep the placeholder intact through extraction and translation. The same word always gets the same replacement (one word = one placeholder). Conversely, if you register ServiceNow and サービスナウ separately and give both [COMPANY-1], the AI sees one company — if that is what you want, line up the replacements yourself.

7. A residual-word check after replacing. Once replacement is done, the whole file (including the parts outside the body text) is searched once more for the registered words (the non-regex ones). Anything found is reported as words still inside the file. This is how you find a word left in a place replacement does not reach — a sheet name, a document property.

What gets redacted

The supported inputs are PPTX / DOCX / XLSX / PDF / .doc / .xls / text (pasted or .txt). Password-protected files cannot be read.

What does not get redacted

• Words you did not register. Nothing is auto-detected. Having the AI hunt for likely proper nouns would mean sending the material to the server to do it — the very thing this exists to avoid. Please check for yourself that nothing has been missed.
• Variant spellings. A different spelling is a different word.
• Text inside images cannot be read — which is why the images are emptied instead.
• Scanned PDFs (where the text is an image) cannot be redacted, so you get an error and nothing is sent. Use a PDF that contains text, or paste the text in.
• Excel sheet names, numbers and formulas.
• Flex's pasted text, web pages, photos and notes are sent immediately; only files are anonymised. Mix one plain-text paste in with redacted files and that batch is no longer “redacted”.
• Restoring the placeholders in the outputs is not available yet (→ next section).

What the outputs look like

• Briefs, summaries and draft translations keep [COMPANY-1] as it is. Read them against your own key. Restoring the names automatically on download is under consideration and not yet available.
• The glossary never contains a term with a placeholder in it. [COMPANY-1] is not something to memorise, and because a glossary entry is stored permanently in the term library once promoted, leaving it there would put it forward on every future assignment for that client. Real terms around a redacted word are extracted as they always were. [COMPANY-1] can still appear inside descriptions and examples as context.
• The AI does not know who [COMPANY-1] is, so the brief cannot cover that company's own background (ownership, history). The effect on the glossary, the summary and the draft translation is small.

To explain this to a client, the data handling policy shows in a diagram where material goes and when it is deleted. It can be printed and handed over.

2. Building a glossary without AI

If anonymisation is preparation for uploading, this is the option where you do not upload at all.

1. On the assignment page open "Add from material" and choose your files.
2. For the mode, choose 🔒 "Match against my library only (no AI)" instead of ✨ "Create with AI".

The material is read inside your browser and matched against your library. Terms found in it enter the assignment glossary with the renderings you settled on. What reaches the server is only the IDs of the terms that matched; not one character of the material leaves your machine, and no AI is called.

22. Account security

• Two-factor authentication — Settings → "Two-factor authentication" registers an authenticator app (Google Authenticator, 1Password, …). With it on, each sign-in asks for a six-digit code.
• Recovery codes — turning 2FA on shows eight single-use codes. They are the only way to get back in by yourself if you lose the authenticator, so print them or put them in a password manager. They can be reissued at any time.
• If you lose both the app and the codes, contact support — we can clear it after verifying your identity.
• Audit log — Settings → Data & privacy shows what was uploaded when, and when it was deleted.

Changing how it looks

Settings → Appearance offers a colour theme and a text size. Both are saved on your account, so the office machine and the laptop you take to the venue look the same.

• Four themes — Pine (the default), Indigo, Wine and Monochrome. Monochrome is the highest contrast of the four, for a badly lit venue or a projector. Each option is drawn in its own colours, so you can see it before you choose it.
• Text size (small / standard / large / extra large) applies to what you read — glossaries, briefs, summaries. The header and tabs keep their size, so the larger text gains the room rather than losing it.
• A theme never changes what a colour MEANS: amber is still a warning, red a danger, violet a client, sky a shared glossary. Your highlight pens keep the colours you chose, and the live sheet still prints black on white.
• Light / Night (2026-08-20): a night mode, on a separate axis from the four themes and working with all of them — "reading in a dark room" and "which colour do I like" are different questions. Switch it in Settings → Appearance or in one click from the user menu, since it is the one appearance choice whose reason changes during the day. Printing is always black on white.

23. Checking, changing and cancelling your plan

Settings → "Plan & billing" holds:

• Your current plan and, on the free plan, how many of this cycle's assignments you have used and when it resets. The cycle runs from your sign-up date, not the calendar month.
• Changing plan — the button opens the payment screen (Stripe).
• Manage payment — change card, download invoices and receipts, edit the billing details a receipt is addressed to, and cancel. Cancelling stops charges from the next renewal; paid features stay available until then.

The Accounting tab (invoices) and your plan

Making invoices is a Basic plan feature. That is why the "Accounting" tab is not shown in the header on the Free plan.

• On Basic (including a trial or a free period), you can create, edit and issue invoices.
• Moving to Free does not delete the invoices you made. If you have any, the Accounting tab stays, and you can open them, export them as PDF or Excel, export the accounting CSV, and delete them - but not create, edit, issue or re-issue.
• Back on Basic, you can create and edit them as before.

Inviting a colleague

The same screen invites colleagues by email. When someone you invited moves to a paid plan, one of your months and their first month are discounted. There is no limit on how many.

• Enter their address and press "Send invitation"; TerpPrep sends the letter.
• Unticking "Include my name" sends it as coming from a TerpPrep member, without naming you.
• Discounts earned while your plan is free are not lost — they are applied afterwards, in order.
• The outcome (sent / already registered / failed to send) is shown on screen. There is an hourly cap on invitations.

24. What TerpPrep cannot do

These are limits of the design. No setting works around them. Where there is a way round, it is given.

There are also caps on size per file and per assignment, and on how many draft translations and regenerations you can run. The numbers per plan are on the pricing page.

25. FAQ

Generation never finishes / it failed

Large material takes a few minutes. You can leave the page — a notification appears at the bottom right when it finishes. If it failed, try again; if it keeps failing, tell us the file type and size.

The terms or renderings are not very good

What is generated is a first draft. Correct it in the editor — your corrections accumulate in the library and feed later assignments and draft translations. If terms you want are consistently missed, or noise is consistently picked up, please tell us.

My upload is refused

Either the file size or the assignment's total volume is over the limit. Try shrinking before upload first (see 6.) — a PowerPoint with video in it can drop to a fraction. If it still does not fit, upload an extract of the pages that matter (we deliberately do not truncate quietly and carry on).

I want to see an old assignment's material again

Source material is deleted after generation, so it is not on TerpPrep. The outputs (the glossary, and agenda-course briefs) are still on the assignment.

I corrected the glossary but the next assignment does not show it

Corrections in an assignment's editor apply to that assignment only, so that a client-specific rendering stays specific. To change the library's rendering, edit it on the term's own page (see 8.).

I forgot my password

Use "Forgot your password?" on the sign-in screen. A reset URL is emailed to your registered address. If nothing arrives within a few minutes, check your spam folder.

I lost my authenticator when I changed phones

Use a recovery code (see 22.). If both are gone, contact support.

Does it work on a phone?

It runs in the browser. The glossary editor and the live sheet assume a wide screen, so preparing on a computer and checking on a phone works best.

26. Contact

Questions and bug reports to support＠terpprep.com. For a bug, (＠ is full-width - please replace it with @ before sending)what you did and what happened — with a screenshot if you can — is the most useful thing you can send.