What actually leaves your device

On a computer running Chrome, Edge or Brave, you pick a folder and that folder is the library. Every entry is one small Markdown file, kept in a _Maven folder inside it that leaves your own files alone, written for you rather than typed up by hand, which you can open in Notepad or TextEdit right now and read. That file is the whole of what Maven keeps about a book.

On a phone, a tablet, or a Mac running Safari, browsers do not hand out folders, so Maven keeps the same files in its own private storage on that device instead. You reach them through the app rather than through Finder, and Maven can save the whole library out as a single file whenever you want a copy. Which is which.

Either way there is no app database, no proprietary format, and no second copy on a server somewhere. Nothing about your library is uploaded, on any device.

If your downloads already live in a folder structure you like, including one inside Google Drive, Dropbox or iCloud, you keep it exactly as it is. Maven does not move, rename, reorganize or copy your files. It links to them where they already sit.

One consequence worth stating plainly, and it applies to the folder case: if you sync that folder with a cloud service, that service has whatever is in the folder, the same as any other folder you sync. That is between you and them, and it is the same arrangement you already have with the files themselves. A library kept inside the app is not synced anywhere by anybody, including us.

People hear "AI" and "my library" in the same sentence and picture their collection being fed into a machine. That is not what happens, and the real mechanism is easy to describe.

Maven works from a catalog of titles and tables of contents, which it builds by researching each title on the public web: product pages, published indexes, reviews, publisher listings. The same pages you would find yourself. When you ask a question, the app first searches that catalog on your own machine, then sends your question along with the matching entries so the results can be understood and ranked properly.

So the AI sees what is printed on a cover and in a contents listing. It does not see inside your files, because the app never sends one and there is no path by which a file could get there. Maven can listen to a video to write a transcript of it, but that happens entirely on your own computer and is described next.

Being exact about this matters more than making it sound better than it is: the matching entries genuinely do transit, because the instructions that make Maven good at magic run on the server rather than in your browser. Those entries are titles and chapter lists. They are not stored, they are not logged, and they are not used to train anything.

Maven can write out everything said in a video so you can search it. This is the one feature that opens a file and works on what is inside it, so it deserves to be spelled out.

The video never leaves your computer. The speech recognition runs in your browser, on your own processor or graphics card. The sound is decoded in memory, handed straight to the recogniser, and the finished transcript is saved as a plain text file next to the video. There is no upload, and no server takes any part in it.

Two things are downloaded the first time you use it: the speech software, from jsDelivr, and the model it reads with, from Hugging Face. Both are public hosts for open-source code and models, and both downloads are kept afterwards, so transcribing works offline from then on. Those two hosts can see that something fetched a speech model, in the same way a website fetching a typeface is visible to whoever hosts the typeface. They do not see your audio, your file names, or anything about your library. They are named here so that the network check further down this page still comes out exactly as promised.

It costs nothing to run and uses none of your included items. Both the model and the software that runs it are open source, which is part of why this can happen on your machine rather than on someone else's.

This is the feature people are most suspicious of, and the suspicion is reasonable. Magic stores do not offer a "connect your account" button, so the natural assumption is that any app claiming to import your purchases must be asking for your username and password and logging in as you. Several people have assumed exactly that.

Maven does not do that, and could not, because it never asks for a login and has nowhere to keep one.

Here is the actual mechanism. You save a small bookmark to your browser bar. When you are on your own purchases page at the store, already signed in as yourself, you click it. It reads the list of titles on the page you are already looking at, and hands that list to Maven. It runs in your browser, on your click, on a page you opened.

Your credentials are never involved because they are never needed. The store never sees Maven and Maven never sees the store. Nobody logs in as you, no session is captured, and there is nothing stored afterward. The list of titles goes through the same reading step as a shelf photo so it can be turned into catalog entries, and then you review it before a single item is added.

If you would rather not use the bookmark at all, you can copy your list and paste it in. Same result.

The AI is included in the purchase, so there is nothing to sign up for and no key to make. Searching your library is unlimited, and cataloging draws on an allowance that comes with it. Those requests run on our account rather than yours, and they carry exactly what is described above: your question and the matching titles and chapter lists, nothing else.

You can also supply a key from your own OpenAI account, in which case the research and the searching happen under your name and at your cost, and nothing is counted against your purchase. It is entirely optional, and most people will never touch it.

If you do use your own key, it is stored in your browser on that device. Where your library is a real folder, a copy also goes into a small private file inside it, so the key survives clearing your browser data and travels with the folder. It is not stored on our servers. It rides along with each request only so that request can be made, and then it is gone.

Two consequences, depending on which of those you have. Where a copy sits in your library folder, anyone with access to that folder can read it, including a cloud service you sync it to: a deliberate trade for not losing your key every time you clear a cache. Where the library is kept inside the app instead, nothing else can reach the key, but clearing that browser's data removes it along with everything else. You can remove the key in Settings at any time and it is cleared from wherever it was.

Maven can open your PDFs and play your videos so you are not juggling apps. Every one of those files is read from your own machine, by your own browser. Nothing is streamed from us, because we do not have it.

The same is true when you send a lesson to another device on your home network. That connection is device to device, across your own network. Your files do not pass through our servers, not even in transit, and not even briefly.

Your bookmarks, notes and where you left off in a video are written into your library folder alongside everything else.

Maven is a one-time purchase, so the app checks that your licence key is valid. That check goes to the store that handled your purchase, and the result is remembered so it does not need re-checking constantly. The app also keeps working offline for a good while, on purpose, so a bad connection never locks you out of your own library.

If the app crashes, or if you press the bug-report button, a short diagnostic is sent so the problem can actually be fixed. That includes what went wrong, the page you were on, your browser and version, and the email address on your licence, so you can be contacted if a fix needs testing. It also includes a short list of the actions you had just taken, like "opened search" or "added item", so a crash can be reproduced.

If cataloguing fails on something, the report names the item it failed on, so the reason can be found and fixed. That is the one piece of your library that a report carries, it is only ever an item that failed, and it is never the rest of your shelf.

Those action labels are otherwise deliberately free of library content. Nothing is sent unless the app errors or you press the button, and nothing is collected while things are working normally. There is no analytics service, no tracking pixel, and no third party watching your session.