Frequently asked questions

A diary you can talk to. You write, an AI model running on your own computer replies, and everything is kept in one encrypted file that only your passphrase or security key opens. Its memory brings back what matters from earlier conversations, so a diary can go on for years without the AI losing track of it.

It’s free for any use, personal or commercial. There are no ads, no account, no subscription and no paid tier. The app has no network code, so it has no way to collect anything about you either.

The license asks one thing: get it from this site rather than passing copies around, so the published checksums stay meaningful.

No. The source code isn’t published. The license does explicitly allow you to take the app apart: decompile it, debug it, watch its network traffic and publish what you find. The security page lists checks anyone can run.

Only to download the app and a model, which you do with your browser. After that it works identically with the network unplugged. The app never connects to anything, not even to check for updates.

64-bit Windows 10 and 11, and 64-bit (x86-64) Linux, with at least 8 GB of RAM (16 GB for the popular 7–9B models). There is no macOS, phone or ARM version yet. See the full requirements.

No. Your diary never leaves your computer, and it is encrypted with a key made from your passphrase or security key. There is no copy of that key anywhere else, no master key and no back door.

No, and nobody else can either. Without the factors that open it, a vault can’t be opened by anyone. That is the whole point, but it means you should write your passphrase down and keep it somewhere safe, or enroll a backup security key or passphrase.

No. The AI is a model file on your own disk, run by the app on your own processor or graphics card. No part of a conversation goes anywhere.

Yes. The sync service only ever receives the encrypted file, although it can see its size and when it changed. Two cautions: lock the vault and let the file finish syncing before you open it on another computer, and never have it open on two computers at once. Two copies changed separately can’t be merged into one.

If the vault was locked, they get an encrypted file they can’t read. If it was open, they see whatever is on the screen until it locks itself, so keep auto-lock short. Keeping your diary private covers the rest.

No. Without one, the model runs on your processor. Replies take longer, but they are just as good and just as private. A smaller model (3–4 billion parameters) keeps things responsive on a processor alone.

Yes. The app uses llama.cpp’s Vulkan backend, which supports NVIDIA, AMD and Intel cards with an up-to-date driver. Only NVIDIA has been tested so far, so AMD and Intel are expected to work rather than confirmed. If a card isn’t picked up, the model simply runs on the processor.

A 7–9 billion parameter instruct model in Q4_K_M suits most computers with a graphics card. Choosing a model has a table by graphics memory.

LM Studio, yes: point the app at LM Studio’s models folder and it finds every model in it, subfolders included. Ollama stores its models under its own file names without the .gguf ending, so they don’t show up. Download the GGUF version of the model instead.

No. It knows what the model learned in training, plus what you have told it. That also means it can’t tell you today’s news, though it does know today’s date and time.

As long as you like. The model can only read a limited amount at once, so before each reply the app puts together the most useful page for it: your latest messages, a summary of earlier ones, pinned facts and anything relevant it finds in the rest of your diary. How memory works explains it, including what it can miss.

Yes. Every vault is a separate file with its own passphrase or key, conversations, memories, personas, settings and model choice. One vault is open at a time. Separate vaults are also the way to keep parts of your life truly apart.

Lock the vault and copy its file, just like any other file. Copy your model too, or download it again. Taking a vault to another computer explains how the app finds the model there.

The app never updates itself, so check this site now and then. Make a backup first, then run the new installer over the old one, or replace the portable folder or AppImage with the new one. Your vaults stay where they are, and a vault from an older version is upgraded the first time the new version opens it. After that, older versions can’t open it, which is why the backup comes first.

The builds aren’t code-signed yet, so Windows SmartScreen doesn’t recognize them. You can check the file’s SHA-256 checksum against the one published here, then follow these steps to run it.

No. Uninstalling removes the program only. Your vaults are wherever you saved them, and they stay there until you delete them yourself. If the app ever crashed, a small crash log may also remain in %LOCALAPPDATA%\MyThoughtSafe; it contains no diary content, and you can delete that folder.

The app’s screens are in English. You can write in any language your model understands, and most current models handle many languages well.