Teach Tatsu how you work
Explaining your role, your clients and your style again in every chat wastes time. Tell Tatsu once, and check what it keeps in one place you can read and change.
Takes about 4 minutes
Steps
Facts and instructions: two kinds of memory
A tool that remembers you keeps two kinds of things. Facts describe you and your work: your role, your clients, how you sign off. Instructions say how you want work done: short answers, page numbers, a formal tone with clients. Some instructions you write yourself; others a tool learns from how you react. Facts help when they are relevant; instructions apply every time, so keep them short.
Whatever a tool remembers, you should be able to read it, correct it and delete it. Google's Gemini, for example, lets you edit or delete each piece of saved info.
What Tatsu can do
- Remember a fact you tell it, such as your role or how you sign off, and say so in one short line.
- Save a preference when you state it (“always…”, “from now on…”) or make the same correction twice, and apply it from then on.
- Follow your “Workspace Instructions” in every chat, and a project's own TATSU.md inside that project.
- Bring in what another tool remembers about you, after you review it.
What Tatsu cannot do
- Treat a pasted or forwarded email as a fact about you. It answers, then offers to save it.
- Turn memory off. Forget a fact, or clear your instructions, to stop Tatsu using them.
- Spot secrets in an import. Uncheck passwords, account numbers and anything private before you add it.
How it fits with your Library, memory and apps
Memory holds how you work, your Library holds your documents, and an app holds your records. Tatsu uses all three: your sign-off and tone shape a reply in Draft a reply and send it, while names and amounts in a reply still come from your files, with a number to open each one. A project can carry its own TATSU.md, so one client's rules apply only in that project's chats.
You stay in charge
- You can correct or forget any fact. A forgotten fact goes to Trash first, where you can restore it for at least 30 days.
- An import adds nothing until you choose what to keep.
- Forgetting a preference removes it. Tatsu can learn it again if you state it again or make the same correction again.
Privacy
- What Tatsu remembers is kept as notes in your workspace, encrypted at rest, until you delete them or your account.
- Your workspace is never sent to training datasets, and the model providers that answer your questions commit to no training and to keeping none of your content afterwards. Reading an import is one more request under the same commitments.
- You can see the notes themselves: in the Library, turn on system files and folders, then open .tatsu, then user.
Questions
- Where can I see what Tatsu remembers about me?
- In Settings, “Memory and instructions”: facts under “What I Know About You”, and how you want work done under “Workspace Instructions”, which also holds the preferences Tatsu saves. Those in its learned list also show under “How Tatsu Adapted to You”, where you can forget each one.
- What belongs in “Workspace Instructions”?
- How you want work done, in a few short lines: length, tone, language, how to cite. Facts about you belong in memory. The box holds up to 2,000 characters and is saved as TATSU.md.
- How do I bring my memory over from ChatGPT?
- On the same page, choose “Import memory”, pick ChatGPT and copy the prompt. Paste it into ChatGPT, paste the reply back, choose “Read my memory”, and add what you keep. Claude and Gemini work the same way.
- Can I turn memory off?
- No. Forget what you do not want used under “What I Know About You”, and “Clear instructions” empties your instructions.
- Is my memory used to train models?
- No. Your workspace, memory included, is never sent to training datasets.