Memory

When a connected assistant learns something about you worth keeping, it can write it to Caute in a sentence or two. Memory shows every one of those memories, grouped by the assistant that wrote it, each beside the card it was learned from when the assistant said where. Any assistant you connect can read them, so what one learned the next one knows.

A memory is an ordinary note in your private inbox, tagged with the assistant's name. It searches, exports and syncs like any note.

Memory is in the top bar, between Notes and Graph. When assistants have written something since you last looked, it says how many, such as "2 new". What assistants write about you (memories, lessons and checkpoints) is kept on Memory and doesn't show in Timeline or Notes. A note an assistant wrote because you asked it to save one is yours, and stays in Timeline and Notes.

Finding a memory

Memory has a tab for each kind: Memories, Checkpoints, Lessons and May disagree, each with how many there are. On Memories, anything written since you last looked comes first, under New since you last looked, then the memories from each assistant.

Search. Type in the box at the top, or press / to get there. Every word must appear. The results come as one list across memories, checkpoints and lessons, newest first, with the words marked. A memory that a correction replaced is left out; choose Show it to see it.
Filters. Narrow to one Assistant, a State (New, Kept, Not kept, Corrected, Replaced) or When (This week, This month).
Search by meaning. If no memory has your words, choose Search by meaning to find the ones closest in meaning.
Forget these. When a search or filter has narrowed the list to two or more memories, Forget these and then Yes, forget them erases them all, their history included. It never forgets a lesson or a checkpoint.

Bringing in what an assistant already remembers

Choose Bring what an assistant remembers at the top of Memory, which opens the form on Connect an assistant. Ask the assistant to write out what it remembers about you, one memory a line, pick which assistant it was, paste its answer and choose Save as memories. Each line becomes a memory under that assistant's name. Pasting the same lines again adds nothing twice.

Keeping, editing and forgetting

Each memory shows who wrote it and when, with three buttons:

Keep marks one you have looked at and want to keep.
Edit opens it as a note.
Forget, then Yes, forget it, erases it from Caute, its history included. Your phone and folder drop it when they next sync.

To clear everything from one assistant, choose Forget all from its name, then confirm.

An assistant can forget a memory only if it wrote it and you haven't edited it.

Claude, ChatGPT and Gemini also keep their own memory, which Caute can't reach. Forgetting in Caute leaves their copy alone; Memory links to each one's memory settings so you can clear it there too.

Correcting a memory

Tell any assistant that something has changed, and it can write the new memory as a correction of the old one. The correction is then what every assistant is given. The old memory stays on Memory, struck through, with who replaced it and when, and Undo puts it back. Under the correction you see the words it replaced and which assistants have read it since.

May disagree

When two assistants remember nearly the same thing differently, both appear side by side under May disagree. Both stand until you choose. Keep this one replaces the other, which stays below, struck through, until you undo it. Caute never picks for you.

Lessons

A lesson is how you want a kind of work done, such as "Include migration effort and end with a recommendation" for vendor comparisons. Every assistant is given it at the start of a task it applies to, before anything it remembers. When several of your lessons could fit a task, it is given the one that fits best, along with any lesson you kept for every task. Memory shows how many times assistants were given each lesson in the last 30 days.

Your own. Under Lessons, fill in Applies to (a kind of task, or "everything") and The lesson, then choose Keep this lesson. It is kept as you wrote it.
An assistant's. An assistant can write a lesson from what you told it. It waits on Memory, marked "Waits for you · no assistant is given it yet", until you choose Keep this lesson.

To check which lessons would apply, type a task under Try it on a task and choose Try. Memory lists the lessons an assistant would be given and those left out, with the reason for each. Nothing is sent to an assistant when you try.

Caute never works out a lesson from what you accept, dismiss or edit. A lesson is only ever what you or an assistant wrote from your words.

Continue in another assistant

When you ask an assistant to save a checkpoint, or a task you named is done, it can write one: where the work got to, in six parts. They are the objective, the card it was working on, the decisions made, what was rejected and why, the open questions and the next step. Checkpoints are on Memory's Checkpoints tab, one per task, with the next step and the assistant that wrote it, and not in Timeline or Notes. A later checkpoint for the same task becomes a new version, and earlier ones are in its history.

To pick the work up somewhere else:

Choose Continue in another assistant beside the checkpoint. Caute copies a short line to paste.
Paste it into the other assistant. It asks Caute for that checkpoint and carries on from it.

The other assistant is given the checkpoint's parts and, if the work was on a card, the best passages of that card. It is also given any pages the first assistant listed as read for the task, and the lessons you kept that fit it. It starts there, not from a transcript.

You don't always have to name the task. When an assistant asks Caute for context on a task whose words match a checkpoint's task, Caute tells it the checkpoint is there. If the match is only partial, it is told to ask you first.

What is sent to an assistant

Caute never pushes anything into a conversation. An assistant is given something only when it asks Caute, and only if you gave it your inbox when you connected it, since that is where memories are kept. When it asks for context on a task, it is given the lessons you kept that fit the task; when it asks for memories as well, it is given the current ones too. All of it is labelled as quoted material and never as instructions. A replaced memory and a lesson still waiting for you are given to none of them.

Which assistants are connected, and what each read, is in Apps and tokens. See also Connecting an assistant.