The Memory tab
The agent's mind drawn as a sky — how to read it, search it, and open one memory at a time.
The Memory button in the menu bar — the brain, just left of Theme and Settings — opens a tab that draws an agent's memory as a sky: every memory a light, linked memories wired together, neighbours in meaning drifting near each other. It runs on the same engine as the 3D view of your code, so it moves the same way: left-drag orbits, wheel zooms, right-drag pans, and the navigation dial and camera buttons sit where they do over the code.
It's a place to look, not to edit. Nothing here creates a memory or deletes one — the first conversation in a project grows its agent, the walker and the agent fill it, and the Project Manager is where memory is cleared.
Whose mind#
At the top centre, mind names the agent you're looking into. With more than one agent on your account it's a dropdown: ★ marks this project's default, · marks the others allowed into this project, and each row carries its memory count. A brain allowed into several projects is one brain; the tab opens on the one this project uses and lets you look into any other.
Reading the sky#
Each memory is coloured by its type — insight, decision, feedback, reference and so on — and the types are listed down the left rail with their counts. Click a type to hide it from the sky; double-click for only that type. Under the types, the importance floor: every memory carries an importance from 0 to 1, and raising the floor sinks the ones below it into the dark. Nothing is hidden from recall by either control — a search still finds them and still lists them.
Lights that are wired together are linked memories — one refers to the other by name. Lights that sit close without a wire are near in meaning.
Switch on Info in the menu bar's Panels bubble and the mind's own numbers appear top right: memories, links, neighbours, and promises — links to memories that haven't been written yet.
Searching it#
The box at the bottom reads Ask the memory…. Type a question and press Recall. Hits rise as a
list beside the sky and their lights come forward; everything the mind didn't answer with sinks.
Clear puts the sky back.
Above the hits, one word says what kind of answer this is. recall means the mind answered — the top hits stood out, and each row carries a score. association means it didn't: what came back is only loosely related, and the scores are replaced by dots so fourteen near-identical percentages can't pretend to be a ranking. Nothing in this mind answers that is also an answer.
Opening a memory#
Click a light, or a hit, and a card opens at the right: the memory's type, its name, its one-line
description, and four numbers — importance, how many times it has been recalled, the date
it's from, and when it was last revised if that differs. The body opens at about ten lines;
Show more unfolds the rest. Any [[links]] at the bottom are clickable and open that memory in
turn; a link to a memory that hasn't been written yet is greyed, not written yet. A quiet
footnote gives the day it was carved when that isn't the day it happened — a memory written weeks
after the fact, from an import.
Escape closes the card.
When the sky is empty#
The tab tells you why, in one line: Semantix Memory is off (with an Open Settings button); No project open; The memory organ could not be reached; or No memory yet — A project's first conversation grows its agent. Come back after you have talked. An agent that exists but has never had a conversation filed shows holds nothing yet — its conversations have not slept.
What it isn't#
It isn't where memory is managed. Making an agent, allowing it into a project, clearing what it remembers or forgetting a project all happen in the Project Manager. And it isn't the code graph: the two skies share an engine and a look, but the code one is your project's files and this one is your agent's mind.