Several things write to a workspace's memory, and they run in different places.
Knowing which is which answers most questions about why something did or did not
happen.

## The AIs people connect write to the memory

Claude Code, Codex, ChatGPT, Claude, Cursor, an OpenClaw of the team's own, or
any other AI someone [connects](/docs/tools/assistants) reaches the workspace
over one address. With Memory at Read & Write it searches what is there, saves
notes, and changes them, while someone is using it. It runs wherever that AI
runs, on a laptop or in the vendor's cloud, and stops when that person stops.

Each AI saves more reliably with the lines under Make it a habit, on
[Connect an AI](/docs/tools/assistants), in its standing
instructions. Every action it takes is on the
[activity record](/docs/tools/activity), under its own connection.

## A workspace's own AI runs the scheduled work

On Team, the workspace has an AI of its own, running on the workspace's
server in the cloud. It answers in **Chats**, in Slack, Telegram and iMessage,
and it runs the skills on the **Schedule** page at their times, with every
laptop closed. It files what it learns as it goes. It spends the workspace's
monthly AI credit.

Personal has no AI of its own, so it has no chat and no Schedule page. A
repeating job there is a scheduled task in the AI you connected, which reads and
writes the memory over the same connection:
[Use the memory from other AIs](/docs/workspace/memory-anywhere)
has the words to paste. Moving the same workspace to Team puts the schedule in
the cloud.

## GBrain makes every note searchable

Whichever AI saved a note, GBrain indexes it for search on the workspace's own
server, usually within minutes, on every plan. Nothing has to be set up for this,
and no AI has to ask for it. Search finds a note by what it means as well as by
its words once it is indexed.

## Overnight upkeep keeps the memory tidy

On Team, the workspace's server looks after the memory once a day, usually
overnight. It checks each note is well formed, mends the links between notes,
pulls the facts and links out of new ones, indexes anything that was missed,
notes the pages nothing links to, and clears out what was deleted. This is the
upkeep the open source GBrain project calls dreaming. It repairs and links what
is already there; it does not write new notes of its own.

On Team, the workspace's own AI then reads the pages from the last day that hold
several ideas at once, such as a pasted memo or a long note, and gives each idea
a note of its own in **Concepts**, in the author's words, citing where it came
from. That is why a workspace fed for a month answers better than a new one.

Personal has no overnight pass, because it runs on the workspace's own AI and
every pass costs a model turn. A note saved there is still filed and searchable;
it is linked and tidied when the workspace moves to Team.

## All of it runs at gbrain.io

These pages are about GBrain at gbrain.io, where GBrain runs each workspace's
server, its memory and its upkeep. Nothing here is installed on a laptop. The
open source GBrain project is for running GBrain on a computer you look after
yourself, and its own setup steps (installing it, `gbrain init`, `gbrain serve`)
are not needed for a workspace.

To see what a workspace has scheduled, open its **Schedule** page. To see what an
AI did, open **Activity** in Tools.