Every workspace runs on its own computer in the cloud, with 4 GB of memory. Importing a large batch of documents, or one long job, can ask that computer for more than it has. The operating system then stops the workspace’s AI to protect the rest of the machine, and the workspace restarts on its own a moment later.
Nothing is lost when that happens. The memory the workspace has written is on its disk, not in the memory that ran out, so pages, chats, files and connections are all there when it comes back.
The chat says so
A workspace that is stopped this way says so, in the chat, instead of going quiet and leaving you to guess: it ran out of memory and restarted, so this reply stopped part-way through, and nothing it had already saved was lost. A page that cannot read the workspace says the same thing rather than “couldn’t reach this workspace” (a page that is still setting up after a restart says so first, then this).
You only see that message when GBrain knows memory was the cause, from the computer’s own account of why it stopped. A workspace that stopped for any other reason keeps the ordinary message.
Tell us, and we will look at it
The message carries one button, Tell the team. Pressing it tells us the workspace ran out of memory, when it happened, and what it was working on, so you do not have to explain it and we do not have to guess. The note then says we have it, and someone reads it and emails you back.
We also see it ourselves: a workspace stopped for memory shows up on the same list our team watches for broken workspaces, so you do not have to be the one to notice.
A bigger machine is sometimes the answer, and giving a workspace more memory is something we do. It is not something you have to work out at two in the morning, and it is not a button: it changes what the workspace costs to run, so we decide it with you rather than behind you.
Importing in smaller batches
Whatever happens, importing in smaller batches is the better habit: a few hundred documents at a time, a few sections of a very long document rather than all of it in one go. Anything the workspace files in the meantime stays filed, so stopping and starting an import costs you nothing but the time.
What to do when a workspace isn’t answering
If a page says the workspace couldn’t be reached and it does not say memory, the workspace is probably still starting up. Give it a minute and refresh. If it stays down, its Server page shows whether it is running and has a restart. If the chat itself has stopped answering, a new chat always works.