Workspaces and Memory Spaces: where things live

Player One keeps your things in two kinds of place, and most confusion about where something went comes from mixing them up.

A workspace is a folder of files: documents, data, the pages of a website, whatever you uploaded this morning and whatever a job produced last night. You browse it, upload to it, download from it and preview things in it, much as you would a folder on a shared drive.

A Memory Space is where knowledge is written down and indexed so it can be found by meaning. It holds what you agreed about discounts on annual contracts, rather than a file named pricing. You rarely put anything there by hand: your assistant writes to it as it learns, and a background pass organises it while you are not working.

Files go in a workspace; what Player One knows goes in a Memory Space. The shared-drive comparison holds right until you want to search: a drive finds filenames, and a workspace does too, but finding by meaning is something you switch on deliberately.

Everyone has one of each

Every agent has its own workspace and its own Memory Space, and both follow that agent wherever it works. The researcher's notes stay the researcher's, and your main assistant's files stay its own. Putting an agent into a project does not tip its things into a shared pile.

A job does not get a place of its own. Its files land in a folder inside the project's workspace, so a finished job leaves its work somewhere you can still find in a month, rather than in a list of one-off spaces. You reach the same folder two ways: the job's own Files tab, or Manage > Files (/workspaces) along with everything else.

Then there are the ones you make. Ask for a workspace for a client, a product or a long-running piece of work and you get a standalone one you can share. Connect it to a project as Use, so agents there can read and search it, or Use and update, so they can write to it as well. A Memory Space you create for a topic rather than for an agent works the same way. Sharing is set on the workspace's own Projects tab, or on Manage Project > Knowledge > Workspaces (/project/knowledge/workspaces).

Where a thing actually lands

Drop a file into the chat and it goes into the workspace of the agent you are talking to, filed under today's date. The card in your message opens it. Nothing is trapped in the conversation: ask about that file a week later and the agent can still read it.

Work an agent does inside a project lands in the project's workspace, not in its own, because the work belongs with the project rather than with whichever agent happened to do it. Outside a project, an agent works in its own space.

When you are unsure, ask. Agents name the workspace they wrote to rather than reciting a path, and they can open the file browser in the preview pane at the exact spot.

A few files in an agent's workspace are read-only to that agent: its own instructions and identity. You can still edit them yourself on the page. It guards against an agent tidying away the thing that tells it who it is.

Cloud drives attach to a workspace

Connect a drive by asking for it, or with Connect storage on Manage > Files (/workspaces); either way you sign in through your browser, then choose which folders you want. Each folder attaches to one workspace with a direction: two-way, download only for reference material you never change, or upload only for output you want published to the drive. After that it is an ordinary folder inside that workspace, and anything an agent writes there appears on your drive.

Google Drive, Dropbox, OneDrive, Box and SharePoint are the usual choices, and which of them you are offered depends on how your installation was set up.

Two things surprise people. Detaching a folder stops future syncing and deletes nothing, in the workspace or on the drive, so it is safe to do when a folder gets noisy. And a code repository does not travel through a sync: the tracked files copy across, but the repository's own hidden bookkeeping never syncs, because a two-way sync would corrupt it rather than copy it. If you want a repository on another machine, clone it there.

Finding by meaning is a choice

An agent being able to read a workspace does not make it searchable by meaning. The two are separate on purpose: indexing a folder of throwaway drafts costs something and buys little.

The rule of thumb is that durable knowledge belongs in a Memory Space, which looks after its own index and its own guidance on what belongs there. A workspace of files can be indexed too, on its Knowledge tab, and that is the right answer when you have a pile of source documents you want an agent to answer from. Anything you want the assistant to simply know, rather than to look up in a named folder, is better saved as memory.

A Memory Space and an indexed workspace both report what has been indexed and which files were skipped. If an agent cannot find something you can plainly see, that report is where to look.

Keeping it tidy

Pin the workspaces you return to and they stay at the top of the list. Delete the ones you have finished with, although Player One will not let you delete a project's default workspace or an agent's own, because too much depends on them.

Both lists follow the project selected in the header, so something missing is more often out of scope than lost. Switch project, or switch to all projects, before concluding anything has gone. Neither page has its own search box: the box in the top bar filters whatever page you are on, and it says which page it is searching.

One habit saves most of the mess: one workspace per real thing, such as a client, a product or a site. Resist making a new one for every task, because the tenth half-used workspace is worse than a slightly crowded one. Memory Spaces want the opposite instinct: a few broad ones you keep feeding beat a dozen narrow ones nobody remembers to search.

Next: Memory, explained covers what a Memory Space actually does, Organising with projects explains the scoping, and Workspaces and files, Memory and Jobs describe the parts. The steps for connecting a cloud drive and building a knowledge base are in the guides.