Start with one personal workspace, not one per AI. Connect the AIs to that same
memory, choose one to save notes, and let the others read. Personal and work can
use the same GBrain sign-in and the same AI connection, but need separate
workspaces in separate organizations if different people should read them.
One shared workspace has no private personal corner.

These pages describe hosted GBrain, not an installation of the open-source
project on a computer someone maintains themselves; hosted plans and connection
steps are not self-hosting instructions.

## One workspace, with its readers checked first

1. **Start or open a workspace.** [Personal](/personal) keeps memory for an AI
   such as Claude or ChatGPT. Personal has no chat or Schedule page of its own:
   do the steps below in the AI that connects to it. If a workspace already
   holds the material, use that one rather than starting empty again.
2. **Check the organization before saving anything private.** Open
   **Organizations**, the workspace's organization, then **Members**. For a
   personal workspace, that list should contain only the person using it.
   Check **Settings**, then **Data protection**, for other ways people reach it.
   Every organization member can read every workspace in it. Buying Personal
   can reuse an existing workspace or an organization already administered by
   the buyer; the plan name does not make it private. If the organization is
   shared, stop and [ask for setup help](/support) before adding personal
   material. [The organization boundary](/docs/workspace/sensitive-teams)
   explains the separation.
3. **Connect one AI as the writer.** Follow [Connect an AI](/docs/tools/assistants)
   from **AI & Agents** under **Integrations**. Set its **Memory** to
   **Read & Write**. Connect the others to the same workspace and set their
   Memory to **Read**. One writer is a starting recommendation, not a restriction
   GBrain imposes: several AIs can be allowed to write.

## Two lines in every AI's standing instructions

Replace `Personal` with the workspace's actual name. Paste the first line into
every AI's standing instructions, and the second into the chosen writer's.
[Connect an AI](/docs/tools/assistants) lists where each AI keeps those
instructions.

```text
At the start of each session and before answering about me or my work, search the GBrain workspace named "Personal" and cite the notes used; do not substitute another workspace.
Save durable facts, decisions and corrections to the GBrain workspace named "Personal" as we work; check existing notes first, update rather than duplicate, and do not copy anything to a work or family workspace unless I explicitly ask.
```

The instructions ask the AI to use memory; they do not grant permission or
guarantee that the AI saves every conversation. GBrain's Memory setting decides
whether writing is allowed. A reader needs only the first line.

## A fresh conversation proves the save

In the writer, ask: *“Save this harmless test fact to the GBrain workspace named
Personal: my setup-test colour is violet. Show the saved note.”* Use the actual
workspace name here too.

Open a **new conversation**, preferably in one of the read-only AIs, and ask:
*“Search the GBrain workspace named Personal for my setup-test colour. Open the
saved note and cite it.”* The test passes when the answer comes from that note,
not from the original conversation. Open **Memory** to inspect the note, and
[Activity](/docs/tools/activity) to check that the write and search happened.

If saving is refused, check the writer's Memory level on its GBrain connection
page. If search is empty, confirm the workspace name and open the saved note
before trying again after indexing. An answer with no saved source is not proof:
ask the AI to search and open the note rather than guess.

Once the test works, add a small amount of useful material through the writer.
For a separate shared memory, follow
[A workspace for a family or company](/docs/workspace/family-or-company).