A workspace does the most when an AI can save to it and some of its work runs
without being asked. These five steps get both done in about fifteen minutes.

## 1. Connect the AI with Read & Write

Open the workspace, then **AI & Agents** under **Integrations**, and pick the AI:
ChatGPT, Claude, Claude Code, Codex, Cursor or another. Point it at
`https://gbrain.io/mcp`. When GBrain asks what the AI may do in the workspace,
leave **Memory** and **Schedules** at **Read & Write**, where they start. Codex
and Cursor take a key instead of signing in, and the same page asks for one
level for both: pick **Read & Write**.

At Read, an AI can search the memory but every save it tries is refused, so an
AI that files notes from mail, documents or chats needs Read & Write. It still
cannot raise its own access. [Connect an AI](/docs/tools/assistants) has the
steps for each AI, and [What a connected AI can
reach](/docs/tools/what-an-ai-can-reach) has what each level allows.

## 2. Paste the standing lines

GBrain tells the AI about the memory each time it connects, but an AI may not
act on that unless a question is plainly about what is remembered. Paste these
lines into the instructions the AI reads in every chat:

```
At the start of every session, and whenever I mention another chat, check GBrain's memory first.
Save decisions to GBrain as they are made.
Save the work facts, decisions, and people and project context my team needs.
Don't save health, family, personal finances, private email or other personal matters to GBrain unless I ask.
GBrain hosts my workspace at gbrain.io, so skip any steps for installing or running GBrain myself.
```

Once the AI is connected, its page under **AI & Agents** has the same lines
under **Make GBrain a habit**, with a Copy button and where that AI keeps them:
`~/.codex/AGENTS.md` for Codex, `~/.claude/CLAUDE.md` for Claude Code,
Settings for ChatGPT and Claude.

## 3. Open Skills and read the first ten

Open [Skills](/skills). Each skill is a finished job: a morning briefing, an
inbox worked through, a person researched before a meeting. Read the first ten
and pick the two or three that match work someone does by hand today.

## 4. Set up the ones that fit

On a skill's page, press its **Set up** button, then pick the workspace. On a
workspace with its own AI, a chat opens for the skill and asks what it needs
to know, such as what to watch and who it is for. On Personal, the page lists
the connectors the skill reads and the words to say in the connected AI.

## 5. Put them on a schedule

A skill that runs every morning does its work before anyone asks. On a
workspace with its own AI, say yes when the setup chat offers to run it on a
schedule, and say when. Then open **Schedule** to see the job, when it last
ran and when it runs next. On Personal the workspace runs nothing on its own:
ask the connected AI to run the skill as one of its own scheduled tasks.
[Skills and scheduled work](/docs/skills) has more on both.

Check back the next day. **Memory** shows what was filed, and
[Activity](/docs/tools/activity) shows each save and search the AI made.
[Where a connected AI files notes](/docs/workspace/where-memory-goes) says
which folder each note lands in.