Skip to main content
An open wooden treasure chest on the floor is overflowing with a brain model, index cards, a claw hammer, a bench plane, a games controller, and a recipe box, with a controller, card, and carpenter’s pencil spilling out in front, and dark blue iron clasps as the accent colour.

A serious amount of machine, for one team

A memory that outlasts any single AI, accounts it can act on, skills that run on a clock, and a server of the team's own. Here is what using it looks like.

Memory

What it knows about the team and the work, in files the team owns.

One memory shared by ChatGPT, Claude, Claude Code and Cursor

Connect any AI that supports MCP, the standard connection, to the same notes. One memory for all.

See how it works

Read-only or full access to the memory, per AI

Read can search and recall. Full can also remember, link and forget. Set it for each AI.

See how it works

Search that finds notes by meaning as well as words

Ask about "the pricing call" and it finds the note titled "Q2 renewal terms."

See how it works

Browse the memory by day and by person

Lately shows what it learned this week. People shows everything it knows about one person.

See how it works

Tools

The accounts it reaches, and what each AI may do with them.

The same tools in every AI that speaks MCP

One address, pasted once into ChatGPT, Claude, Claude Code or Cursor. Each gets the same accounts and rules.

See how it works

Gmail and Calendar from the command line, for people and coding agents

Gmail, Calendar and web search from a terminal. Claude Code and Codex run the same commands.

See how it works

Passwords and API keys never reach the AI

Connected accounts are held by gbrain. An AI gets the result of a call, never the key.

See how it works

A permission level for every application an AI can reach

Gmail has four levels, Calendar two, web search is on or off. Set per application; every AI follows.

See how it works

A log of every call an AI makes with connected accounts

One row per call. Who asked, which AI, which account, what it did, whether it was allowed.

See how it works

A kill switch for every AI, and an expiry date

Turn an AI off and its next call is refused. Every AI's access also has an end date.

See how it works

Skills

The jobs it knows how to run, on demand or on a schedule.

A directory of jobs gbrain already knows how to do

Every job on the shelf, with what each one does and needs. Read one, turn it on.

See how it works

Skills you switch on, and switch off again

One switch per skill on the workspace. On, it runs for everyone; off, it stops.

See how it works

Work that runs on a schedule, and reports back

Give a skill a time. It runs, writes what it found to the memory, and tells the thread.

See how it works

Workspace

The room a team works in, and the server it runs on.

One conversation the whole team is in, with the assistant

Everyone in the thread, everyone's messages, one assistant answering all of them.

See how it works

Watch the assistant work, and steer the answer before the end

The steps appear as it takes them. Add a note mid-answer and it takes the note.

See how it works

The workspace keeps working with the laptops closed

Hand it a job at night. Open the thread in the morning and the work is there.

See how it works

A private server per workspace, kept up to date by gbrain.io

The memory and skills run on a machine of the workspace's own, wrapped and kept current by gbrain.io.

See how it works

Sign in with Google, or an email link, and no passwords

The whole team gets in with the accounts they already have. There is nothing to reset.

See how it works

Run the workspace on the model you prefer, from Anthropic or OpenAI

Pick the model a workspace thinks with, and change it later. Everything else stays.

See how it works

Spin up a gbrain

All of it, in a new workspace, in about a minute.