Skip to main content
gbrain
Sign in

Your memory should outlive the assistant.

Every assistant now offers to remember you, and each one remembers you only to itself. What you explained to one on Tuesday is gone when you open another on Friday, and all of it is gone the day you switch. Keep it somewhere else: one memory you own, that all of them read and write.

What is wrong with the memory you already have

Nothing, while you use one assistant, forever, alone. That is not what anybody does.

It does not travel

The assistant in your editor cannot read what the one in your browser learned. You explain the same project twice, then a third time next month.

You cannot read it

A summary of what it thinks it knows about you is not the same as the record. You cannot correct a line, and you cannot tell why it answered the way it did.

It leaves when you do

Switching tools costs you every accumulated thing the old one knew. That is a lock-in bill, paid in the one asset that took you a year to build.

Written once. Read by all of them.

Your assistants stop being separate strangers who each know a third of the story.

One note, and everything you use can see it

You tell one assistant how your team ships. Every other assistant you use knows it from then on, because they are all reading the same memory instead of each keeping a private guess.

It works the other way too. A correction made anywhere is a correction everywhere, so you fix a wrong assumption once rather than in four places you will not remember to visit.

Brad's MacBook On 4m ago
Claude Code On 22m ago
Cursor Not set up Never
Hermes Mac Mini Off 6d ago
The clients hooked up to one person: a terminal, an editor and an assistant, all reading the same memory.

It is a folder of files

Not rows in a system you cannot reach. Plain markdown, one thing per file, that you can read, correct, diff and take away.

Read it, fix it, keep it

The memory a tool will not show you is a memory you cannot correct. This one is a directory.

One level of a brain's memory: the folders, then the notes, each a file you can open.
brad@macbook · gbrainio 
ls ~/memoryhow-we-ship.md          prefers-short-answers.mdpricing-decisions.md    q3-goals.mdteam-and-roles.md       tools-we-pay-for.mdcat ~/memory/prefers-short-answers.md---name: prefers-short-answersdescription: Answer in a paragraph, not a report---Lead with the verdict. Hold the detail until asked.cp -r ~/memory ~/Dropbox/memory-backup
One file per thing, named after the thing. Nothing is hidden from you and nothing needs an export request.

Why it should not come from the assistant

An assistant's memory is built to make that assistant harder to leave. Yours should be built to make leaving free.

Neutral

It belongs to no model vendor, so it has no reason to keep what it knows away from a competitor you would rather use.

Legible

You can read every line of it. A record you cannot inspect is not a memory, it is a rumour.

Portable

Files, on disk, in a format that outlives all of this. Copy the folder and you have moved.

Start it in one line

The open source memory is free and yours. A workspace is what adds the team, the schedule and the machine it runs on. 

# Install the command line
curl -sSL https://gbrainio.terminalwire.sh | bash

# Point your assistants at your memory
gbrainio login

The installer is a shell script. Read it first. macOS and Linux.

Keep what you have explained

You have already told an assistant most of what it needs to know. Start writing it somewhere it will still be next year.

Sign in See where a team's memory lives