Private beta — invite only

AI is powerful. Chat is the bottleneck.

Select what you're working on, choose what you want AI to do, and keep moving — without rebuilding the context every time.

Early beta. No spam.

The interface

Your workspace should follow your reasoning, not the apps you use.

One command on the mouse and GalaxyBrain takes the whole screen — translucent. You still see the page you were on, and you can still point at it.

Summon

One command on the mouse, from whatever app you are in.

See through

Your screen stays visible underneath. You never lose your place.

Act directly

Point at what is on screen instead of describing it in words.

Everything below happens here
act on any selectionuse your own reusable actionscombine selected context

One translucent screen you call up from wherever you are — and everything below runs inside it.

How it works
1
Select

Highlight anything, wherever you already work.

2
Choose an action

Explain, Summarize, Rewrite, Ask — or one of your own.

3
Keep working

The answer lands in place. No tab switch, no re-explaining.

Your thinking isn't linear. Why should your AI workflow be?

Today
SelectCopyChatPastePromptAnswerCopyReturn
With GalaxyBrain
SelectAct
If you've caught yourself thinking any of these, you're who we're building for.

“Why am I the one carrying context between my tools?”

“Why is my work cut up along app borders instead of along how I think?”

“I already worked out how I want this done. Why do I have to explain it again?”

Your actions, reusablein the overlay

Write the prompt once. Keep it as a command.

Your best prompts live in a note you copy-paste from. Here you never save one: the moment a prompt runs, it becomes a command — same instruction, with an empty slot where the thing you pointed at used to be.

What a command is made of
rewrite?in the voice of?
The instruction

Kept word for word, exactly as you typed it the first time it worked.

A slot per thing you pointed at

What you selected is dropped and left empty, so the command fits the next thing instead of the last one.

A way to fire it

From the overlay, from the menu that appears on a selection, or from a shortcut you assign.

1You just run a prompt
rewritethis paragraphin the voice ofmy style guide
2The nouns are lifted out
rewrite?in the voice of?

The verb stays, the nouns go. Nothing to configure.

3It is waiting for you next time
rewrite ? in the voice of ?tighten ? to three lines
One slot
It stays armed

Tap the command, then keep tapping things down the page. Each one gets it. Esc when you are done.

Two slots
They fill in order
rewrite ? in the voice of ?rewrite this intro in the voice of ?

The command shows what you have filled so far. The last tap runs it — you never press send.

Already selected
Then just tap it

Point first, tap the command second. It runs once and does not stay armed — no mode to escape from.

You already worked out how you want AI to do this. The point is not to make you explain it again.

When one selection isn't enoughin the overlay

Point at two things. Then keep the move.

Highlight A, type what you want done, highlight B, send. GalaxyBrain saves the request as an action with empty slots — next time you tap it and just point at the next two things.

1Highlight A
A
2Say what to do, point at B
compareAthe paperwithBmy note
3It stays as an action
compare ? with ?rewrite ? in the voice of ?

None of this is invented from scratch. Pointing at objects instead of describing them, and keeping a prompt that worked as a reusable command, come out of published HCI research.

DirectGPT — Masson et al., CHI 2024

Help shape GalaxyBrain.

We're looking for the first people who feel this problem every day.

Early beta. No spam.