// why bramo
A terminal that respects how you already work.
You live in the terminal all day. Most new terminals ask you to change something in return: an account, a new shell, telemetry, a single platform. These are the rules Bramo follows instead.
// 01
A plain terminal is always underneath.
Bramo uses the terminal engine from Ghostty, libghostty-vt. Programs see a standard terminal, so vim, Neovim, tmux, htop, less and ssh behave as they do anywhere else. Blocks, the command editor and the panels are drawn around the terminal grid, never inside it.
Every new behavior can be turned off. Hand the keyboard back to your shell with one switch, or run without shell integration and get a classic scrollback.

// 02
Your terminal is yours.
There is no account and no login. The app sends nothing about what you type or see, and it has no telemetry by default. It has no AI built in either: you bring the agents you want, and Bramo works with them.
Your settings are a text file. Your sessions run on your machine, in a service you can stop with one command.
bramo.tomlPlain text you can keep in your dotfiles.// 03
It behaves like a desktop app.
Select text with the mouse and copy it without padding spaces. Undo in the command editor. Click a file to read it. Find any action in the palette. These are things every other app on your computer does, and a terminal should too.
The keyboard reaches everything. Every panel and action has a shortcut or a palette entry, and focus is always visible. Shortcuts follow each system's conventions, so Windows and Linux don't get Mac shortcuts with the keys renamed.
// 04
Your sessions outlive the window.
Closing a terminal window shouldn't kill a build, a server or an agent halfway through. In Bramo, shells run in bramod, a background service that used about 1 MB of memory in our tests. Quit the app or update it, and the programs keep running. Open it again and you're back where you were, blocks included.
After a restart of the computer, every split comes back in its place with its earlier output, so you can see what you were doing.
// 05
Windows and Linux are not an afterthought.
About half of professional developers use Windows (Stack Overflow Developer Survey 2025), and many use Linux, yet new terminals often launch on the Mac only. Bramo builds Windows, Linux and macOS from the same code, and its test runs open the real app on all three. Menus, dialogs and shortcuts follow each system.
Phone apps to check on your sessions come after the desktop release.
// 06
Speed has a budget, and we measure it.
No interaction in Bramo should take more than 100 ms, and memory and startup have budgets too. A benchmark tool launches the real app, times each action from the input to the frame that shows the result, and fails when a number goes over. It only counts runs on physical machines, never virtual ones.
| Action | Budget | Measured1 |
|---|---|---|
| Open the palette | 100 ms | 29 ms |
| Switch sessions | 100 ms | 17 ms |
| Show the search bar / first match in 100,000 lines | 100 ms | 13 ms / 32 ms |
| Show a new split | 100 ms | 15 ms |
| Files panel on a 10,000-file folder | 100 ms | 19 ms |
| Memory, one idle session | 90 MB | 72 MB |
| Memory, four sessions and the files panel | 90 MB | 75 MB |
| Startup to the first prompt | 300 ms | 584 ms2 |
Get an invite to the alpha.
Invites go out in waves. People whose setup matches what we are building first go first.
We use your email for the invite and a few product updates. Privacy note.
- Machine: MacBook Pro 2019, Intel Core i7-9750H, macOS 26.7.1, external 1080p display at 60 Hz. Measured on 24 October 2026. Each figure is the median of 3 runs, and other programs were running at the time, so the numbers are approximate. ↩
- Startup misses its budget: 584 ms from launch to the first prompt for a new session running
zsh -f, against 300 ms. About 270 ms of that passes before any Bramo code runs, while the UI toolkit starts and macOS sets up text input, and the first frame then waits on macOS; Bramo's own work takes about 20 ms. The number has not been measured again since IBM Plex Mono was bundled with the app; the check for that change found startup within run-to-run noise. ↩


