@textui/terminal

What @textui/core needs to reach an actual terminal: adapters, capability detection, ANSI writing and input decoding.

npm install @textui/terminal

import { createApp } from '@textui/core';
import { createNodeTerminal } from '@textui/terminal';

const app = createApp({ terminal: createNodeTerminal(), root: <App /> });
await app.start();

Two adapters

createNodeTerminal() is the real one - stdin, stdout, resize, and the signal handling that puts the screen back the way it found it when the process dies.

createVirtualTerminal() has no tty at all. It is what the test harness and --static run on, and it is the reason a component that only works interactively is a component nobody can test cheaply.

Capabilities are detected, and can be lied to

Unicode level, colour depth, mouse, hyperlinks, the kitty keyboard protocol - detection is right almost always, and the times it is not are the times nobody is watching. So every capability can be forced, which is how you see the terminal somebody else is sitting at:

createNodeTerminal({ capabilities: { unicode: 'ascii', colors: 4 } });

Under the kitty protocol, ctrl+shift+f is a distinct stroke from ctrl+f. Without it the terminal sends the same byte for both, and no amount of decoding recovers the difference.

A frame you can keep

The screen is the output, so the next redraw destroys the frame that was wrong. Two functions write one out instead - both from the buffer the runtime last painted, neither needing a tty.

captureBuffer gives a frame a terminal can replay: every cell in order, no cursor control, so it can go in a file or an issue. colors: false strips the colour, which is the copy a diff can read.

bufferToSvg gives one a repository page can show, because an .ans file is only a screenshot on a terminal:

const svg = bufferToSvg(app.buffer(), {
  background: app.theme.colors.canvas,
  foreground: app.theme.colors.text,
});

Self-contained - no font, no stylesheet, no script, nothing fetched - which is what survives GitHub's image proxy. And it is text, so a committed screenshot diffs and CI can check it instead of somebody re-taking it.

What it needs from a stream

TerminalInput and TerminalOutput, which say what the adapter uses rather than which stream it is - setRawMode, columns, write, and half a dozen more. process.stdin and process.stdout satisfy them, so nothing changes for a caller, and so do the streams of runtimes that are not node.

That is also why the published types name no Node interfaces: a package with no dependencies should not need @types/node installed to compile.

Runtime

No dependencies beyond @textui/core. Uses process for stdio and signals, and no other Node API. Node 22+ and Bun.

Documentation

https://softov.github.io/textui/

Part of TextUI - documentation - getting started

@textui/kit one install · @textui/core the runtime · @textui/widgets the catalog · @textui/terminal adapters and input · @textui/testing the harness · @textui/cli the CLI