The streaming-text typewriter built for LLM apps on Compose Multiplatform. Renders a Flow<String> of tokens with live progressive Markdown, syntax-highlighted code blocks that build up as tokens arrive, three speed curves (linear / ease-out / natural), a configurable blinking cursor, tap-to-skip, graceful stop-mid-stream, selectable text, and a screen-reader- friendly live region — on every CMP target.

Pairs naturally with prompt-bar — drop both in and you have a ChatGPT-quality chat UI on Android, iOS, Desktop, and Web in ~20 lines. PromptBar's Send/Stop button auto-syncs with state.isStreaming.

Existing CMP typewriters take a static String — they're animations, not stream renderers. LlmTypewriter is built around the AI chatbot streaming-token use case that nobody else has shipped:

// settings.gradle.kts — already on Maven Central, no extra repos.
dependencies {
    implementation("io.github.nadeemiqbal:llm-typewriter:0.1.0")
}

@Composable
fun ChatBubble(chatViewModel: ChatViewModel) {
    val state = rememberStreamingTypewriterState()
    StreamingTypewriter(
        tokens = chatViewModel.responseFlow,   // Flow<String>
        state = state,
        renderer = rememberMarkdownTypewriterRenderer(),
        cursor = TypewriterCursor.Line,
        speedCurve = SpeedCurve.Natural,
    )
    Button(onClick = { state.stop() }, enabled = state.isStreaming) { Text("Stop") }
}

TypewriterText(
    text = "Build LLM-powered apps for Android, iOS, Desktop, and Web.",
    cursor = TypewriterCursor.Block,
    speedCurve = SpeedCurve.Natural,
)

CyclingTypewriterText(
    phrases = listOf("Type", "Stream", "Render", "Anywhere"),
    holdMs = 1200L,
)

StreamingTypewriter(
    tokens = tokens,
    cursor = TypewriterCursor.Custom {
        Box(Modifier.size(12.dp).background(MaterialTheme.colorScheme.primary, CircleShape))
    },
)

The Markdown parser is prefix-stable: a **bold mid-stream renders as plain text until the closing ** arrives, but every token before that opening ** stays exactly where it was. Code fences highlight progressively as tokens stream in — no waiting for the closing ```.

StreamingTypewriterState is fully usable without composition — handy for tests and for hosts that want to drive the typewriter from a background coroutine:

val state = StreamingTypewriterState()
state.appendToken("Hello, ")
state.appendToken("world!")
state.completeSource()
state.skipToEnd()      // flush buffer
state.stop()           // freeze mid-stream
state.resume()         // pick up where we left off
state.reset()          // start fresh

./gradlew :sample:desktopApp:run                 # Desktop
./gradlew :sample:androidApp:assembleDebug       # Android
./gradlew :sample:webApp:wasmJsBrowserDevelopmentRun   # Web

The sample includes a fake-LLM that streams pre-canned responses with Markdown + code blocks, a speed-curve switcher, and Stop/Resume/Skip controls.

Apache 2.0 — see LICENSE.