Skip to content

Halogen

LLM-generated Material 3 themes for Compose Multiplatform

On-device with Gemini Nano on Android. Cloud-powered everywhere else. Just a Gradle dependency.

Quick Start

val halogen = Halogen.Builder()
    .provider(GeminiNanoProvider())
    .cache(HalogenCache.memory())
    .build()

halogen.resolve(key = "coffee", hint = "warm coffee shop vibes")

HalogenTheme { App() }

That's it. Halogen generates a full Material 3 theme from a natural language prompt - colors, typography, and shapes - cached locally so the LLM is called once per key, ever.

Features

  • Pluggable LLM providers - Ship with Gemini Nano on-device, or use OpenAI/Claude for cloud, or write your own. Set one provider per engine instance.
  • Keyed contextual themes - Any string maps to a cached theme. A subreddit, a route, a category, a brand - each gets its own generated look and feel.
  • Full Material 3 coverage - 49 color roles, 15 typography styles, 5 shape sizes. The LLM generates 6 seeds + hints; the library expands them using HCT tonal palettes.
  • In-memory caching - LRU cache with configurable max entries. Themes resolve in nanoseconds after first generation. The LLM is never called twice for the same key. Optional persistent caching via halogen-cache-room (backed by Room KMP) survives process death and app restarts.
  • Kotlin Multiplatform - Android, iOS, Desktop (JVM), and Web (WasmJs). Gemini Nano is Android-only; all other platforms use cloud providers.
  • WCAG AA contrast validation - Generated color schemes are validated for accessibility. Primary/secondary/tertiary contrast against their on-color counterparts meets minimum ratios.

Platform Support

Platform LLM Provider Cache Room Cache Compose UI
Android Gemini Nano (on-device) or cloud Memory (LRU) Yes (optional) Material 3
iOS Cloud providers Memory (LRU) Yes (optional) Material 3
Desktop (JVM) Cloud providers Memory (LRU) Yes (optional) Material 3
Web (WasmJs) Cloud providers Memory (LRU) / localStorage -- Material 3

Artifacts

// Core primitives — no LLM dependencies
implementation("me.mmckenna.halogen:halogen-core:0.1.0")

// Compose Multiplatform UI components
implementation("me.mmckenna.halogen:halogen-compose:0.1.0")

// Engine — caching, orchestration, prompt construction
implementation("me.mmckenna.halogen:halogen-engine:0.1.0")

// Gemini Nano provider (Android only)
implementation("me.mmckenna.halogen:halogen-provider-nano:0.1.0")

// Optional: Persistent theme cache backed by Room KMP (survives app restarts)
implementation("me.mmckenna.halogen:halogen-cache-room:0.1.0")

Learn More