Vega SDK

Vega OS is Amazon's new TV operating system (Fire TV's next-gen runtime, distinct from Fire OS which is Android-based). It uses the Kepler runtime and its own SDK. The template's apps/vega/ is a separate app that shares UI via packages/shared-ui but builds on a different toolchain.

What Vega is (and isn't)

• Fire OS (legacy): Android-based. Apps built like Android TV apps. Uses apps/expo-multi-tv/.
• Vega OS (newer): Not Android. Custom runtime called Kepler. JavaScript-based. Uses apps/vega/.

Both are "Fire TV" in the marketing. The difference is real. A Fire OS APK does not run on Vega. A Vega bundle does not run on Fire OS.

If the AppSpec targets Fire TV in general, target both by default. Drop one only if the user explicitly excludes it.

apps/vega/ layout

apps/vega/
├── src/
│   ├── App.tsx              # Vega entry
│   ├── screens/             # Vega-specific screens (if any)
│   └── ...
├── package.json             # Has @amazon-devices/* deps
├── vega.config.js           # Vega bundler config
└── assets/                  # Vega-specific assets

It imports from @multi-tv/shared-ui like the Expo app does. The Metro .kepler.ts resolver picks Vega-specific files automatically.

Key @amazon-devices/* packages

These are the Vega-specific ones the agent will encounter:

• @amazon-devices/kepler-app — runtime, lifecycle, system events
• @amazon-devices/kepler-ui-components — native Vega UI primitives (use these for Vega-only screens for best perf)
• @amazon-devices/kepler-media — media playback APIs (use instead of react-native-video on Vega)
• @amazon-devices/kepler-pointer — input events, including remote
• @amazon-devices/kepler-navigation — system back, exit handling

The shared UI catalog's <Player> has a .kepler.ts override that swaps in @amazon-devices/kepler-media. Don't bypass it.

What ports cleanly from Expo app → Vega

• React + JSX. Same.
• React Navigation. Works.
• React TV Space Navigation. Works.
• Styling via the theme system. Works.
• packages/shared-ui components. Mostly works; .kepler.ts overrides exist where they don't.

What does NOT port

• react-native-video → use @amazon-devices/kepler-media.
• expo-font → fonts are loaded via Vega manifest, not at runtime.
• expo-image → use Vega's image primitive or plain <Image>.
• Any package that depends on Android Native Modules or iOS Native Modules. These will install but fail at runtime. Audit package.json deps before adding.
• Background tasks / push notifications differ — out of scope for v1.

If you add a new dependency to packages/shared-ui (which both apps consume), check it works on Vega. The safe rule: pure-JS packages port; native-bridged packages probably don't.

Build commands

# From repo root
yarn workspace vega install

# Dev build (deploys to connected Vega device or sim)
yarn workspace vega dev

# Production bundle
yarn workspace vega build

# Bundle artifact ends up in apps/vega/dist/

The vega_build tool wraps these. The agent should not invoke Vega builds in parallel with Expo builds — both compete for emulator/device handles and the failure modes are confusing.

Vega manifest essentials

apps/vega/vega.json (or equivalent — check current template) declares:

• App ID and version (mirror the Expo app's bundle ID for store consistency)
• Required capabilities (media playback, network)
• Entry point
• Icon and banner assets (Vega has its own size requirements — typically the same 400×240 banner works)

When customize_app_metadata runs, it must update both apps/expo-multi-tv/app.json and apps/vega/vega.json. Keeping them in sync is the agent's responsibility.

Common Vega failure modes

Decision: should I add a Vega-specific override file?

Add .kepler.ts only if:

1. The default file imports a non-portable package.
2. The behavior must differ on Vega (UI guideline, system integration).

Don't add it for:

• Cosmetic tweaks. Use theme tokens with .kepler.ts overrides instead.
• "Just in case." Each override doubles maintenance.

Testing on Vega

Virtual Device (VDA)

The Vega Virtual Device Agent is the primary testing target:

npx kepler device start       # Start VDA
npx kepler device list        # Verify it's running
npx kepler install            # Deploy the built app
npx kepler launch             # Start the app
npx kepler screenshot <path>  # Capture screen
npx kepler key right          # Send D-pad right
npx kepler key down           # Send D-pad down
npx kepler key center         # Send select/enter
npx kepler key back           # Send back
npx kepler stop               # Stop the app

Don't ship to a real device on first run. VDA catches common issues faster.

MCP-based diagnostics

The @amazon-devices/amazon-devices-buildertools-mcp package provides additional diagnostic tools:

• analyze_perfetto_traces for app launch time analysis (TTFF, TTFD)
• get_app_hot_functions for CPU profiling
• symbolicate_acr for crash report analysis
• search_documentation for Vega-specific guidance

Appium (for advanced automation)

Vega supports Appium via the Vega Appium Driver for more complex test scenarios. Use locator strategies specific to Vega. For the QA loop, the Kepler CLI commands above are sufficient.

Vega log access

npx kepler logs              # Stream app logs (Vega equivalent of logcat)
npx kepler logs --filter=error  # Filter for errors only

Anti-patterns

• Forking packages/shared-ui for Vega. Defeats the monorepo. Use .kepler.ts overrides on specific files.
• Conditional logic if (Platform.OS === 'kepler') in shared files. Use platform extensions; the bundler is the routing layer.
• Skipping Vega in v1 "to ship faster". If the user wanted Fire TV reach, half is not the answer. Either include Vega or drop Fire TV entirely.
• Treating apps/vega/ like a copy of apps/expo-multi-tv/. It's a different runtime. Its entry, lifecycle, and APIs differ.