App Actions for Google Assistant: Development and Publishing

User says "open order #123" — your app should open the order card. Without App Actions, the Assistant will simply suggest opening the app, and the user will have to navigate manually. This is a loss of time and UX: in 90% of cases, users abandon the app if they don't find the needed screen within three taps. We design App Actions so voice commands work on the first invocation. In 3–5 days turnkey: from configuring shortcuts.xml to publishing in Actions Console and testing on a real device. Compared to manual navigation, App Actions reduces time to reach a screen by 50% and improves user retention. Typical project cost ranges from $1,500 to $5,000 depending on complexity. Get a free project assessment — contact us.

Google Developers: "App Actions let users invoke your app's features using the Google Assistant."

App Actions is a mechanism through which an Android app registers supported voice commands in the Google Assistant ecosystem. The user says "start workout in [App]" — Assistant launches the exact screen directly, without manual search. Developing App Actions means creating a capabilities configuration, mapping NLU parameters, and handling intents inside the app.

Common Issues We Resolve

Three frequent problems:

• NLU fails to recognize the command — before configuring inventory, recognition accuracy is about 65%; after implementing inline inventory, it reaches 95%.
• Intent not handled on cold start — due to incorrect Activity launch mode. In 20% of projects, this causes a crash.
• Publication is rejected due to non-compliance with App Store Review Guidelines (Section 4.2) — requires correct deep link and parameter handling.

Our Configuration Approach

The central file is shortcuts.xml in res/xml/. It contains <capability> elements, each describing one supported Built-in Intent (BII). The file is linked to the manifest and uploaded to Actions Console upon publication.

Capability structure:

<shortcuts xmlns:android="http://schemas.android.com/apk/res/android">
  <capability android:name="actions.intent.START_EXERCISE">
    <intent
      android:targetPackage="com.example.fitness"
      android:targetClass="com.example.fitness.WorkoutActivity"
      android:action="START_EXERCISE">
      <parameter
        android:name="exercise.name"
        android:key="exercise_type"
        android:mimeType="text/plain" />
    </intent>
    <slice-presentation
      android:shortcutId="start_running"
      android:title="Start run" />
  </capability>
</shortcuts>

exercise.name is the BII parameter that Google's NLU engine extracts from the voice command. If the user says "start a run", the value will be running. The Activity receives it via intent.getStringExtra("exercise_type").

Which Fulfillment Type to Choose

Deep link is 2x more effective when integrating with the web — we choose it in 80% of projects.

Inline Inventory for Better Recognition

For BIIs that accept specific values from a closed set (e.g., list of tracks or workout categories), you can declare a <shortcut> bound to the capability. Assistant suggests these values as hints and confirms selection before launching:

<shortcut android:shortcutId="yoga_workout" android:shortcutShortLabel="@string/yoga">
  <capability-binding android:key="actions.intent.START_EXERCISE">
    <parameter-binding
      android:key="exercise.name"
      android:value="@array/yoga_synonyms" />
  </capability-binding>
</shortcut>

yoga_synonyms is a string-array with variants: "йога", "yoga", "stretching". NLU matches them to the yoga_workout shortcut.

List of popular BIIs:

• actions.intent.START_EXERCISE
• actions.intent.OPEN_ORDER
• actions.intent.SEND_MESSAGE
• actions.intent.PLAY_MUSIC
• actions.intent.GET_RESERVATION

Testing Without Publishing

1. Install gactions CLI.
2. Run gactions deploy preview --action_package app_actions.xml.
3. On a test device (with the same Google account) invoke the command in Assistant.
4. Verify that the Intent launches the correct screen.

This method saves up to $2000 on cloud debugging.

Handling Intents in Your App

The Activity declared as targetClass must correctly handle the Intent in all launch modes: cold start, return from background, transition from another screen. Parameters that NLU couldn't extract will come as null — the app should show a selection screen instead of crashing.

Analytics: each App Actions launch should be logged with parameters to understand which BIIs are actually used. Firebase Analytics or Amplitude, custom event assistant_app_action with parameter bii_name.

What's Included in Our Work

We develop App Actions turnkey. As a result, you get:

• Configuration of shortcuts.xml with required BIIs and inline inventory.
• Integration of intent handling into the app code.
• Testing via Actions Console and a real device.
• Publication to Production (if needed).
• Documentation for maintenance and analytics setup.

Contact us to discuss your project. We'll assess the scope and provide timelines. Get a free consultation on inventory configuration — no charge.

Process and Timelines

Our typical workflow:

1. Requirement analysis — gather voice commands and target user flows.
2. App audit — review existing deep links, activities, and manifest setup.
3. Design — map each command to a BII, create shortcuts.xml, define inventory.
4. Estimation — provide fixed timeline and cost (based on complexity).
5. Development — integrate code, configure intents, handle edge cases.
6. Testing — preview with gactions CLI, full test on physical devices.
7. Deployment — publish to Actions Console and Google Play.

From start to live deployment, typical projects take 3–5 business days. Larger integrations may require up to 2 weeks. Contact us for a precise assessment.

Typical Mistakes to Avoid

• Missing null checks — NLU may return null for optional parameters. Always handle gracefully.
• Wrong launch mode — using singleTop or singleTask incorrectly can cause crashes on cold start. Prefer standard unless deep links require otherwise.
• No analytics — without logging, you cannot measure which commands users actually use.
• Skipping preview testing — many issues are caught only on real devices. Always run gactions deploy preview.
• Ignoring deprecated BIIs — Google occasionally deprecates intents. Stay updated via Google Developers.

Why Trust Us

Our experience: 5+ years of Android development, 30+ projects with voice integrations. We guarantee stable App Actions across different Android versions and Assistant versions. We provide post-deployment support and can add new commands as needed.

Ready to implement voice commands in your app? Let's talk. Get in touch for a free consultation.

Why is native Android development with Kotlin the production standard?

RecyclerView with DiffUtil.calculateDiff() on main thread, a list of 500 items, an average older Android phone – the user gets 200–400 ms freezes on every data update. Move the diff calculation to a background thread via AsyncListDiffer – the problem disappears. These things aren't obvious without a profiler and understanding Android’s threading model. According to Wikipedia (Android development), improper threading is one of the top causes of ANRs. We encounter such pitfalls daily, so our team bakes profiling and optimization into every sprint. One day of downtime due to ANR can cost an app with 100 000 DAU significant revenue losses – refactoring threading pays off within a week.

Kotlin + Jetpack Compose + Coroutines is the current production standard for native Android development. XML and View system haven’t disappeared, but we start new projects only with Compose. The result: fewer bugs, faster iterations, 30% less code compared to the classic approach. Want to estimate savings on your project? Contact us – we’ll do a free code audit within half a day.

How does recomposition work in Jetpack Compose and why is it important?

Compose is a declarative UI framework. Instead of TextView.setText() and adapter.notifyItemChanged() – composable functions that describe UI as a function of state. When state changes, Compose recomputes only the affected parts of the tree. This is called recomposition.

Problem: recomposition can be too frequent. If you pass a lambda created on every recomposition of the parent to a composable, the child composable will recompose every time, even if the visible data hasn’t changed.

// Bad – new lambda on each recomposition, child component thinks parameter changed
@Composable
fun ParentScreen(viewModel: MyViewModel = hiltViewModel()) {
    val items by viewModel.items.collectAsState()
    ItemList(
        items = items,
        onItemClick = { id -> viewModel.selectItem(id) } // created anew each time
    )
}

// Good – remember stabilizes the lambda
@Composable
fun ParentScreen(viewModel: MyViewModel = hiltViewModel()) {
    val items by viewModel.items.collectAsState()
    val onItemClick = remember { { id: String -> viewModel.selectItem(id) } }
    ItemList(items = items, onItemClick = onItemClick)
}

Stability and @Stable/@Immutable

Compose determines whether to recompose a composable by checking the stability of parameters. A type is considered stable if Compose can guarantee: if two values are equal by equals(), their UI representation is the same.

Primitives, String, data classes with val fields of stable types are automatically stable. List<T> is unstable because it’s an interface. MutableList can change without notification. Solution: use ImmutableList from kotlinx.collections.immutable or annotate a data class with @Immutable.

// List<Item> is unstable – LazyColumn will recompose excessively
@Composable
fun ItemList(items: List<Item>) { ... }

// ImmutableList is stable – Compose skips recomposition if items haven't changed
@Composable
fun ItemList(items: ImmutableList<Item>) { ... }

For diagnosing recomposition issues we use Compose Compiler Metrics. Add flags -P plugin:androidx.compose.compiler.plugins.kotlin:reportsDestination=... to build.gradle and get a report: which composables are restartable, which are skippable, why a parameter is unstable.

LazyColumn and list performance

LazyColumn is the RecyclerView equivalent in Compose. key in items { } is mandatory for any list where items can move or be deleted. Without key, Compose cannot distinguish moving an item from deleting one and adding another, breaking animations and potentially causing unexpected cell state reset.

LazyColumn {
    items(
        items = messages,
        key = { message -> message.id } // stable identifier
    ) { message ->
        MessageItem(message = message)
    }
}

contentType is an additional optimization. With multiple cell types, Compose can reuse composition for cells of the same type. It’s analogous to getItemViewType in RecyclerView.

How to avoid common mistakes when using coroutines?

Coroutines are structured concurrency with a clear scope and lifecycle.

viewModelScope is a coroutine scope tied to the ViewModel lifecycle. When the ViewModel is cleared (onCleared()), all coroutines in the scope are automatically cancelled. This eliminates a whole class of leaks typical for callback-based approaches.

@HiltViewModel
class OrderViewModel @Inject constructor(
    private val orderRepository: OrderRepository
) : ViewModel() {

    private val _uiState = MutableStateFlow<OrderUiState>(OrderUiState.Loading)
    val uiState: StateFlow<OrderUiState> = _uiState.asStateFlow()

    fun loadOrder(orderId: String) {
        viewModelScope.launch {
            _uiState.value = OrderUiState.Loading
            try {
                val order = orderRepository.getOrder(orderId) // suspend function
                _uiState.value = OrderUiState.Success(order)
            } catch (e: IOException) {
                _uiState.value = OrderUiState.Error(e.message)
            }
        }
    }
}

What to choose: StateFlow or LiveData?

StateFlow and SharedFlow are the recommended replacements for LiveData in Kotlin projects. LiveData is lifecycle-aware but tied to the Android platform. Flow is pure Kotlin, testable without Android dependencies.

collectAsState() in Compose subscribes to StateFlow and triggers recomposition on new value. lifecycleScope.launch { flow.collect { } } is for collection in Fragment or Activity with lifecycle awareness via repeatOnLifecycle(Lifecycle.State.STARTED).

repeatOnLifecycle is important. Without it, the flow will be collected even when the app is in the background, potentially causing UI event processing when the window is not active. Apps that ignore this see up to 40% more battery drain and missed UI updates.

Dispatchers and structured concurrency

Dispatchers.IO for network requests and file operations. Dispatchers.Default for CPU-intensive tasks (parsing, sorting, encryption). Dispatchers.Main for UI.

withContext(Dispatchers.IO) switches the coroutine to the appropriate dispatcher without creating a new scope. This is more efficient than launch(Dispatchers.IO) inside another launch.

// Correct pattern in Repository
suspend fun getOrders(): List<Order> = withContext(Dispatchers.IO) {
    orderDao.getAll() // Room automatically suspend, but explicit IO dispatcher is good practice
}

Hilt and dependency injection

Hilt is the official DI framework for Android built on top of Dagger 2. It eliminates Dagger boilerplate: no need to write Component and manually connect Module with Component.

@HiltViewModel + @Inject constructor – ViewModel with dependency injection without factories. @Singleton, @ActivityScoped, @ViewModelScoped – proper lifecycle for dependencies.

A common mistake: using @Singleton for a repository that holds an Activity context. This leaks the Activity. Rule: @Singleton only for dependencies that need Application context or don’t store Android-specific state.

Want to implement DI without headaches? Contact us – we’ll set up Hilt within an hour on any existing project.

WorkManager and background tasks

WorkManager for guaranteed background tasks that must execute even after app or device restart. Data sync, analytics upload, file downloads.

CoroutineWorker is the suspend version of Worker. It runs on Dispatchers.IO by default.

Android 14 tightened background execution requirements. FOREGROUND_SERVICE_TYPE is mandatory for foreground services. WorkManager correctly handles constraints (network, charging) and doesn’t require foreground service for most tasks.

Tools

Android Studio Profiler – CPU profiler with System Trace shows everything: coroutine suspension points, RenderThread, MainThread. Memory profiler – heap dump, allocation tracking. Network profiler – all HTTP requests with bodies.

Compose Layout Inspector – composable tree with recomposition counts. Shows which composables recompose too often – more precise than any logging.

LeakCanary – automatic memory leak detection in development builds. Shows reference chain to the leak. Added with one dependency, works without configuration.

Firebase Crashlytics + Performance Monitoring – crash-free rate by version, network request traces, custom traces for critical operations.

What’s included in native Android development: our process

1. Requirements audit and architecture design – diagrams, stack selection, prototype.
2. Implementation with Kotlin + Jetpack Compose – StateFlow, Hilt, Coroutines, Navigation.
3. Backend integration – REST/GraphQL, WebSocket, push notifications (FCM), Android App Links.
4. Testing – unit tests (JUnit, MockK) with 85%+ coverage, UI tests (Compose Test), load testing.
5. CI/CD – GitHub Actions / GitLab CI with automated builds, linters, and publication to Google Play Console.
6. Documentation – README, ADR (Architecture Decision Records), code comments.
7. Post-release support – monitoring, crashlytics, hotfixes, updates.
8. Code warranty – 3 months of free support after delivery.

From real projects we’ve seen: missing key in LazyColumn causes broken animations and binding resets; @Singleton repository with Activity context leads to memory leaks; flows collected without repeatOnLifecycle process events in background; using Dispatchers.Main for IO results in ANR; unstable types in Compose cause excessive list recomposition; manual cache management without Room or DataStore creates chaos. After refactoring these issues, clients report a 40% reduction in crash rate within the first month, and API response time drops from 1200 ms to 400 ms due to proper dispatcher handling and caching.

Timelines

Cost is calculated after requirements analysis and specification. Estimate is free. Get a consultation – we’ll prepare a detailed commercial proposal with stage breakdown.

Why trust us

5+ years on the market, 70+ completed Android projects (from startups to enterprise). Our team includes a Lead Android Developer with experience at Google and Associate Android Developer certification. All projects undergo Code Review with Checkstyle and Detekt, ensuring code quality. For production builds, we use ProGuard/R8 with custom shrink rules, reducing APK size by 25–35% without loss of functionality. With us you get a predictable result – contact us to see how your app can improve.