Ask an LLM — one button, one answer

A small Java / View Binding example with Firebase AI Logic

Secure Google-side setup · עברית

Press Ask LLM, see Thinking…, then see the generated answer in one label. No new Git repository, data table, model class, or JSON parser is needed. Start with a Java Empty Views Activity, package com.example.topic, and complete the separate setup page first. Keep all unrelated template code and dependencies. Use Android API 24 or higher; keep a higher template minimum if already set. Use a separate emulator if another TopicProjects exercise has the same application ID.

1. One button and a scrollable label

In Gradle Scripts > build.gradle.kts (Module :app), enable View Binding inside the existing android block. If buildFeatures already exists, add only its viewBinding setting.

android {
    // Keep the existing namespace, SDK levels, build types, etc.
    buildFeatures {
        viewBinding = true
    }
}

In app > res > layout > activity_main.xml, replace only the Hello World TextView with the following children. Keep the root ConstraintLayout and its main ID. The ScrollView lets longer answers remain readable.

android:layout_height="match_parent"
     tools:context=".MainActivity">

-    <TextView
-        android:layout_width="wrap_content"
+    <Button
+        android:id="@+id/ask_llm"
+        android:layout_width="0dp"
         android:layout_height="wrap_content"
-        android:text="Hello World!"
+        android:layout_margin="16dp"
+        android:text="Ask LLM"
+        app:layout_constraintTop_toTopOf="parent"
+        app:layout_constraintStart_toStartOf="parent"
+        app:layout_constraintEnd_toEndOf="parent" />
+
+    <ScrollView
+        android:layout_width="0dp"
+        android:layout_height="0dp"
+        android:layout_margin="16dp"
+        app:layout_constraintTop_toBottomOf="@id/ask_llm"
         app:layout_constraintBottom_toBottomOf="parent"
-        app:layout_constraintEnd_toEndOf="parent"
         app:layout_constraintStart_toStartOf="parent"
-        app:layout_constraintTop_toTopOf="parent" />
-
+        app:layout_constraintEnd_toEndOf="parent">
+        <TextView
+            android:id="@+id/answer_label"
+            android:layout_width="match_parent"
+            android:layout_height="wrap_content"
+            android:text="The answer will appear here."
+            android:textIsSelectable="true" />
+    </ScrollView>
 </androidx.constraintlayout.widget.ConstraintLayout>

View Binding turns XML IDs ask_llm and answer_label into the Java fields askLlm and answerLabel. The binding class is generated; do not write it yourself.

2. Connect the screen and the prepared model

In app > kotlin+java > com.example.topic > MainActivity, add these imports below the existing imports:

import androidx.core.content.ContextCompat;
import com.example.topic.databinding.ActivityMainBinding;
import com.google.firebase.FirebaseApp;
import com.google.firebase.appcheck.FirebaseAppCheck;
import com.google.firebase.appcheck.debug.DebugAppCheckProviderFactory;
import com.google.firebase.ai.FirebaseAI;
import com.google.firebase.ai.GenerativeModel;
import com.google.firebase.ai.java.GenerativeModelFutures;
import com.google.firebase.ai.type.Content;
import com.google.firebase.ai.type.GenerateContentResponse;
import com.google.firebase.ai.type.GenerativeBackend;
import com.google.common.util.concurrent.FutureCallback;
import com.google.common.util.concurrent.Futures;
import com.google.common.util.concurrent.ListenableFuture;

Inside onCreate, change the existing layout references as shown; keep the insets listener’s body unchanged:

         EdgeToEdge.enable(this);
+        ActivityMainBinding binding = ActivityMainBinding.inflate(getLayoutInflater());
         setContentView(R.layout.activity_main);

Before

setContentView(R.layout.activity_main);
ViewCompat.setOnApplyWindowInsetsListener(findViewById(R.id.main), (v, insets) -> {
    Insets systemBars = insets.getInsets(WindowInsetsCompat.Type.systemBars());

After

setContentView(binding.getRoot());
ViewCompat.setOnApplyWindowInsetsListener(binding.main, (v, insets) -> {
    Insets systemBars = insets.getInsets(WindowInsetsCompat.Type.systemBars());

After the insets listener’s closing });, initialize Firebase, install the classroom App Check provider, and create the Java model wrapper:

FirebaseApp.initializeApp(this);
// Use the classroom token registered on the setup page.
FirebaseAppCheck.getInstance().installAppCheckProviderFactory(
        DebugAppCheckProviderFactory.getInstance());
GenerativeModel ai = FirebaseAI.getInstance(GenerativeBackend.googleAI())
        .generativeModel("gemini-3.8-flash");
GenerativeModelFutures model = GenerativeModelFutures.from(ai);

The model name is a dated example from the supported-model list, checked on 24 October 2026. If it is retired later, use a currently supported text model. No Gemini API key is passed to this constructor.

3. Add the click action

Immediately after the model initialization, still inside onCreate, add:

// Android calls this lambda when the user clicks ask_llm.
binding.askLlm.setOnClickListener(v -> {
    binding.answerLabel.setText("Thinking...");
    binding.askLlm.setEnabled(false); // One request at a time.
    Content prompt = new Content.Builder()
            .addText("Explain a callback in two simple sentences.")
            .build();

    // Start asynchronously: future is a handle, not the answer.
    ListenableFuture<GenerateContentResponse> future = model.generateContent(prompt);
    // Attach ONE callback object with TWO possible outcome methods.
    Futures.addCallback(future, new FutureCallback<GenerateContentResponse>() {
        /** Called on success; result is supplied by the callback mechanism. */
        @Override
        public void onSuccess(GenerateContentResponse result) {
            if (isFinishing() || isDestroyed()) return;
            String text = result.getText(); // SDK already decoded the response.
            binding.answerLabel.setText(text != null ? text : "No text returned.");
            binding.askLlm.setEnabled(true);
        }

        /** Called on failure; error describes what went wrong. */
        @Override
        public void onFailure(Throwable error) {
            if (isFinishing() || isDestroyed()) return;
            binding.answerLabel.setText("Request failed: " + error.getMessage());
            binding.askLlm.setEnabled(true);
        }
    }, ContextCompat.getMainExecutor(this)); // Run BOTH callbacks on the UI thread.
    // This click callback ends now; the screen does not wait for the answer.
});

Build and run now. If this is the first run with your debug App Check provider, register its private token as described on the setup page, then retry. Press Ask LLM: the label should change immediately, and the button should become available again after success or failure.

flowchart LR
    A["Click Ask LLM"] --> B["Start request: receive a future"]
    B --> C["Screen stays responsive"]
    B --> D{"Future completes"}
    D -->|success| E["onSuccess: display text"]
    D -->|failure| F["onFailure: display error"]

Futures.addCallback receives three arguments: which future, which callback object, and which executor. The callback methods run later on the main thread because we explicitly chose ContextCompat.getMainExecutor(this). This does not move networking onto the UI thread. The isFinishing / isDestroyed checks skip updates to an Activity that has ended. This classroom example does not preserve a pending request through rotation; leaving the screen does not undo server work.

Predict and check

  • Change the prompt to request an answer in Hebrew. Predict which line supplies the request text and which line displays the result.
  • Turn off networking and try once. Expect a visible failure and a re-enabled button; restore networking afterwards. Generated wording varies between successful calls.
  • Keep the answer selectable: copy it into Notepad++ and inspect it as text.
  1. Is the future the answer? No: it represents an asynchronous operation; the callback receives the eventual result.
  2. Why two overridden methods? The anonymous FutureCallback object implements success and failure callbacks. Exactly one outcome is delivered for that future.
  3. Where is the JSON? The SDK handles HTTP and decodes the response into GenerateContentResponse. getText() extracts text from that object. Use the Petstore lab to inspect raw JSON yourself.

References: Firebase’s Java request example, Android’s main-thread executor.