Use server-side Remote Config with Cloud Functions and Vertex AI

This guide describes how to get started using 2nd gen Cloud Functions with server-side Remote Config to make server-side calls to the Vertex AI Gemini API.

In this tutorial, you'll add Remote Config to a chatbot-like function that uses the Gemini model to answer user questions. Remote Config will manage Gemini API inputs (including a prompt that you'll prepend to incoming user queries), and you can update these inputs on-demand from the Firebase console. You'll also use the Firebase Local Emulator Suite to test and debug the function, and then, after verifying that it works, you'll deploy and test it on Google Cloud.

Prerequisites

This guide assumes that you're familiar with using JavaScript to develop applications.

Set up a Firebase project

If you do not already have a Firebase project:

  1. Sign into the Firebase console.

  2. Click Create project, and then use either of the following options:

    • Option 1: Create a new Firebase project (and its underlying Google Cloud project automatically) by entering a new project name in the first step of the "Create project" workflow.
    • Option 2: "Add Firebase" to an existing Google Cloud project by selecting your Google Cloud project name from the drop-down menu in the first step of the "Create project" workflow.

  3. When prompted, you do not need to set up Google Analytics to use this solution.

  4. Continue to follow the on-screen instructions to create your project.

If you already have a Firebase project:

Proceed to Configure your development environment.

Configure your development environment

You'll need a Node.js environment to write functions, and you'll need the Firebase CLI to deploy functions to the Cloud Functions runtime.

  1. Install Node.js and npm.

    For installing Node.js and npm, we recommend using Node Version Manager.

  2. Install the Firebase CLI using your preferred method. For example, to install the CLI using npm, run this command:

    npm install -g firebase-tools@latest

    This command installs the globally-available firebase command. If this command fails, you may need to change npm permissions.

    To update to the latest version of firebase-tools, rerun the same command.

  3. Install firebase-functions and firebase-admin and use --save to save them to your package.json:

    npm install firebase-functions@latest firebase-admin@latest --save

You're now ready to proceed to implementation of this solution.

Implementation

Follow these steps to create, test, and deploy your 2nd gen Cloud Functions with Remote Config and Vertex AI:

  1. Enable Vertex AI recommended APIs in the Google Cloud console.
  2. Initialize your project and install Node dependencies.
  3. Configure IAM permissions for your Admin SDK service account and save your key.
  4. Create the function.
  5. Create a server-specific Remote Config template.
  6. Deploy your function and test it in the Firebase Local Emulator Suite.
  7. Deploy your function to Google Cloud.

Step 1: Enable Vertex AI recommended APIs in the Google Cloud console

  1. Open the Google Cloud console, and when prompted, select your project.
  2. In the Search field at the top of the console, enter Vertex AI and wait for Vertex AI to appear as a result.
  3. Select Vertex AI. The Vertex AI dashboard appears.

  4. Click Enable All Recommended APIs.

    It might take a few moments for API enablement to complete. Keep the page active and open until enablement finishes.

  5. If billing isn't enabled, you'll be prompted to add or link a Cloud Billing account. After enabling a billing account, return to the Vertex AI dashboard and verify that all recommended APIs are enabled.

Step 2: Initialize your project and install Node dependencies

  1. Open a terminal on your computer and navigate to the directory where you plan to create your function.

  2. Log into Firebase:

    firebase login

  3. Run the following command to initialize Cloud Functions for Firebase:

    firebase init functions

  4. Select Use an existing project and specify your project ID.

  5. When prompted to select the language to use, choose Javascript and press Enter.

  6. For all other options, select the defaults.

    A functions directory is created in the current directory. Inside, you'll find an index.js file that you'll use to build out your function, a node_modules directory that contains the dependencies for your function, and a package.json file that contains the package dependencies.

  7. Add the Admin SDK and Vertex AI packages by running the following commands, using --save to ensure that it's saved to your package.json file:

    cd functions
npm install firebase-admin@latest @google-cloud/vertexai --save

Your functions/package.json file should now look like the following, with the latest versions specified:

  {
    "name": "functions",
    "description": "Cloud Functions for Firebase",
    "scripts": {
      "serve": "firebase emulators:start --only functions",
      "shell": "firebase functions:shell",
      "start": "npm run shell",
      "deploy": "firebase deploy --only functions",
      "logs": "firebase functions:log"
    },
    "engines": {
      "node": "20"
    },
    "main": "index.js",
    "dependencies": {
      "@google-cloud/vertexai": "^1.1.0",
      "firebase-admin": "^12.1.0",
      "firebase-functions": "^5.0.0"
    },
    "devDependencies": {
      "firebase-functions-test": "^3.1.0"
    },
    "private": true
  }

Note that if you're using ESLint, you'll see a stanza that includes it. In addition, make sure that the node engine version matches your installed version of Node.js and the version you ultimately run on Google Cloud. For example, if the engines stanza in your package.json is configured as Node version 18 and you're using Node.js 20, update the file to use 20:

  "engines": {
    "node": "20"
  },

Step 3: Configure IAM permissions for your Admin SDK service account and save your key

In this solution, you'll use the Firebase Admin SDK service account to run your function.

  1. In the Google Cloud console, open the IAM & Admin page, and locate the Admin SDK service account (named firebase-adminsdk).
  2. Select the account and click Edit principal. The Edit access page appears.
  3. Click Add another role, select Remote Config Viewer.
  4. Click Add another role, select AI platform developer.
  5. Click Add another role, select Vertex AI user.
  6. Click Add another role, select Cloud Run Invoker.
  7. Click Save.

Next, export the credentials for the Admin SDK service account and save them in your GOOGLE_APPLICATION_CREDENTIALS environment variable.

  1. In the Google Cloud console, open the Credentials page.
  2. Click the Admin SDK service account to open the Details page.
  3. Click Keys.
  4. Click Add key > Create new key.
  5. Ensure that JSON is selected as the Key type, then click Create.
  6. Download the key to a safe place on your computer.

  7. From your terminal, export the key as an environment variable:

    export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/service-account-key.json"

Step 4: Create the function

In this step, you'll construct a function that handles user input and generates AI-powered responses. You'll combine multiple code snippets to build a comprehensive function that initializes the Admin SDK and Vertex AI Gemini API, configures default parameters using Remote Config, fetches the latest Remote Config parameters, processes user input, and streams a response back to the user.

  1. In your codebase, open functions/index.js in a text editor or IDE.

  2. Delete the existing content and then add the Admin SDK, Remote Config, and the Vertex AI SDK and initialize the app by pasting the following code into the file:

    const { onRequest } = require("firebase-functions/v2/https");
const logger = require("firebase-functions/logger");

const { initializeApp } = require("firebase-admin/app");
const { VertexAI } = require('@google-cloud/vertexai');
const { getRemoteConfig } = require("firebase-admin/remote-config");

// Set and check environment variables.
const project = process.env.GCLOUD_PROJECT;

// Initialize Firebase.
const app = initializeApp();
    index.js

  3. Configure default values that your function will use if it can't connect to the Remote Config server. This solution configures textModel, generationConfig, safetySettings, textPrompt, and location as Remote Config parameters that correspond with Remote Config parameters that you'll configure further on in this guide. For more information about these parameters, see Vertex AI Node.js client.

    Optionally, you can also configure a parameter to control whether or not you access the Vertex AI Gemini API (in this example, a parameter called vertex_enabled). This setup can be useful when testing your function. In the following code snippets, this value is set to false, which will skip using Vertex AI while you test basic function deployment. Setting it to true will invoke the Vertex AI Gemini API.

    // Define default (fallback) parameter values for Remote Config.
const defaultConfig = {

  // Default values for Vertex AI.
  model_name: "gemini-1.5-flash-preview-0514",
  generation_config: [{
    "stopSequences": [], "temperature": 0.7,
    "maxOutputTokens": 64, "topP": 0.1, "topK": 20
  }],
  prompt: "I'm a developer who wants to learn about Firebase and you are a \
    helpful assistant who knows everything there is to know about Firebase!",
  safety_settings: [{
    "category":
      "HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT",
    "threshold": "HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE"
  }],
  location: 'us-central1',

  // Disable Vertex AI Gemini API access for testing.
  vertex_enabled: false
};
    index.js

  4. Create the function and set up server-side Remote Config:

    // Export the function.
exports.generateWithVertex = onRequest(async (request, response) => {

  try {

    // Set up Remote Config.
    const rc = getRemoteConfig(app);

    // Get the Remote Config template and assign default values.
    const template = await rc.getServerTemplate({
      defaultConfig: defaultConfig
    });

    // Add the template evaluation to a constant.
    const config = template.evaluate();

    // Obtain values from Remote Config.
    const textModel = config.getString("model_name") ||
        defaultConfig.model_name;
    const textPrompt = config.getString("prompt") || defaultConfig.prompt;
    const generationConfig = config.getString("generation_config") ||
        defaultConfig.generation_config;
    const safetySettings = config.getString("safety_settings") ||
        defaultConfig.safety_settings;
    const location = config.getString("location") ||
        defaultConfig.location;
    const vertexEnabled = config.getBoolean("is_vertex_enabled") ||
        defaultConfig.vertex_enabled;
    index.js

  5. Set up Vertex AI and add the chat and response logic:

      // Allow user input.
  const userInput = request.query.prompt || '';

  // Instantiate Vertex AI.
    const vertex_ai = new VertexAI({ project: project, location: location });
    const generativeModel = vertex_ai.getGenerativeModel({
      model: textModel,
      safety_settings: safetySettings,
      generation_config: generationConfig,
    });

    // Combine prompt from Remote Config with optional user input.
    const chatInput = textPrompt + " " + userInput;

    if (!chatInput) {
      return res.status(400).send('Missing text prompt');
    }
    // If vertexEnabled isn't true, do not send queries to Vertex AI.
    if (vertexEnabled !== true) {
      response.status(200).send({
        message: "Vertex AI call skipped. Vertex is not enabled."
      });
      return;
    }

    logger.log("\nRunning with model ", textModel, ", prompt: ", textPrompt,
      ", generationConfig: ", generationConfig, ", safetySettings: ",
      safetySettings, " in ", location, "\n");

    const result = await generativeModel.generateContentStream(chatInput); 
    response.writeHead(200, { 'Content-Type': 'text/plain' });

    for await (const item of result.stream) {
      const chunk = item.candidates[0].content.parts[0].text;
      logger.log("Received chunk:", chunk);
      response.write(chunk);
    }

    response.end();

  } catch (error) {
    logger.error(error);
    response.status(500).send('Internal server error');
  }
});
    index.js

  6. Save and close the file.

Step 5: Create a server-specific Remote Config template

Next, create a server-side Remote Config template and configure parameters and values to use in your function. To create a server-specific Remote Config template:

  1. Open the Firebase console and, from the navigation menu, expand Run and select Remote Config.

  2. Select Server from the Client/Server selector at the top of the Remote Config page.

    • If this is your first time using Remote Config or server templates, click Create Configuration. The Create your first server-side parameter pane appears.
    • If this is not your first time using Remote Config server templates, click Add parameter.

  3. Define the following Remote Config parameters:

    Parameter name Description Type Default value
    model_name Model name
    For up-to-date lists of model names to use in your code, see Model versions and lifecycles or Available model names.     		String gemini-1.5-pro-preview-0514
    prompt Prompt to prepend to the user's query. String I'm a developer who wants to learn about Firebase and you are a helpful assistant who knows everything there is to know about Firebase!
    generation_config Parameters to send to the model. JSON [{"stopSequences": ["I hope this helps"],"temperature": 0.7,"maxOutputTokens": 512, "topP": 0.1,"topK": 20}]
    safety_settings Safety settings for Vertex AI. JSON [{"category": "HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "HarmBlockThreshold.BLOCK_LOW_AND_ABOVE"}]
    location Location to run the Vertex AI service and model. String us-central1
    is_vertex_enabled Optional parameter that controls whether queries are sent to Vertex AI. Boolean true

  4. When you've finished adding parameters, double-check your parameters and that their data types are correct, then click Publish changes.

Step 6: Deploy your function and test it in the Firebase Local Emulator Suite

Now you're ready to deploy and test your function locally with the Firebase Local Emulator Suite.

  1. Make sure that you've set GOOGLE_APPLICATION_CREDENTIALS as an environment variable as described in Step 3: Configure IAM permissions for your Admin SDK service account and save your key. Then, from the parent directory of your functions directory, deploy your function to the Firebase emulator:

    firebase emulators:start --project PROJECT_ID --only functions

  2. Open the emulator's logs page. This should show that your function has loaded.

  3. Access your function by running the following command, where PROJECT_ID is your project ID and LOCATION is the region that you deployed the function to (for example, us-central1):

    curl http://localhost:5001/PROJECT_ID/LOCATION/generateWithVertex

  4. Wait for a response, then return to the Firebase Emulator logs page or your console and check for any errors or warnings.

  5. Try sending some user input, noting that because is_vertex_enabled is configured in your Remote Config server template, this should access Gemini through the Vertex AI Gemini API and that this may incur charges:

    curl http://localhost:5001/PROJECT_ID/LOCATION/generateWithVertex?prompt=Tell%20me%20everything%20you%20know%20about%20cats

  6. Make changes to your Remote Config server template on the Firebase console, then re-access your function to observe changes.

Step 7: Deploy your function to Google Cloud

After you've tested and verified your function, you're ready to deploy to Google Cloud and test the live function.

Deploy your function

Deploy your function using the Firebase CLI:

firebase deploy --only functions

Block unauthenticated access to the function

When functions are deployed using Firebase, unauthenticated invocations are allowed by default if your organization's policy doesn't restrict it. During testing and before securing with App Check, we recommend blocking unauthenticated access.

To block unauthenticated access to the function:

  1. In the Google Cloud console, open Cloud Run.

  2. Click generateWithVertex, then click the Security tab.

  3. Enable Require authentication and then click Save.

Configure your user account to use the Admin SDK service account credentials

Because the Admin SDK service account has all the necessary roles and permissions to run the function and interact with the Remote Config and the Vertex AI Gemini API, you'll want to use it to run your function. To do this, you must be able to create tokens for the account from your user account.

The following steps describe how to configure your user account and the function to run with the Admin SDK service account privileges.

  1. In the Google Cloud console, enable the IAM Service Account Credentials API.
  2. Give your user account the Service Account Token Creator role: From the Google Cloud console, open IAM & Admin > IAM, select your user account, and then click Edit principal > Add another role.

  3. Select Service Account Token Creator, then click Save.

    For more detailed information about service account impersonation, refer to Service account impersonation in the Google Cloud documentation.

  4. Open the Google Cloud console Cloud Functions page and click the generateWithVertex function in the Functions list.

  5. Select Trigger > Edit and expand Runtime, build, connections and security settings.

  6. From the Runtime tab, change the Runtime service account to the Admin SDK account.

  7. Click Next, then click Deploy.

Set up the gcloud CLI

To securely run and test your function from the command line, you'll need to authenticate with the Cloud Functions service and obtain a valid authentication token.

To enable token generation, install and configure the gcloud CLI:

  1. If not already installed on your computer, install the gcloud CLI as described in Install the gcloud CLI.

  2. Obtain access credentials for your Google Cloud account:

    gcloud auth login

  3. Set your project ID in gcloud:

    gcloud config set project PROJECT_ID

Test your function

You're now ready to test your function in Google Cloud. To test the function, run the following command:

curl -X POST https://LOCATION-PROJECT_ID.cloudfunctions.net/generateWithVertex \
  -H "Authorization: bearer $(gcloud auth print-identity-token)" \
  -H "Content-Type: application/json"

Try again with user-supplied data:

curl -X POST https://LOCATION-PROJECT_ID.cloudfunctions.net/generateWithVertex?prompt=Tell%20me%20everything%20you%20know%20about%20dogs \
 -H "Authorization: bearer $(gcloud auth print-identity-token)" \
 -H "Content-Type: application/json"

You can now make changes to your Remote Config server template, publish those changes, and test different options.

Next steps