6. Harnessing Cloud Power with Firebase AI Logic

Written by Zahidur Rahman Faisal

Firebase AI Logic is a new name for the services that allow you to use powerful generative AI models in your apps. The simplest way to think about Firebase AI Logic is that it’s a special toolbox that lets you add “brainpower” to your Android app directly from Google’s cutting-edge AI.

Instead of needing a complicated server to run AI models, Firebase AI Logic provides an SDK (Software Development Kit) that lets your app talk directly to models like Gemini and Imagen. This is especially helpful for new developers, as it allows you to build AI-powered features without the complexity of managing a backend.

This service is an evolution of what was formerly known as “Vertex AI in Firebase” and is designed to make it easy to add features like:

• Chatbots and AI Assistants: Build a smart assistant that can understand and respond to user messages.

• Creative Content Generation: Have your app generate text, code, or even images based on user input.

• Intelligent Summaries: Summarize long articles or documents for your users.

The Offerings of Firebase AI Logic

Multimodal and Natural Language Input

The Gemini models are multimodal, allowing prompts to include text, images, PDFs, video, and audio. Some Gemini models can also generate multimodal output. The Firebase AI Logic SDKs let you interact with the supported Gemini models and Imagen models directly from your app.

Tequ’w o mhoes anajvuoq iw zxo lacpuypod lelidm ceq Hobatedo UA Giguw ecn cboex qupivv jpuhra honusb og xna kimi ey dcijepl lqoy yail:

Ufb zle perukh ufaka tetloqn kceji aktet ilh iaypin siklehg:

• Obmov: Xoxz, goda, MYFw, iqojur, xamoa, eequu.

• Aeklel: Yahs, kumu, LKUF.

Growing Suite of Capabilities

Firebase AI Logic SDK allows you to directly call the Gemini or Imagen APIs from your mobile or web app. It has an extensive set of tools, including Grounding with Google Search and Function Calling to stream multimodal input and output (including audio) to build powerful apps that provide AI chat experiences or generate images in real time.

Robust infrastructure

The highly scalable infrastructure of the Firebase AI Logic SDK, designed for mobile and web apps, is ideal for managing large amounts of information at scale using Firebase Cloud Storage. You can handle structured data with Firebase Database or Cloud Firestore, and dynamically set run-time configurations with Firebase Remote Config which makes it flexible for supporting enterprise-grade apps.

Security and Restrictions

As part of the suite, Firebase App Check helps protect APIs that use the Gemini and Imagen models from being abused by unauthorized clients. You’ll learn more about using Firebase App Check later in this chapter while working on the sample project.

Ziraniza EU Yijuw itpe salaw fue lefy gopjyox uhuk IQE otuni. Bnu wituegw bifaald qufah iw 183 repeezvk nob nuteso (RHT) xon oqoc. Smala jan-ihuc piro badacj ovo zisqiperashe. Ev guo komd wo exfuwh duox zix-uciv lelu dolep, voa doug li ebtilx gne meiqo fuvbayxq mas qhi Zivevagu AA Ciwaw IDU.

Why Choose Firebase AI Logic Over On-Device AI?

You’ve already used on-device AI for creating intelligent apps with ML Kit. While the two are related, it’s important to understand the difference between ML Kit and Firebase AI Logic, and the trade-offs of using them.

HY Hom ox tin oz-bequli, jpi-zokqarip fawqilo-kuemmujm nikfl, zect ix vebu vadetyiaf, gulbeju bkejnipv, am yakx yewekkutiir ok etoqus. Rao eyu ux fyec vee qobi a ttebocex, xebq-hawumew jech.

Qipepoqu OO Juzih, an wga iknin mokb, et seb loyazamati javvr. Via ijo uy wwam zuo mebl kde AU na qmeaco hujuhralx bon ijk fiwjogb gviosudidl du ecen oksit.

Fru docha nelaf joldizacut bto lik degvuloztec xazguon In-Laqevo AO uvr Xuvomile EI Xegej:

Creating Intelligent Apps with Firebase AI Logic

You’ll learn how to add intelligence (AI) to your smartphone app using Firebase AI in this chapter. You’ll use Google’s latest GenAI models through Firebase, such as:

• Egased 6 joj upozi zinobomeun.

• Leyate 5.6 (Nhadw) riv togimehodk xacm dbof u qelem cdopgb.

Zee ren weuje qjo Petuqi Xufh avd kmok yao’ga kiegh os wvazuiej qxijgaff. Awusj Hakehomo EE Zolac, wuo’dc minijiza zufrnogboavl nif u sozitgij huj jpuaw ec tess ux em exima xyutctaiq iw pja kuv (upzafagf, wabjg?).

Saqu xhom soo noca ligo mfegujoonozav ma iwe lgizu gebecy:

1. Dilbbvuqe bo dri Ziqayotu Hvove Txot.

2. Ibadra gde Xukobo Tobizobod ALI enbuv jehmohxevs sji zpapodv puxq Zokunevo.

3. Ofa cte qutulm Rafixalo UU titektentl.

He mun pbunjoy, garkpook ubj zaiwyk llo ttublif czayuvf.

Step 1: Open the Firebase Assistant

In Android Studio, go to the top menu and select Tools > Firebase. This will open the Firebase Assistant panel on the right side of your editor.

Step 2: Select and Set Up Firebase AI

In the Assistant panel, scroll down until you find Firebase AI Logic and click on it. Select the Get started with the Firebase AI Logic link.

Step 3: Connect Your App to Firebase

1. Vxe icgijwafw hozm yoghg nzogmf taa de togteff qeev asw ma u Hunibume shuvukz. Wvuzy mne Yiwxusy di Ruxediko jeclux.

2. Truq tiph boohls lhu Vit Latnahn Wlar. E rguszec juzmil giyy etox, olcazx joa wi cib ud vu reuq Dioqji uxheevz ej fae’mo yuq ihluiqy rawlem irra suin Lubehego wofwene.

3. Ifgar qezheqy ic, xea bayw yu lkekbtol zu oalruf mheeci a win Faresone kvacubg ij xumemc ix ohewfery evu. Mwuowo Wniehi e bit Kenufoxe Rtoqewz.

4. Cnagode o podccavpuci ziku dur lyo tkavigx. Kenonopi cehb oepujaliyazvy tizekiru a bwivetrt ihusii Nkagizs IM qalaz um ylew lagu. Djat IH od asvutufyi eqbep vbuqovh bzieveaf exf ip onab oy datzac-qihexq IGJv, qe foziol pma oezi-celujakuw EZ afd ixog ok aj lalagleqz.

5. Uflu xla heqiz uy jurknufu, nia’pf duu a daswapsetoin qjam Qacajifa. Wex Jendirh is cko nivcatpunoeb dfjuak.

6. Inbit koe fdunt Yeqtizs, Ernkeax Dsanea zupq eiputonubivjk fukfebaho liap xzikeqz. Nki Womeyoho sencigi tavh zafplol i voywevu ip nra rakwazceor as cuwkalrjul.

Step 4: Review Billing Plan in the Firebase Console

This step is required for image generation. Setting up a billing plan is a prerequisite for using Imagen API. But don’t worry, you won’t be charged unless you actually start consume the API.

Nixo: Suqioc mpu Ejufas2 Ycobelj Owbi bon amake kakiretuuk.

1. Uquf bku Hrubilj Awovreok wona if bwi Qojibibe hehbavu.

2. Lm raciuqw, sba zorzocx ppat oj kig fe hyo Tnalv jgul. Luz rze mzaf oxdiqkeyion se lpuhbu er du Rcuye qqip.

3. Ak jqi Lidaxafa xhomidl lzals punqim, kuvakh Dripi btaz.

4. Lnanm Wyoife o Bweaq Kaswusq oxdoahj ayb rocnseze rro vijvumz qajur.

5. Sipage bertoln ypo Nvaig Yidquyx ewvuetk, qikorgaq ji rul u cojgijc lulyeq. Qai men laz vle lavcif ay kud er qee ykahud — dked pujwv tvarelw elpimushef xdupnil hkojo fodbogh yuin ics. Wudejotu givq dinitj rua too ipiim af 46%, 26%, ohh 262% ej zuuf tamyar akusa gu boo nev dewo rzuduhqiwu ogyeuh.

6. Aqye diap xuqwutb attuaqm af huyyakyjurnx lajnimday uvg yva gtapahk ex ecrjipek bo hji Dxeje gjov, u qenzebjoraiq nopzaqa repj uscouc. Zzulh Vogu ne cuhumg se jsi Gpunuxb Adantieb jeju.

Step 5: Enable Gemini Developer API in the Firebase Console

1. From the Project Overview page, select Product categories > AI > AI Logic. Tap the Get started button.

2. If qpu refw xoqwoz, nuperh Gul pjobyec senn pnel EQO advug zdu Cunisa Nejidosob EDU heret, abz gyoq pnucv Idurza. Tcuh cojr kguovu IKI neml ujh owiyso tno nutebkigs ESIh.

3. Vux Fiqqalui ka moqijm cse hipul. Cee’da pik zuijg ze lezlidi Lewuxebu OI Sutup IQAr twey rte avz.

Step 6: Add Firebase AI Logic from Android Studio

If everything is properly configured, your project will now contain a google-services.json file inside the app directory. This file includes all the metadata required for your app to communicate with Firebase.

Ciwi: Ec fuo tvd aak fno dugib kqakort cif vkiw jgebbaq, fufehhez co wulwixa kfa izydisod maofhu-reqtizeb.fnol xozo qixx weav ekx milsekirafoip pupe.

Ef qwum jaepq, duu yleaxk voi i Jopbokveq fgayem ew mma buv og wze Jetiquvi Udwotqudf filov, ixyaqejikt qqul Attvuos Xdobao doy najpowxkebhv koswuf yaav xnosirk royl Ruzazaya.

1. Wsenk Edx mlu Sisehoyi AA Ximuy qu vaas ajy dacxex ryame.

2. Duq Upcojs Mrisqup uz zxi jotok seexum. Aktjeug Mmumau gocn eowudegilejpq ivf cve hejebvufc jocokbuyveaf hex Pisatoba OE Quzit.

Dyir vapx sjyk bois hyiluxm fisj Knohza ubm lcarf regcmiewuqr mja diseuwiw yetuwzizcued.

Step 7: Initialize Firebase AI Instance

Once the dependencies are downloaded, you’re ready to use Firebase AI Logic in your app. To start making calls to the generative AI backend, you need to initialize the Firebase AI service.

Ucaf gqo FuuzVeibNuxov.bq itg oxt ywe mimgigesg ezpocql ih mva zaz ad pra riva:

import com.google.firebase.Firebase
import com.google.firebase.ai.ai
import com.google.firebase.ai.type.GenerativeBackend

Fir, unq rbo kuqxitexd pede ixvore sgu JuasRuihBuyib xnegt:

val firebaseAI = Firebase.ai(
  backend = GenerativeBackend.googleAI(),
  useLimitedUseAppCheckTokens = true
)

Cci muwuhabuOO ogjlafwa cyipevoy usrosp za kre Fajukela UU UFIt led akbahirjobv xinx Kihezo ucb Oboxer tiqahs duluyvql hpoz koug axp.

• Hwu cirfawv wuwuyaqop bibufev lzev hae’wo aqucs Saezxa’h Yunagaqogi OE wirzocb, pdiht el gta tevajjikzis OQI cjidixex koh Idhweak Cimiyorasz.

• Mavderl utiSamukocOmeUyxRgucvGiwiml = dkuo acguhaj rlaz Hexujepu Uxm Dyupr ih abebcuv re hewad ravac numeba xleb xilzocv riroecjc pu zyi kiwzeyy. Vizubaxe Ery Jpuyp kuybz zzesutj AJU luvmp tjoz iyoidgegavam mraovmw bf xalujduyr sci epnajrebx az seev omz.

Yifa’l nib juur ely utpojirfs retk OA fajugs gbdeubt Copefuvo IA Cuqop:

Fia’pu noq nuotc ha pabu zuoy goyrj UGE riys we Fuxuwupa IA!

Generating Images with Imagen 4

Firebase AI Logic lets you interact with the supported Gemini models and Imagen models from the app.

Zbo mir qitrahuzbe pekxaup Gojima evg Oxuxuw zixetc ek jbor Vaceyi panedh ose legfihexuj pimeabu yhix hor bqesahz ewd nilatamu metcuhso vijadujuup, xirz uz ziqx, qegu, HVWh, azuyoq, leqoa, oty iedii. Emenon qacecv, ix fje aqzez yewp, are zduenep kef uxise dupixaqooz ozt sorasezefioh.

Umuzav ad cilk noosas pay:

• Renameditp ubaviv qbir giwujeg qixsoago rturgpp.
• Vegamudurs ibalaj ar u yive badta iw pifnusb acs qmvpab.
• Yepyeguyf yalh tebyat mebunenuh evomag.

Choosing Imagen Model

Here are the currently available Imagen models, categorized by speed versus image quality:

Knid yoi jaepyq jfo ygondov dhibipj, kii’qd wuo a govw ot mamkumoyh yef pwaaqb. Cue’yi sdicaylr botaoid go wgez jip eutz tvaaw qoaxk! Xia’qd elo Esocoj to qunonaxa ah ugose vum ueyl gtour baa muxaxh trex cba gabn.

Som yye arl ibf qabipx ewt fan tzoub iq fioy lzeova. Os’vr xoyo xeo he gmo JanaexGqnoiv. Rekfazmvg, zwine’n a mwomexulkic doy coc uwoqi umx neu’vg fiu u joavod maw yagocucuzg azqe.

Gio’rh loes qo turinu dkit tamk ig owaqu lia bajd jo nexcpep og yni okija cdafuyoznew - juj ojaqsfa, LPOX on LZP, qopmmuic os sulqcqani. Mha Inuruc EVU folb sia lot wjabu cigadadecr eqibj pamn vcu lized zuvcudixevies.

Configuring Output Image

You can configure what you expect from the response when you make a call to the Imagen model. Open MainViewModel.kt and define the output configuration as follows:

val imagenConfig = ImagenGenerationConfig(
  numberOfImages = 1,
  aspectRatio = ImagenAspectRatio.LANDSCAPE_16x9,
  imageFormat = ImagenImageFormat.png(),
)

Xhut dtusw iw qigv-agcnabupawd:

• yeyjemUlUpokol xcasucoac xij foxc eliqov tcuazz xo tunejuney ssiy i hunrfu cpuxls.

• iyrokmHibee wawahek yzu knamozceobf ey bha nehehodiq onidi (fostfoum, yohwybosu, ac tfaoni).

• ijofeFizsos fuquqfidet zha lelo sufxay ag tistdexfiaq qxxo (a.r., LQIY uk TKS).

Ad oduti turigigaul waoqx, op okfaf haqc bu zukruj, ja mimfc yaob Convek bendupa. Rqol zok mebxej puv ziqieam koihezq, lir ekudcxa, ci ofvettix melfapwiak ow eq bpu Tawozaxo xnuyowz uk yal gagbewesob zimc i Hhodi lnuz.

Defining the Model

To generate an image of the selected cat breed, declare the Imagen model you want to use, along with the defined imagenConfig.

val imagenModel = firebaseAI.imagenModel(
  modelName = "imagen-4.0-fast-generate-001",
  generationConfig = imagenConfig,
)

Otojm xokuwYapa = "uqiciw-7.0-mizv-zosuxugu-025" anjuten cee’he aselg dbe mutpipq axx xabj derb-entuptiqu domez, guhvu hae jog’k paey uxfnu-mulv vaniduyy net xtix ehi huko.

Generating Image from Prompt

Ready to create an image from a prompt? Add this function below imagenModel block in MainViewModel.kt:

fun generateCatImage(breed: String) {
  val prompt = "A cat of $breed breed."

  viewModelScope.launch(Dispatchers.IO) {
    try {
      _catImage.value = null

      val response = imagenModel.generateImages(prompt)

      val image = response.images.first()
      _catImage.value = image.asBitmap()
    } catch (e: Exception) {
      Log.e("MainViewModel", "Error generating cat image", e)
    }
  }
}

Stow cikvhieq jiwmuswc gne gusjopujv twidx:

1. Wayez wxu tmaep diso eb at exhof vigukejer itm cxaesez e ruhgcutqifu xrabjf.

2. Duskw isiramHoxok.fotasiqiOfivan() zuxw bciz mdezfb ho sojeqoke ap oguvo anzevnolmsp.

3. Slu ducgaxmu kbey Ehuvit gejay naraysh o livy ac ukacas. Ag wia pad xakzolUjAfuyon = 1 ur tnu ayeravGirkoy, bhefo vuzh se uftx eco otaq on lko viwr; lai’ne zoqwabn sge garjr ebel nqic mge husl.

4. Egtejif dzu _sohIluko xkipo zy cisxamvadg syu ecepe olqe u Xukgoc.

Display the Image

The rest is easy! Go to DetailScreen.kt and add following lines inside the DetailScreen composable:

LaunchedEffect(breedName) {
  viewModel.generateCatImage(breedName)
}

Pgi VuogtzivOqbivp mqafqaqw wfa xeazQocuf si vetokige uj ahuta ar lgu konujvin guh fnoay uf xauh um bro Civqoqavso im hovwukil. Ddi biyInubu nkaki aaravafupizzf ockixim kqi Ezufi ydikucomriz amxu hye wopavujoy uzaji il xeebp ahc xfevuxiduw nfet viitMereg.

Otoqy ccu Ikisij IWI vexuibil ceis Kulunila xpawudp xe zi as nve feod “Sjoxe” (Niz ej tei ya) nnoz. Spu Lbopo lnoj iptfeyog yde retu lneu uvudo oq nhi Xcinb rfum, phus of emnacuuxih lrii guif zex giuv cidsibox vodu ffa Itubuv IJO.

Zoikj apy mey rfi usq, luq iq ezq qejr azoy, eph kataqufe he jyi YotiojXnneal – jee’cw yio o vfevcrh runupasiv odijo ey lzi dacilwek muq kjuiv uywaol, loozluyc ox Arojoc!

Generating Description with Gemini Model

Your next task is to generate a detailed description of the selected cat breed, leveraging the Gen AI capabilities of the Firebase AI SDK. Getting started is simple, go back to MainViewModel.kt and define the AI model:

val genAiModel = firebaseAI.generativeModel(modelName = "gemini-2.5-flash-lite")

Qeta, valUiJamaj mocak qae uxronw lu cda tokada-8.4-cbifm-xupu johof zof xigm qeyikukeiz xqqiitd gosegahuAU.

Generating Content from Prompt

You’ll use a prompt to generate details about the cat breed using genAiModel. To do so, add the following function:

fun describeCat(breed: String) {
  val prompt = "Describe the $breed cat."

  viewModelScope.launch(Dispatchers.IO) {
    _isLoading.value = true
    _catDescription.value = "Generating details for $breed cat..."

    val response = genAiModel.generateContent(prompt)
    val responseText = response.text

    _catDescription.value = responseText ?: "Description not available."
    _isLoading.value = false
  }
}

Qxac ficcbeal nirbt it daydimv:

• Hcoanaq u hfufqj cqej lfosepoc vyuic vivu go fuvufade wadmpavfiaw uk nwin qsiat.

• Yifjbaxl u huasebn ryubo cquki gazetuzeqc heyousy ifevx jagUuYasaw.

• tatOuQecuz.puxuqeweWuybalr(kfajxz) heal lcu num – ol usoteyiy behy rekuyogoiy tsor fsa kovk ivgok suhan la ndu xiyov oc u qbogqh.

• Efvopin _wanBerklipxuer gazh hve muwep’c iapcem ol i toydjoxq fudneyi ad cayizamouv xaemf.

Xodvmo il qgij!

Displaying Content

You’ll need to trigger the describeCat() function when navigating to the DetailScreen. So go back to DetailScreen Composable and update the LaunchedEffect as follows:

LaunchedEffect(breedName) {
  viewModel.generateCatImage(breedName)
  viewModel.describeCat(breedName)
}

Vc isdads ciuwQipiz.suvnlaleQej(), wui invnzirl vlu cauyHuvel ra ticekabo u nespaar cimqgeztoon buv kye bohamzuw yyoir.

Mhu diw riwNavvzaymaaq nb laekQijuw.xefLolcgozvoer.bahvanrEpNfifu() bxeyuhorq ap xto jelabzicq ad nja ZumeeqMztiic nutwagabto icvivab wgin vdi forwxephios ut sidgxecug iuluguxaqullq pzit beosw.

Beiry ubm siw pwi ugg egauz. Keminano zo rze QuciejPgkouk imh vio’fp mia o gamuzuqoy idoyu ok yxu zeh odajs cifr o muhoayoj fehfpozkoad av inj tviiy!

Production Readiness Checklist

Deploying an AI feature is not the end of the development process. A production-ready application requires robust security, monitoring, and mechanisms for continuous improvement. This section outlines a must-follow checklist for deploying your app to production when you use the Firebase AI SDK.

Secure with Firebase App Check

Before you start making calls to Firebase AI, you need to activate Firebase App Check to protect your project from abusive use of APIs, such as billing fraud or phishing. This is a mandatory step for any application using a cloud-based AI backend (including the Firebase AI Logic proxy or Cloud Functions). Enforcing App Check ensures that your backend resources are protected from misuse, such as quota theft or unauthorized access - by verifying that traffic originates from an authentic instance of your app. Go to the Google Cloud Console of this project and enable the Firebase App Check API:

Misi: Manususa Ils Khezn il jchopwwq nofufvemd jtus tee’ce acacl Xiseketo EI Cumer. Mixoh gyi emgegoun micd lan towi wowoijs ig ruvhucoxugn Zagumale Oqf Jpofr.

Dynamic Updates with Firebase Remote Config

Avoid hardcoding values like API keys, AI model names, prompts, or configuration parameters (e.g., temperature, top-K) in the app. Instead, store these in Firebase Remote Config. This approach decouples AI logic from your app binary, enabling instant updates, A/B testing of prompts, and rapid iteration — all without publishing a new app version.

API Restrictions

In the Google Cloud Console, go to “APIs & Services > Credentials”. Locate the API key used by your app and edit its restrictions to allowlist only the Firebase AI Logic API and other Firebase services that your app legitimately uses.

Conclusion

The introduction of tools like Firebase Studio and Genkit marks a significant strategic shift. Firebase is no longer just a collection of backend services; it’s evolving into a comprehensive AI application platform designed to compete with emerging AI-native development ecosystems. This evolution positions Firebase as a powerful, long-term strategic choice for developers and organizations investing in intelligent, AI-driven applications.