# on-device Published articles for on-device. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Snap Announces New "anticipatory" AI Service & Apps for First Consumer 'Specs' AR Glasses DevFeed: [Snap Announces New "anticipatory" AI Service & Apps for First Consumer 'Specs' AR Glasses]() Original publisher: [Read original article]() Author: Scott Hayden Published: 2026-09-16T23:40:00Z Content type: news Language: en Sources: [Road to VR]() Topics: [Artificial Intelligence](), [AI Models](), [iOS](), [iphone](), [Cloud](), [Security](), [Streaming]() Tags: [ai](), [ai-models](), [cloud](), [ios](), [iphone](), [on-device](), [security](), [streaming](), [xr-industry-news]() ### AI overview Snap announced Specs Intelligence, an anticipatory AI service for its upcoming consumer Specs AR glasses. The service is designed to work across Specs, iPhone, and Mac, using connected apps and tools to build context and surface relevant information. Snap also announced AR experiences, streaming features, Spotify integration, and partnerships including HBO Max, the NBA, and the WNBA. ### Source excerpt Snap today announced new experiences, services and partnerships for SPECS, the company's upcoming pair of consumer AR glasses. Snap's big Specs livestream today wasn't technically a launch event--they're still slated to arrive in the US, UK and France later this fall starting at $2,195--although the company did give a little more insight into what sort [...] The post Snap Announces New "anticipatory" AI Service & Apps for First Consumer 'Specs' AR Glasses appeared first on Road to VR. ## On-Device AI Series (Part 5): LiteRT-LM DevFeed: [On-Device AI Series (Part 5): LiteRT-LM]() Original publisher: [Read original article]() Author: Oğuzhan Aslan Published: 2026-09-14T05:59:12Z Content type: tutorial Language: en Sources: [ProAndroidDev - Medium]() Topics: [LiteRT](), [On-device AI](), [Large Language Model](), [Inference](), [Streaming](), [GPU](), [LLM evaluation / benchmarking](), [PyTorch]() Tags: [ai](), [android](), [android-development](), [benchmarking](), [gpu](), [inference](), [litert](), [llama](), [llm](), [multimodal](), [on-device](), [on-device-ai](), [programming](), [pytorch](), [qwen](), [software-development](), [streaming]() ### AI overview This tutorial explains LiteRT-LM for running large language models on-device. It covers the Engine/Session API, streaming output, system prompts, tool calling, multimodal inputs, thinking mode, and CPU-versus-GPU benchmarking. The article also discusses tradeoffs involving privacy, network independence, latency, memory, sampling configuration, and model capability compared with cloud APIs. ### Source excerpt Put your phone in airplane mode. Open the app, type a question, and watch the answer arrive one token at a time -- no spinner waiting on a network round-trip, no API key, no per-token bill, and nothing you typed ever leaving the device. LiteRT-LM removes the genuinely hard parts of running an LLM on-device -- KV-cache management, token streaming, backend selection -- but it doesn't remove your job so much as relocate it. What's left on your plate is a short, specific list: sizing a combined input+output token budget, owning your own sampling defaults, hand-building system prompts and tool calling out of raw text, and one native-library collision that presents as a SIGSEGV rather than a build error. Know those going in and the API itself is a clean three-step pattern. We'll get there in that order: Why you'd choose this runtime and what it costs you versus the cloud. The Engine/Session model you need to read the code at all. Real implementation samples -- streaming, system prompts and tool calling, multimodal inputs, thinking mode, and CPU-vs-GPU benchmarking. The anti-patterns to avoid. A developer-friendliness rating on the same rubric as Parts 1-4. Why Use LiteRT-LM? You reach for LiteRT-LM instead of hand-rolling generation on top of raw LiteRT when: You need multi-turn conversation, not single-shot inference -- session state and KV-cache bookkeeping are handled for you, and resetting a conversation is a session swap, not a model reload. You need streaming output -- token-by-token delivery for a responsive chat UI, instead of a blocking call that returns everything at once. You're choosing between CPU and GPU per device -- the explicit backend parameter turns that into a runtime decision instead of a build-time guess. You want a pre-converted model without doing your own PyTorch-to-LiteRT conversion work -- the Model Zoo covers Gemma, Qwen, Llama, and more out of the box. You're willing to own sampling -- the engine won't pick sane decoding defaults for you; that's on the ## Mastering Edge AI on Raspberry Pi with LiteRT and Gemma DevFeed: [Mastering Edge AI on Raspberry Pi with LiteRT and Gemma]() Original publisher: [Read original article]() Author: Lu Wang; Terry Heo; Naushir Patuck; José María Casanova Published: 2026-09-12T11:04:33.891311Z Content type: tutorial Language: en Sources: [Google Developers Blog]() Topics: [LiteRT](), [On-device AI](), [Local AI](), [AI, ML & Data Engineering](), [Google AI](), [cpu](), [Security, Privacy and Abuse Prevention]() Tags: [cli](), [cpu](), [edge](), [gemma](), [gpu](), [inference](), [litert](), [offline](), [on-device](), [real-time](), [robotics]() ### AI overview The article explains how to deploy Gemma models with LiteRT on a Raspberry Pi for local, real-time edge AI applications such as robotics. It highlights LiteRT-LM, CPU and GPU optimization, and reported performance figures for Gemma 4 E2B on Raspberry Pi 5. ### Source excerpt Deploying secure, real-time Edge AI on Raspberry Pi is now simplified using LiteRT and lightweight Gemma open models. LiteRT optimizes CPU and GPU performance, delivering fast token speeds for models like Gemma4, enabling real-time local reasoning for robotics. Developers can quickly convert, quantize, and run these models using the lightweight LiteRT CLI tool. Support for Hailo AI accelerators is also coming very soon. ## Setup On Device Gemma 4 Android Development on Emulator DevFeed: [Setup On Device Gemma 4 Android Development on Emulator]() Original publisher: [Read original article]() Author: Elye - A Dev By Grace Published: 2026-09-11T14:27:55Z Content type: tutorial Language: en Sources: [Mobile App Development Publication - Medium]() Topics: [gemma4](), [Android](), [Emulator](), [android-development]() Tags: [android](), [android-app-development](), [android-development](), [androiddev](), [artificial-intelligence](), [emulator](), [gemma](), [gemma-4](), [google](), [mobile-app-development](), [on-device]() ### AI overview A tutorial explaining how to set up and run the Gemma 4 large language model locally on an Android Emulator, including downloading and testing the Google AI Edge Gallery sample app. It notes that emulator storage and memory may affect model use. ### Source excerpt Create Android Project with Downloadable Android On Device Gemma 4 Model Continue reading on Mobile App Development Publication " ## Android Studio Quail Using On Device Gemma 4 Experience DevFeed: [Android Studio Quail Using On Device Gemma 4 Experience]() Original publisher: [Read original article]() Author: Elye - A Dev By Grace Published: 2026-09-09T11:28:45Z Content type: tutorial Language: en Sources: [Mobile App Development Publication - Medium]() Topics: [Android Studio](), [gemma4](), [Large Language Model](), [Development]() Tags: [agent](), [android](), [android-development](), [android-studio](), [development](), [gemma](), [gemma-4](), [llm](), [local](), [memory](), [mobile-app-development](), [on-device]() ### AI overview This article describes using the on-device Gemma 4 model in Android Studio Quail for Android development. It outlines selecting the local model option, choosing Gemma in Model Providers, and downloading a model, while noting the need for a fast machine with substantial memory. ### Source excerpt How's Android Studio Quail support for On-Device LLM Model Doing? Continue reading on Mobile App Development Publication " ## Engineering log: fine-tuning Gemma 4 E4B with LoRA to bring FormAI's coaching on-device DevFeed: [Engineering log: fine-tuning Gemma 4 E4B with LoRA to bring FormAI's coaching on-device]() Original publisher: [Read original article]() Published: 2026-08-28T23:00:00Z Content type: article Language: en Sources: [John O'Reilly]() Topics: [gemma4](), [Fine-tuning](), [lora](), [On-device AI](), [multiplatform](), [LiteRT](), [Android](), [Artificial Intelligence](), [API]() Tags: [ai](), [android](), [api](), [data](), [data-quality](), [engineering](), [fine-tuning](), [gemini](), [gemma](), [gemma-4](), [inference](), [kotlin-multiplatform](), [litert](), [lora](), [multiplatform](), [on-device]() ### AI overview This engineering log describes a prototype that distils FormAI's Gemini-based sports coaching feedback into Gemma 4 E4B fine-tuned with LoRA. The local pipeline uses seed videos, filters inadequate critiques, extracts frames, trains and merges the adapter, then converts the model to LiteRT-LM for possible on-device Android inference. ### Source excerpt ⚠ Note: this post is AI-generated. The text below was written by Claude, and documents findings from a series of Claude Code sessions working on this project -- the experiments, bugs and measurements described are ones that came out of those sessions. The engineering work is real and the numbers were measured rather than estimated, but the write-up is the model's own account of what it did, not a human's independent retelling of it. ## CodeScanner.scan(): Barcode Scanning Without Rebuilding the Camera Pipeline DevFeed: [CodeScanner.scan(): Barcode Scanning Without Rebuilding the Camera Pipeline]() Original publisher: [Read original article]() Author: Shai Almog Published: 2026-08-26T00:00:00Z Content type: release Language: en Sources: [CodeName One]() Topics: [Barcode](), [QR Code](), [webcam](), [Code]() Tags: [camera](), [code](), [component](), [on-device](), [pipeline](), [vision]() ### AI overview Codename One adds CodeScanner and VisionCameraView, higher-level APIs for full-screen barcode scanning and embedded live vision analysis. The update also introduces typed results, coordinate helpers, image bridging, and analyzer-specific native dependency selection. ### Source excerpt CodeScanner and VisionCameraView put full-screen scanning and embedded live analysis above Codename One's on-device vision APIs, with typed results and simulator scripting. ## PROOF-Gen: From Optimized Data to Better Distillation DevFeed: [PROOF-Gen: From Optimized Data to Better Distillation]() Original publisher: [Read original article]() Published: 2026-08-26T00:00:00Z Content type: article Language: en Sources: [Apple Machine Learning Research]() Topics: [Fine-tuning](), [Prompt optimization](), [Artificial Intelligence](), [gemma4](), [Programming]() Tags: [agent](), [ai-agents](), [evaluation](), [fine-tuning](), [gemma](), [generate](), [models](), [on-device](), [post-training](), [prompt-optimization]() ### AI overview PROOF-Gen improves tool-calling model distillation by using per-scenario prompt optimization to recover successful trajectories from failed teacher attempts. The method strips corrective guidance before training, producing clean demonstrations and improving benchmark, deployed-pipeline, and on-device model performance. ### Source excerpt Supervised fine-tuning on teacher-generated trajectories is the standard first stage for distilling tool-calling capabilities into deployable models. Post-training pipelines that drive shipped tool-calling agents re-run this stage on a daily or weekly cadence, paying the frontier-teacher cost each cycle, yet the mechanism is generate-and-filter (keep the teacher's passing trajectories, discard the rest) and each cycle leaves behind the same hard scenarios because failures supply no signal. On τ 2-bench, 57% of teacher trials fail, two-thirds of them near-misses (most tool calls correct, undone... ## Up to 3.2x Faster Inference with LFM2.5-DSpark DevFeed: [Up to 3.2x Faster Inference with LFM2.5-DSpark]() Original publisher: [Read original article]() Author: Xx; Leonie Monigatti; Fernando Fernandes Neto; Tarek Dakhran; Nathan Ranchin Published: 2026-08-20T16:52:57Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Inference](), [Large Language Model](), [Latency](), [Streaming](), [sglang](), [GPU](), [Code]() Tags: [agentic](), [benchmark](), [code](), [gpu](), [inference](), [latency](), [llama](), [llama-cpp](), [llm](), [memory](), [on-device](), [streaming]() ### AI overview The article presents LFM2.5-DSpark, a speculative decoding approach that accelerates LLM inference by using a lightweight draft model and verifier. It reports up to 3.18x GPU throughput improvement and up to 2.87x on-device improvement, along with reduced function-calling latency and support for llama.cpp and SGLang. ### Source excerpt - Faster inference: up to 3.18 throughput improvement on a GPU and up to 2.87x on-device. - Toward on-device agentic inference: cuts function-calling latency by 57% on average for LFM2.5-2.6B - Day-one support for llama.cpp and SGLang: LFM-compatible DSpark integration is open-sourced upstream The decode phase in LLM inference is traditionally memory-bound. Most latency comes from streaming weights from DRAM into SRAM, not from intense computation. ## Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control DevFeed: [Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control]() Original publisher: [Read original article]() Author: Michelle Horton Published: 2026-08-19T16:00:00Z Content type: tutorial Language: en Sources: [NVIDIA Developer](), [NVIDIA Technical Blog]() Topics: [Cosmos](), [AI, ML & Data Engineering]() Tags: [agentic-ai-generative-ai](), [cosmos](), [deployment](), [edge](), [edge-computing](), [featured](), [foundation-models](), [inference](), [jetson](), [latency](), [nvidia](), [on-device](), [physical-ai](), [post-training](), [robotics](), [robotics-simulation](), [robots](), [simulation-modeling-design](), [thor](), [tutorial]() ### AI overview A tutorial on post-training NVIDIA Cosmos 3 Edge as an on-device robot manipulation policy, serving it on Jetson Thor, running receding-horizon inference, and evaluating it in closed-loop simulation. ### Source excerpt Robots need policies that can adapt to their sensors, environments, and tasks while running on onboard computing hardware. World models offer a foundation for... ## Scanning bus stop codes with ML Kit and Vision in the GalwayBus Compose Multiplatform app DevFeed: [Scanning bus stop codes with ML Kit and Vision in the GalwayBus Compose Multiplatform app]() Original publisher: [Read original article]() Published: 2026-08-06T23:00:00Z Content type: tutorial Language: en Sources: [John O'Reilly]() Topics: [compose-multiplatform](), [ML Kit](), [Android](), [cameraX](), [iOS](), [Code](), [ui]() Tags: [android](), [apple](), [camera](), [camerax](), [code](), [compose-multiplatform](), [ios](), [ml-kit](), [multiplatform](), [native](), [ocr](), [on-device](), [platform](), [recognition](), [ui]() ### AI overview This tutorial explains how the GalwayBus app scans six-digit bus stop codes using on-device ML Kit on Android and Apple's Vision framework on iOS. Camera preview, matching logic, and UI are shared through Compose Multiplatform, while platform-specific implementations handle camera access and OCR. Recognized six-digit runs are matched exactly to stops, and overlapping recognition requests are avoided by dropping newer frames. ### Source excerpt Every bus stop in Galway has a plate with a 6-digit stop code printed on it. We recently added a "Scan" tab to the GalwayBus app that lets you point the camera at that plate and jump straight to the stop's departures. The text recognition runs entirely on device (ML Kit on Android, Apple's Vision framework on iOS), with the camera preview, matching logic and UI all living in the shared Compose Multiplatform code. ## Codename One Adds On-Device AI and Loopback MCP Support DevFeed: [Codename One Adds On-Device AI and Loopback MCP Support]() Original publisher: [Read original article]() Author: Shai Almog Published: 2026-08-02T00:00:00Z Content type: release Language: en Sources: [CodeName One]() Topics: [On-device AI](), [Model Context Protocol](), [Artificial Intelligence](), [LiteRT](), [Inference](), [Mobile](), [supply-chain-security]() Tags: [applications](), [inference](), [litert](), [mcp](), [mobile](), [on-device](), [on-device-ai](), [tensorflow-lite]() ### AI overview Codename One has added on-device vision, language, and LiteRT inference to its core, alongside a guarded loopback MCP transport for inspecting and operating applications. The article explains platform support, asynchronous OCR, model sessions, secure model downloads, and privacy limitations of local inference. ### Source excerpt Codename One now exposes on-device vision, language, and LiteRT inference in the core, while a guarded loopback MCP transport lets an LLM inspect and drive real applications. ## Gemini Robotics 2 brings whole body intelligence to robots DevFeed: [Gemini Robotics 2 brings whole body intelligence to robots]() Original publisher: [Read original article]() Author: Carolina Parada Published: 2026-07-28T13:21:37Z Content type: article Language: en Sources: [Google DeepMind News]() Topics: [Robotics](), [AI Models](), [Artificial Intelligence](), [vlm](), [multimodal]() Tags: [agent](), [ai](), [ai-models](), [collaboration](), [gemini](), [models](), [multimodal](), [on-device](), [robotics](), [robots](), [teamwork](), [vision]() ### AI overview Gemini Robotics 2 is presented as an intelligence layer for adaptable robots, enabling whole-body control, dexterous manipulation, multi-robot teamwork, and adaptation to new robotic bodies. The article describes three models: a vision-language-action model for motor control, an embodied reasoning vision-language model for communication and multi-step planning, and an on-device model optimized for local operation. ### Source excerpt From feet to fingertips -- we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks. ## Build intelligent Android apps: On-device inference DevFeed: [Build intelligent Android apps: On-device inference]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: article Language: en Sources: [Android Developers Blog]() Topics: [Android](), [On-device AI](), [ML Kit](), [Inference](), [gemma4](), [Google](), [Mobile]() Tags: [android](), [efficiency](), [features](), [gemini](), [gemma](), [gemma-4](), [google](), [inference](), [ml-kit](), [on-device](), [speed]() ### AI overview This Android developer article explains how to use Gemini Nano through ML Kit's Prompt API to build on-device features in the Jetpacker demo app. It covers itinerary summarization, expense management, and voice notes, emphasizing local processing, privacy, offline reliability, and avoiding cloud inference costs. The article reports that prompt iteration reduced response time from 13 seconds to under 2 seconds. ### Source excerpt Posted by Caren Chang, Developer Relations Engineer, Android Developer Relations Welcome back to the blog post series "Build intelligent Android apps" where we take a basic Android app and transform it into a personalized, intelligent, and agentic experience. In our previous post we introduced Jetpacker, the demo app we'll use throughout this series. In this blog post, we will share how you can use Gemini Nano through ML Kit's Prompt API to build intelligent on-device features. Building intelligent on-device features refers to the ability to process prompts and data directly on a device without sending data to a server. This offers a few advantages: User data can be processed locally on the device, preserving user privacy Functionality of the model is reliable even with spotty or no internet connection No additional cloud inference cost, since everything runs on the user's hardware With the benefits of on-device in mind, we identified three features to add in Jetpacker that can improve the user experience: summarizing trip itineraries, managing expenses, and capturing voice notes. On-device features in Jetpacker: Summarizing trip itineraries, managing expenses, and voice notes High quality tailored summarization of short texts The itinerary screen gives users a quick overview of all activities for a given trip. Since this screen contains a lot of information, it can quickly become overwhelming. To help users prepare without feeling overwhelmed, we can add a 'Get ready for your trip' section at the top. The romantic Paris trip is summarized as a classic Parisian adventure blending art, sights, and delicious food. A tip and some useful phrases are also added. By inputting a trip itinerary and asking an LLM to summarize it, we can generate a quick summary of the trip along with packing tips and useful local phrases. This is a great use case for an on-device model for several reasons: Performance and quality: Both the input and output text are relatively short. With tha ## Build intelligent Android apps: Cloud and hybrid inference DevFeed: [Build intelligent Android apps: Cloud and hybrid inference]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: tutorial Language: en Sources: [Android Developers Blog]() Topics: [Android](), [Firebase](), [Artificial Intelligence](), [Inference](), [Cloud](), [ML Kit]() Tags: [ai](), [ai-models](), [android](), [cloud](), [firebase](), [firebase-ai-logic](), [gemini](), [grounding](), [hybrid-inference](), [inference](), [on-device]() ### AI overview This tutorial explains how to build cloud-hosted and hybrid AI features in Android apps with Firebase AI Logic. Using the Jetpacker app as an example, it covers web-grounded museum assistance, hybrid restaurant review drafting with Gemini Nano and cloud fallback, and custom-routed live translation for hotel support. ### Source excerpt Posted by Thomas Ezan, Jolanda Verhoef, Caren Chang, Senior Developer Relations Engineers, Android Developer Relations Welcome back to the blog post series "Build intelligent Android apps" where we take a basic Android app and transform it into a personalized, intelligent, and agentic experience. In our previous post we explored how to build intelligent on-device features using Gemini Nano through ML Kit's Prompt API. In this post, we will look at how you can leverage Firebase AI Logic to build cloud-hosted and hybrid AI features: Grounding answers in real-world context Routing requests dynamically between cloud and local execution using hybrid inference Translating content with custom routing systems Sometimes a use case requires AI models with greater world knowledge, a much larger context window, or the ability to handle complex queries. In those scenarios, we can leverage cloud models. Other times, you want the best of both worlds: using hybrid inference to run on-device when available to lower costs, while falling back to the cloud to ensure compatibility for all devices. Cloud and hybrid features in Jetpacker: Museum assistant with web grounding, hybrid restaurant review drafting, and support chat featuring custom-routed live translation. Let's look at how we implemented three cloud and hybrid features in Jetpacker: a museum assistant with web grounding hybrid restaurant review drafting hotel support chat featuring custom-routed live translation. Use LLM grounding for up-to-date informationMuseum assistant chatbot with LLM grounding The Museum assistant is an interactive chatbot designed to help users plan their museum visits. It provides visitors with up-to-date details regarding specific exhibits, current opening hours, ticket pricing, and more. Museum assistant is a chatbot that answers questions, such as 'How can I get a ticket discount for Le Louvre?' When building AI features, getting the model to answer with fresh, accurate, and specific real-world info ## Build intelligent Android apps: Introduction to Jetpacker DevFeed: [Build intelligent Android apps: Introduction to Jetpacker]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: tutorial Language: en Sources: [Android Developers Blog]() Topics: [android-development](), [Mobile](), [On-device AI](), [Inference](), [genai](), [Open Source](), [App](), [ui](), [Google]() Tags: [android](), [android-development](), [apps](), [genai](), [google](), [inference](), [on-device](), [open-source](), [ui]() ### AI overview This introduction to a technical blog series presents Jetpacker, an open-source Android showcase app built for Google I/O. It outlines choices involving on-device, cloud, and hybrid inference, Android system integration, and agentic flows, with later posts promising implementation guidance and code examples. ### Source excerpt Posted by Jolanda Verhoef, Senior Developer Relations Engineer, Android Developer Relations Building GenAI features in your app usually means navigating through various models, APIs and architecture choices: Execution location: Where does your model run? On device, in the cloud, or both? Complexity: How complex is your setup? Are you doing a single inference call or do you need a more agentic flow? In-app or Android System: Should your feature be built into your Android app or does it fit better as an Android system integration? In this blog post series we'll navigate these choices with you. We will take you along on a journey, starting with a basic mobile app and transforming it into a personalized, intelligent, and agentic experience. Jetpacker: a demo travel app Jetpacker is a technical showcase app that our team built from the ground up for this year's Google I/O (built using Antigravity). At its core, Jetpacker helps users plan, explore, and enjoy their next big adventure. It shows an overview of your trips, the itinerary of each trip, and details of each event on that trip. Of course following all best practices of Android development, including a beautifully expressive Material UI design. And best of all? It's fully open source! Today we are publishing a series of technical blog posts diving deep into each of these features. We'll provide detailed implementation steps, code snippets, and architectural insights to help you build your own intelligent Android applications. On-device intelligence On-device features in Jetpacker: Summarizing trip itineraries, managing expenses, and voice notes Using an on-device model comes with no additional cloud inference costs, means you don't have to worry about internet connectivity, and lets users be confident that private information will be processed locally, on the device, without any of their data being sent to the cloud. In Jetpacker, we chose on-device inference for three of our features: The trip overview feature tra ## Build intelligent Android apps: Cloud and hybrid inference DevFeed: [Build intelligent Android apps: Cloud and hybrid inference]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: tutorial Language: en Sources: [Android Developers Blog](), [Android Developers Blog]() Topics: [AI Chat]() Tags: [ai](), [ai-models](), [android](), [api](), [apps](), [cloud](), [firebase](), [gemini](), [how-to](), [inference](), [llm](), [on-device](), [routing](), [web]() ### AI overview A tutorial on building Android AI features with Firebase AI Logic, including web-grounded chat, hybrid on-device/cloud review generation, and custom-routed translation. ### Source excerpt Posted by Thomas Ezan, Jolanda Verhoef, Caren Chang, Senior Developer Relations Engineers, Android Developer Relations Welcome back to the blog post series "Build intelligent Android apps" where we take a basic Android app and transform it into a personalized, intelligent, and agentic experience. In our previous post we explored how to build intelligent on-device features using Gemini Nano through ML Kit's Prompt API. In this post, we will look at how you can leverage Firebase AI Logic to build cloud-hosted and hybrid AI features: Grounding answers in real-world context Routing requests dynamically between cloud and local execution using hybrid inference Translating content with custom routing systems Sometimes a use case requires AI models with greater world knowledge, a much larger context window, or the ability to handle complex queries. In those scenarios, we can leverage cloud models. Other times, you want the best of both worlds: using hybrid inference to run on-device when available to lower costs, while falling back to the cloud to ensure compatibility for all devices. Cloud and hybrid features in Jetpacker: Museum assistant with web grounding, hybrid restaurant review drafting, and support chat featuring custom-routed live translation. Let's look at how we implemented three cloud and hybrid features in Jetpacker: a museum assistant with web grounding hybrid restaurant review drafting hotel support chat featuring custom-routed live translation. Use LLM grounding for up-to-date informationMuseum assistant chatbot with LLM grounding The Museum assistant is an interactive chatbot designed to help users plan their museum visits. It provides visitors with up-to-date details regarding specific exhibits, current opening hours, ticket pricing, and more. Museum assistant is a chatbot that answers questions, such as 'How can I get a ticket discount for Le Louvre?' When building AI features, getting the model to answer with fresh, accurate, and specific real-world info ## Build intelligent Android apps: Introduction to Jetpacker DevFeed: [Build intelligent Android apps: Introduction to Jetpacker]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: article Language: en Sources: [Android Developers Blog](), [Android Developers Blog]() Topics: [AI Bots](), [User interface design]() Tags: [agentic](), [android](), [apis](), [cloud](), [genai](), [inference](), [on-device](), [orchestration]() ### AI overview An introduction to the Jetpacker technical showcase and a blog series about building intelligent Android applications with on-device, cloud, hybrid, and agentic AI approaches. ### Source excerpt Posted by Jolanda Verhoef, Senior Developer Relations Engineer, Android Developer Relations Building GenAI features in your app usually means navigating through various models, APIs and architecture choices: Execution location: Where does your model run? On device, in the cloud, or both? Complexity: How complex is your setup? Are you doing a single inference call or do you need a more agentic flow? In-app or Android System: Should your feature be built into your Android app or does it fit better as an Android system integration? In this blog post series we'll navigate these choices with you. We will take you along on a journey, starting with a basic mobile app and transforming it into a personalized, intelligent, and agentic experience. Jetpacker: a demo travel app Jetpacker is a technical showcase app that our team built from the ground up for this year's Google I/O (built using Antigravity). At its core, Jetpacker helps users plan, explore, and enjoy their next big adventure. It shows an overview of your trips, the itinerary of each trip, and details of each event on that trip. Of course following all best practices of Android development, including a beautifully expressive Material UI design. And best of all? It's fully open source! Today we are publishing a series of technical blog posts diving deep into each of these features. We'll provide detailed implementation steps, code snippets, and architectural insights to help you build your own intelligent Android applications. On-device intelligence On-device features in Jetpacker: Summarizing trip itineraries, managing expenses, and voice notes Using an on-device model comes with no additional cloud inference costs, means you don't have to worry about internet connectivity, and lets users be confident that private information will be processed locally, on the device, without any of their data being sent to the cloud. In Jetpacker, we chose on-device inference for three of our features: The trip overview feature tra ## Build intelligent Android apps: On-device inference DevFeed: [Build intelligent Android apps: On-device inference]() Original publisher: [Read original article]() Author: Android Developers (noreply@blogger.com) Published: 2026-07-21T13:00:00Z Content type: article Language: en Sources: [Android Developers Blog](), [Android Developers Blog]() Topics: [AI Chat]() Tags: [android](), [api](), [gemini](), [inference](), [llm](), [model](), [on-device](), [performance](), [privacy]() ### AI overview The article explains how to use Gemini Nano through ML Kit's Prompt API to add on-device LLM features to an Android travel app. Its example summarizes itineraries and generates preparation tips and local phrases while processing data locally. ### Source excerpt Posted by Caren Chang, Developer Relations Engineer, Android Developer Relations Welcome back to the blog post series "Build intelligent Android apps" where we take a basic Android app and transform it into a personalized, intelligent, and agentic experience. In our previous post we introduced Jetpacker, the demo app we'll use throughout this series. In this blog post, we will share how you can use Gemini Nano through ML Kit's Prompt API to build intelligent on-device features. Building intelligent on-device features refers to the ability to process prompts and data directly on a device without sending data to a server. This offers a few advantages: User data can be processed locally on the device, preserving user privacy Functionality of the model is reliable even with spotty or no internet connection No additional cloud inference cost, since everything runs on the user's hardware With the benefits of on-device in mind, we identified three features to add in Jetpacker that can improve the user experience: summarizing trip itineraries, managing expenses, and capturing voice notes. On-device features in Jetpacker: Summarizing trip itineraries, managing expenses, and voice notes High quality tailored summarization of short texts The itinerary screen gives users a quick overview of all activities for a given trip. Since this screen contains a lot of information, it can quickly become overwhelming. To help users prepare without feeling overwhelmed, we can add a 'Get ready for your trip' section at the top. The romantic Paris trip is summarized as a classic Parisian adventure blending art, sights, and delicious food. A tip and some useful phrases are also added. By inputting a trip itinerary and asking an LLM to summarize it, we can generate a quick summary of the trip along with packing tips and useful local phrases. This is a great use case for an on-device model for several reasons: Performance and quality: Both the input and output text are relatively short. With tha ## Meeting Notes - A Free Desktop App for Tracking 1:1s DevFeed: [Meeting Notes - A Free Desktop App for Tracking 1:1s]() Original publisher: [Read original article]() Author: Dean Hume Published: 2026-06-29T10:16:48Z Content type: article Language: en Sources: [Dean Hume]() Topics: [App](), [meetings](), [Local-First](), [asr](), [Whisper](), [Open Source](), [Markdown]() Tags: [app](), [autosave](), [desktop](), [free](), [meetings](), [notes](), [offline](), [on-device](), [open-source](), [recording](), [speech-to-text](), [technical-leadership](), [technical-program-manager](), [whisper](), [writing]() ### AI overview The article introduces Meeting Notes, a free, open-source Windows desktop app for organizing 1:1 notes by person. It supports topic tags, Markdown, autosave, voice recording, and offline speech-to-text using Whisper, with data and audio kept on the device. ### Source excerpt Meeting Notes is a free offline desktop app for 1:1s. Organise notes by person, tag topics, and transcribe meetings on-device with Whisper. ## Accelerating Gemini Nano models on Pixel with frozen Multi-Token Prediction DevFeed: [Accelerating Gemini Nano models on Pixel with frozen Multi-Token Prediction]() Original publisher: [Read original article]() Published: 2026-06-26T18:30:00Z Content type: article Language: en Sources: [The latest research from Google]() Topics: [AI Inference](), [On-device AI](), [Google](), [LLMs](), [gemma]() Tags: [ai](), [efficiency](), [energy](), [gemini](), [inference](), [machine-intelligence](), [mobile-systems](), [natural-language-processing](), [on-device](), [on-device-ai](), [phones]() ### AI overview Google Research describes a method for retrofitting Multi-Token Prediction onto frozen Gemini Nano v3 production models to accelerate on-device inference on Pixel phones. The approach targets mobile energy and memory constraints, improving the speed and energy efficiency of features such as notification summaries and proofreading without requiring separate drafting models. ### Source excerpt Machine Intelligence ## Experimenting with the proposed Cross-Origin Storage API in Transformers.js DevFeed: [Experimenting with the proposed Cross-Origin Storage API in Transformers.js]() Original publisher: [Read original article]() Author: Thomas Steiner Published: 2026-06-23T00:00:00Z Content type: tutorial Language: en Sources: [Hugging Face - Blog]() Topics: [AI, ML & Data Engineering]() Tags: [ai-models](), [announcement](), [api](), [artificial-intelligence](), [asr](), [browser](), [cache](), [chrome](), [chrome-extension](), [cos](), [cross-origin-storage](), [inference](), [javascript](), [on-device](), [open-source](), [storage](), [transformers](), [transformers-js](), [wasm](), [web](), [web-apps](), [web-developers](), [web-standards](), [webassembly]() ### AI overview A tutorial that uses Transformers.js browser inference examples to examine duplicate model and WebAssembly downloads across origins and the proposed Cross-Origin Storage API. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Why I Think On-Device AI Changes Mobile Architecture DevFeed: [Why I Think On-Device AI Changes Mobile Architecture]() Original publisher: [Read original article]() Author: Akansha Jain Published: 2026-06-03T17:56:53Z Content type: opinion Language: en Sources: [Flutter Community - Medium]() Topics: [Artificial Intelligence](), [On-device AI](), [Mobile](), [Flutter](), [Android](), [data]() Tags: [ai](), [android](), [api](), [artificial-intelligence](), [dart](), [data](), [flutter](), [llm](), [mobile](), [on-device](), [on-device-ai]() ### AI overview This opinion article argues that on-device AI changes mobile application architecture by reducing dependence on network APIs. It discusses connectivity, per-call costs, and user-data concerns, drawing on a Flutter camera-demo crash involving a model already loaded on an Android phone. ### Source excerpt What a crashed app taught me about on-device AI -- and why I think you should care. I was building the demo I'd been planning for weeks. Flutter app running. Model loaded. Chat screen looking clean. I opened the camera -- the feature I was most excited about -- pointed it at something on my desk, and the app just crashed. Not a Dart exception. Not a red error screen. The whole thing just died. I stared at my Redmi Note 7 Pro for a second, opened it again, tried once more. Same result. Every single time I launched the camera with the model already loaded, Android killed my app. It took me longer than I'd like to admit to understand why. And when I finally did, I realised I'd been thinking about AI in mobile apps in a way that was quietly wrong for a long time. Before that, let me back up. For the past couple of years, whenever I wanted to add an AI feature to an app, the answer was pretty obvious: use an API. Need a chatbot? API. Need to analyze an image? API. Need text summarisation? API. A few lines of Dart, a request goes out, and something surprisingly intelligent comes back. It worked well enough that I stopped questioning the architecture behind it. Every one of those API calls travels over the network, gets processed on hardware you don't own, and comes back. Which means three things are always true, and always have been: The feature needs internet. Which is fine until you're in a metro tunnel, on a flight, or somewhere with patchy coverage. Your app just stops being smart. Users notice -- and they don't file bug reports, they just leave. You pay for every call. At low volume it's fine. At scale, it genuinely isn't. 10,000 users x 50 queries a day x per-token pricing. At low volume you barely notice it. At scale, those costs become a product decision. User data leaves the device. The health app where someone is logging symptoms. The finance app where they're describing their expenses. The journal app where they're writing things they wouldn't say out loud. For tho ## Holo3.1: Fast & Local Computer Use Agents DevFeed: [Holo3.1: Fast & Local Computer Use Agents]() Original publisher: [Read original article]() Author: Maxime Langevin; Hamza Benchekroun; Axel Moyal; Emrick Sinitambirivoutin; Antonio Loison; Avshalom Manevich; Tony Wu; Pierre-Louis Cedoz; Aurélien Lac; Ronan Riochet Published: 2026-06-02T14:13:23Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [computer-use](), [On-device AI](), [Local AI](), [Deployment](), [model-deployment](), [qwen](), [Benchmark](), [Cloud](), [NVFP4](), [browser]() Tags: [agent](), [agents](), [benchmark](), [benchmarks](), [browser](), [cloud](), [computer-use](), [deployment](), [devices](), [inference](), [json](), [model](), [models](), [nvfp4](), [on-device](), [performance](), [qwen](), [software]() ### AI overview Holo3.1 is a family of computer-use models designed to operate across web, desktop, and mobile environments and integrate with different agent frameworks. The release adds quantized checkpoints for local inference, native function-calling support, and model sizes ranging from 0.8B to 35B-A3B, targeting private, cost-effective, and high-performance deployments. ### Source excerpt Users want to run the same computer-use capabilities across desktop and mobile environments, with seamless integration with different agent frameworks. They want deployment flexibility, from cloud inference to fully local execution on end-user devices. This is why we are releasing the Holo3.1 family. Holo3.1 improves robustness across the three dimensions that matter most in production: environments (web, desktop, mobile), agent frameworks, and deployment targets. [Next page]()