Run LLMs Locally for Free Using Google’s Latest App!

Riya Bansal. Last Updated : 12 Jun, 2025
9 min read

The world of AI has just taken a gigantic leap forward by Edge Gallery Google. Just in the last week, Google quietly launched AI Edge Gallery, a democratizing application for AI. Google Edge AI enables the execution of powerful language models directly on our smartphones, eliminating dependency on the cloud and offering no subscription fees. It lets your device function as a powerful AI workstation without compromising personal privacy or data security. While this release marks the dawn of fully private and accessible AI, it comes with a lot of implications that go beyond just convenience factors. Let’s understand it better through this article.

Google AI Edge Gallery is an experimental application that transforms Android mobiles. The app acts as a bridge connecting users and Hugging Face Models. It allows for direct downloading and local execution of generative AI models, ensuring your assistant is there for you, entirely within your setup.

This platform removes the traditional walls between users and the best of AI technology. You no longer need the technical know-how or have language model capabilities at your disposal. You can now easily manage the difficult model juggling act with an easy-to-navigate user interface. Google’s Edge Gallery lets users try and test various AI models without requiring any external help or encountering restrictions. 

Source: Google

To put it simply, Edge Gallery is Google’s Initiative to bring AI to everyone, anywhere across the globe. The new update provides additional features and better compatibility with the models. Further, the experimental status enables rapid innovation and integration of user feedback.

Also Read: 5 Ways to Run LLMs Locally on a Computer

The standout features of Edge Gallery are:

  • Complete Offline Functionality: The most important feature is its complete independence from any form of connectivity after setup. Once the models are downloaded, your device becomes a standalone entity. All AI processing occurs locally, and no data is ever sent out to an external server.
  • Extensive Model Library Access: Google Edge Gallery is directly connected with Hugging Face’s huge model repository, so users can easily compare and download models suited for their tasks. The platform can perform text generation, image analysis, and code assistance. By allowing switching models, the platform encourages experimenting with different approaches and capabilities available in the AI world.
  • User-Friendly Experience Design: The application boasts a clean messaging-type interface familiar to its users. With all the power AI can give, navigation is kept fairly straightforward. Model management, settings, and conversations are all interwoven throughout the design. The platform is designed in such a way that even non-techies will be able to use it effectively from the get-go.
  • Maximum Privacy Protection: Since everything is processed locally, the threat of any data breach or unauthorized access is done away with. Your conversations never go to any external server or, for that matter, any third-party database. The completely private information is controlled by the user alone, always. Hence, it can claim better privacy protection than most of the cloud-based AI services can provide.

Benefits of Running LLM Locally

There are many LLMs available in the market, but why does this app stand out is because it allows us to run LLM locally, and that too in an offline way. Here are some benefits of running LLM locally:

  • Uncompromising Data Privacy: Local execution puts an end to personal information ever leaving the device boundary. Therefore, what one says in sensitive conversations, the details in the documents, and research queries or information remain entirely private and protected. Processing thus becomes a technical impossibility for corporate espionage or government surveillance. Your data stays in your complete custody without any third-party dependency.
  • Significant Cost Functionalities: Because of running the models locally, all subscription fees and pay-based pricing are eliminated. Thereafter, the downloaded model will work for you without considering any follow-up costs or payments. It essentially means unlimited access to AI without ever having to worry about financial constraints.
  • Performance Independence and Reliability: Problems with Internet connectivity do not allow for disruption in your AI workflow or task. Irrespective of network complexity, performance remains continuous to enable reliable AI assistance. Your processing power on a device defines how fast the response can be, rather than the quality of the connection. 
  • Unlimited Creative Freedom: Try a model for something different with no concern for usage limitations. Local execution facilitates unlimited creativity and searching without barriers. Users can push boundaries without someone else’s eyes monitoring or filtering their content.

Also Read: Top 12 Open Source Models on HuggingFace in 2025

Here are the steps to start experimenting with Edge Gallery. Using these steps, you can set up the app in your preferred system and start experimenting in no time.

Step 1: Check System Requirements

Currently, the application supports Android devices with fairly adequate processing power specifications. Minimum RAM of 4GB is advised to enjoy smooth functioning of the model and stability. There has to be at least 8GB of space available for free, where the downloads will be stored. Modern smartphones with 64-bit processors would offer ultimate user satisfaction.

Step 2: Check if Your Device is Supported

Android phones with version 8 or above are considered fully compatible. Flagship devices from reputed manufacturers offer the best level of performance and stability. Tablets with adequate specifications can also carry the application well. Compatibility of a device solely depends on RAM availability and processing power.

Step 3: Installation and Setup Process

Download the APK file from the official source or authorized app stores. This application will require an installation from unknown sources, so go to your Security settings and turn it on. Developer options are also required for the setup and configuration at the initial stage.

Grant all the permissions needed for smooth app operation and model administration. Proceed with the installation wizard for a smooth process of installation. The initial installation process depends on device performance and usually takes about 5-10 minutes. The first model download might take some time, depending on the model.

After installation of the Edge Gallery App, you’ll see an interface, so let’s see what we can do about it using some of its features:

Main Dashboard Navigation

The primary interface presents an assorted view of features that exist, such as Ask Image, Prompt Lab, and AI Chat. Below those features, model categories carry a label so that they may be quickly accessed and selected. Download statuses and model statuses appear on the main dashboard.

Quick action buttons allow for immediate access to a searched AI model. Storage usage indicators help users keep track of space on their devices. Shortcuts to settings also help in easy navigation to all customization options and preferences. The dashboard also updates dynamically depending on user and model activity.

Source: Google

Model Browser and Selection

As you can see in the interface, there are three to four model types for each feature from which you can choose. Filtering options help in narrowing down based on size, capability, and requirements. Model description provides detailed information about capability and performance characteristics. Previewing them allows users to test models before downloading them to consume their local storage.

You can also see ratings and community feedback, which helps users determine the model they can select. Popular ones are highlighted to assist users in finding the most commonly used ones. Advanced filtering options allow users to precisely match models for their requirements.

Interactive Chat Interface

It has a conversational interface familiar to users from any commercial messaging app for convenience. It supports typing questions, image uploads, and multi-turn conversations with real-time response generation as the user types. Contextual preservation offers multi-exchange and session flow, thereby maintaining conversation continuity.

It allows for model switching during conversation, which may be performed for comparative and testing purposes. Chat history is maintained on the local machine for reference and continuation. Exporting conversations can save essential chat content and AI-rendered content. It also supports voice input for hands-free interaction with AI models.

Comprehensive Settings Management

It offers customized model parameters, performance-related, and app-related preferences through interfaces. Manage downloaded models storage, usage, and update preferences using provided controls. Advanced users can access further customization to adjust model behavior and their response traits.

Privacy settings guarantee data handling up to personal requirements and standards. It also has performance-oriented settings that allow balancing between quickness and battery drain for use. Models and app updates can be managed automatically.

We have talked so much about the Edge Gallery, but now let’s see how it performs in action. The tasks using its standout features are as follows:-

Task 1: Document Text Extraction using Ask Image

The task demonstrates how the offline functionality of Edge Gallery can help in image analysis when provided a contextual prompt.

  1. Open the Ask Image feature in Edge Gallery
  2. Download your preferred model.
  3. For this task, we’ll be downloading ‘Gemma-2n-E2B-it-int4’.
  4. After downloading, click on ‘Try it’.
  5. Take a clear photograph of any handwritten note, receipt, or printed document.
  6. Upload the image to the selected vision model.
  7. Type the prompt “Extract all text from the image and summarize the key information”.
  8. Then, as the AI is taking its time to process the image and provide an appropriate response, if you’re not satisfied with the response, then you can ask it to make it better by providing detailed feedback.
  9. You can ask a follow-up question: “What are the main points given in the document/Image?”
  10. To test the offline feature, you can just disconnect from the internet and then ask these questions.
  11. Save the results for future reference.

Task 2: Professional Email Rewriting using Prompt Lab

The task demonstrates how Edge Gallery can support professional communication or provide results to our questions/prompts on a completely offline basis.

  1. Choose the Prompt Lab option from the three options on the dashboard.
  2. Choose your preferred text generation model from the available options and download it.
  3. For our task, we’ll be choosing ‘Gemma-3n-E2B-it-int4’ out of the four options, namely, ‘Gemma-3n-E4B-it-int4’, ‘Gemma3-1B-IT-q4’, ‘Qwen2.5-1.5B-Instruct q8’.
  4. After the model is downloaded, click on Try it.
  5. Write a simple, informal email draft (e.g., “Hey, can you send me that report? Thanks.”)
  6. Prompt: “Rewrite this email in a professional, polite tone: [paste your draft].”
  7. Generate the response and analyse it. If you’re not happy with the response then you can ask it to format it in a better and more structured way.
  8. You can experiment by entering the prompt: “Make it more formal and add a proper subject line suggestion.”
  9. Compare the result by choosing another model and entering the same prompts.
  10. Use the export function to store the best version.

Here are some of the advantages of using Google’s Edge Gallery:

  • Revolutionary Privacy Protection: It offers data sovereignty, which means the information will never leave a device boundary. Corporate and government surveillance will be made technically impossible to process. 
  • Zero ongoing Operational Costs: There are no subscription fees or contextual limitations or hidden costs attached to it after installation. Unlimited use of AI power becomes possible also without any additional financial obligations over time.
  • Full Network Independence: Be productive even under poor connectivity or coverage conditions. This means airplane mode compatibility, wherein uninterrupted AI assistance will be provided worldwide while traveling. 
  • Huge Variety and Flexibility of Models: Allowing access to thousands of Hugging Face models for different tasks. Easy switching of models depending on what each specific task requires or when a user prefers one over another. 

Whenever a new launch happens, it comes with lots of advantages over existing models, but there are some limitations to it as well. Here are some of the limitations of Edge Gallery:

  • Hardware Performance Dependencies: With power-hungry AI processing, older smartphones no longer stand a chance. Lesser processing power implies slower response times and significantly undermines the entire goodwill of the user.
  • Big Storage Needs: LFMs require adequate storage space on a device. The storage available on smaller devices gets quickly eaten away by multiple models. So users have to consciously allocate their storage for the best performance.
  • Model Compatibility Constraints: Not every Hugging Face model is compatible with the mobile platform. Some require more resources than a smartphone can practically provide. The sets offered are limited compared to cloud-based AI services.
  • Platform Limitation Availability: Currently limited to Android, with iOS coming, Apple users have to wait for the official release date of the iOS version. As of now, no official timeline has been confirmed for ios app to be available.
  • Slow Responses: As the model is running on the CPU, it is taking a huge amount of time as compared to LLM in providing the response to our prompts.

Also Read: How to Run LLM Models Locally with Ollama?

Comparison with Other Latest Local LLM Options

Let’s compare some of the most popular and recent Local LLMs available today. These platforms allow users to run powerful LLMs directly on their devices, but their features vary depending on the platform.

FeatureGoogle Edge GalleryOllamaLM Studio
Platform SupportAndroid (iOS Coming)Desktop/Server OnlyDesktop Only
Model RepositoryHugging Face DirectCustom/MultipleMultiple Sources
InstallationSimple APK InstallCommand Line SetupGUI Installer
Offline CapabilityFully OfflineFully OfflineFully Offline
Model ManagementEasy In-AppCommand BasedGUI Interface
Resource UsageMobile OptimizedHigh PerformanceHighly Configurable
User InterfaceMobile NativeTerminal/Web UIDesktop GUI
Model VarietyHugging Face SubsetExtensive LibraryWide Selection
PerformanceDevice DependentHardware OptimizedFully Customizable
Learning CurveBeginner FriendlyTechnical UsersModerate Difficulty
Community SupportGrowing RapidlyLarge CommunityActive Development
UpdatesAutomatic UpdatesManual UpdatesIntegrated Updates
CostCompletely FreeCompletely FreeCompletely Free

Conclusion

Google Edge Gallery represents a major shift in making AI more security-conscious. The experimental app takes working generative AI creations to its users. The approach works in unison to safeguard the privacy of users and yet provide them with state-of-the-art AI. The local processing removes barriers traditionally constructed between users and advanced technology.

While there are device compatibility and model selection-related limitations, the value-based advantages overshadow these limitations. This free-of-charge, privacy-centered solution enables everyone to access advanced AI. It is particularly useful for the educational sector, researchers, and privacy-conscious users alike. Such an approach means that developing regions have equal access to AI technology and infrastructure.

Gen AI Intern at Analytics Vidhya 
Department of Computer Science, Vellore Institute of Technology, Vellore, India 

I am currently working as a Gen AI Intern at Analytics Vidhya, where I contribute to innovative AI-driven solutions that empower businesses to leverage data effectively. As a final-year Computer Science student at Vellore Institute of Technology, I bring a solid foundation in software development, data analytics, and machine learning to my role. 

Feel free to connect with me at [email protected] 

Login to continue reading and enjoy expert-curated content.

Responses From Readers

Clear