This experimental application, available now on Android and soon on iOS, allows you to run large language models (LLM) directly on your smartphone, in offline mode, thus ensuring greater confidentiality of your exchanges.
AI at your fingertips, far from servers
One of the main attractions of AI Edge Gallery lies in its ability to free the user from the constraints of remote servers. No more having to route your requests through online platforms, which, by their very nature, raise questions about the management of your personal information. With this new application, data processing is entirely carried out locally on your device, offering greater control over the confidentiality of your interactions with AI.
Live on GitHub, AI Edge Gallery is much more than just an offline chatbot. Google has designed a turnkey solution, allowing you to download and use different language models directly on your smartphone. The application already offers four default models, including three variations of Google's recent Gemma-3 and a Qwen model from Alibaba, with the possibility of importing others from platforms like Hugging Face.
A multifaceted application
The application's interface is divided into three main functions. The first mode, Ask Image, allows you to analyze an image and obtain relevant information using AI, without the information leaving your phone. It could be a plant, a monument, or even a dish.
With the Prompt Lab mode, we find solutions more suited to creatives and developers, such as the ability to synthesize or rewrite text, and even generate computer code. A sort of toolbox for text and development projects. Finally, AI Chat is positioned in the same vein as ChatGPT or Gemini by offering a traditional chatbot experience. This mode allows you to communicate with the AI, ask questions, and get instant answers, all without an internet connection.
Limitations to be aware of
Running language models locally on a smartphone has obvious advantages, but there are also disadvantages and limitations to consider. Indeed, the response time to a query will depend heavily on the smartphone model used, with some devices being faster than others depending on their configuration. It is also important to consider the weight of these models, often measured in gigabytes, which will require sufficient storage space on your device. According to our findings, the size varies between 554.7 MB and 4.4 GB, with several models easily exceeding 1 GB.
However, Google has optimized these models for mobile use. Unlike the colossal versions running in data centers, these lightweight models have fewer parameters, making them compatible with the ARM chips in our smartphones. The trade-off? Sometimes less nuanced or creative responses than their cloud counterparts, but unmatched portability and privacy.
A promising and completely free alpha version
Currently in "alpha" version and available via an APK download on GitHub, AI Edge Gallery is not yet available on the Play Store. This approach allows Google to gather feedback from "early adopters" before a general release. It is aimed more at an informed audience, even if manual installation on an Android smartphone remains simple to perform.
Despite some current limitations, such as the lack of real-time voice interaction, the potential of this app for privacy-focused offline AI experimentation is undeniable. An iPhone version is also planned, but its release date remains unknown.
0 Comments