I put a Gemini alternative on my Android phone that doesn’t need the internet

Released Oct 30, 2026, 4:00 PM EDT

Tashreef’s fascination with customer innovation started in the school library when he came across a tech publication, CHIP, which eventually influenced him to pursue a degree in Computer Science. Because 2012, Tashreef has actually expertly authored over a thousand how-to posts, adding to Windows Report and How-To Geek. He presently focuses on Microsoft Windows material at MakeUseOf, which he has actually been utilizing considering that 2007.

With hands-on experience structure sites and innovation blog sites, he brings useful designer insights to his technical writing. You can see his total work portfolio at itashreef.com.

You may likewise come across his brief how-to video explainers, streamlining intricate subjects. Beyond composing, Tashreef delights in developing brief explainer videos, video gaming, and checking out animated programs.

Gemini on Android works terrific. It can address concerns, read your screen, and even assist you get more out of functions you didn’t understand existed. It requires the web to do any of that. What if you’re on a flight, in a remote location, or simply wish to keep your discussions personal and off another person’s servers? A regional little language design can deal with that, which’s precisely what I attempted.

I downloaded PocketPalfilled an offline little language design, and checked it versus Gemini to see how far it might go. For the a lot of part, it succeeded at fast lookups and fundamental jobs that didn’t need web, though it was visibly slower.

PocketPal brings regional AI designs to your phone

An open-source app that runs language designs without the cloud

Credit: Tashreef Shareef/ MakeUseOf

Android has several methods to run regional AI designs. MLC Chat and Google AI Edge Gallery are 2 popular alternatives, however PocketPal makes it the simplest to start with regional AI on your phone. It’s an open-source app for Android and iOS, and it does not require the web other than to download the designs.

As soon as a design is on your phone, whatever runs in your area. Chats remain on your gadget together with your complete chat history, and you can modify messages and retry actions if the output isn’t what you desired. There’s a voice assistant alternative too, however that’s a paid function, and I’ll get to that later on.

PocketPal likewise has actually a function called Buddieswhich are conserved personalities. Consider them as system triggers bundled with a particular design and additional settings, so you do not need to retype directions each time you begin a brand-new chat.

Utilizing PocketPal on Android

How to set up, download designs, and get talking

You can begin by downloading the app from the Google Play Store. After a brief walkthrough, tap Download ModelBy default, it reveals 3 designs and suggests Well balancedwhich is Gemma 3 1B at around 1,020 MB. You can likewise select a smaller sized design like LFM2.3.1 2B at 700MB or opt for the very best choice, Qwen3.5 3B, at 2.5 GB.

I began with the advised Gemma 3 1B, as it promised to work great on a lot of mobile phones. For recommendation, I checked these on my Galaxy Z Flip 6 with 12GB of RAM and a Note 20 Ultra with comparable memory however an older processor. If your phone has more RAM, do not hesitate to choose the larger designs.

After the download surfaces, you require to pack the design from the chat menu. Tap the up arrow icon, open the Design tab, and select your design. You can likewise do this from the three-dot menu>> Models or the Designs tab in the hamburger menu.

Talking with PocketPal

How the regional designs managed 8 standard triggers

Talking with PocketPal seems like any other chatbot, whether that’s Gemini, Claude, or ChatGPT, other than the reactions are available in slower. That’s not unexpected when the processing takes place totally on your phone’s hardware rather of an information.

I ran a standard test with 3 designs: Gemma 3 1B (1.07 GB), Gemma 3 4B (2.36 GB), and Gemma 4 E2B (4.03 GB). Each design got the very same 8 triggers in a fresh chat with default settings. The triggers covered basic understanding, mathematics, reasoning, summing up, coding, and a purposeful technique concern to check for hallucination.

Here’s how they scored:

All 3 designs nailed the essentials. They properly determined Canberra as Australia’s capital, determined a 25% discount rate on 800 rupees, fixed a transitive reasoning puzzle, and composed a working Python function. They even counted the letter “r” in “strawberry” properly, which is a timely that journeys up some cloud designs.

The one trigger that tripped all 3 was a trap: I asked to sum up a speech Albert Einstein provided to Congress in 1987. Einstein passed away in 1955, so no such speech exists. Every design with confidence composed a plausible-sounding summary rather of flagging the incorrect facility. That’s the primary dependability threat with regional designs, and larger design size didn’t assist.

Speed-wise, the 1B design was the fastest and most responsive, that makes it a great default for fast concerns. The E2B design utilizes a “Think” action before addressing, that made some brief triggers take 30 to 40 seconds regardless of faster raw token speeds. For casual chat, that’s a genuine tradeoff.

Personalizing PocketPal

Buddies, generation settings, and tweak how the design reacts

Proceeding, we have FriendsA Pal is a conserved package of a design, a system timely, and a name, so you do not need to retype guidelines for each chat. By default, PocketPal includes 2 pre-installed Pals. Is Pipa general-purpose assistant that works on Gemma 3 1B, suitable for brief triggers. Second is Lookiea real-time video assistant that explains your electronic camera feed utilizing SmolVLM-500M, getting a frame every 3,000 ms.

You can find more Pals produced by the neighborhood from palshub.aithough you require to visit to conserve them to your own library.

In the settings, you can set up context size, advanced design choices, and memory management. There’s even an alternative to let PocketPal carry out a web search, though it’s switched off by default.

To the right of the chat, tap the three-dot menu>> Generation settingsThis is where you manage how the design composes its reply. You can toggle whether a thinking design’s thinking text remains in the history, change Temperature level Leading K Leading Pand Minutes P to manage just how much the AI stays with safe word options or takes more imaginative threats, modify repeating charges to assist it break out of loops, and set N Predict to top the reply length so a rambling design does not fill your whole context window with unimportant text.

PocketPal is a strong offline buddy, however it’s not changing Gemini

PocketPal likewise has a brand-new voice discussion function that lets you talk hands-free. It’s a Plus-only function, however, so you ‘d need to invest a minimum of $5 to open it. That stated, while optional, it’s a helpful addition if you discover yourself depending on PocketPal routinely.

Now, PocketPal isn’t a direct competitor to Gemini. Gemini can do even more than simply address text triggers: it reads your screen, works throughout Google apps, automates multi-step jobs, and pulls from real-time web information. PocketPal can’t do any of that. What it can do is provide you a personal AI assistant that works without the cloud, operates on hardware you currently own, and does not cost a month-to-month membership. You are not training an LLM with your information.

Read More