🚀 New Plugin: GemmaLLM – Run Gemma AI Models Locally in Your DroidScript Apps

[Jumar]

We're excited to introduce GemmaLLM, a new DroidScript plugin that lets you run Google Gemma language models completely offline on Android devices.

With GemmaLLM, you can integrate powerful on-device AI into your DroidScript applications without relying on cloud APIs or an internet connection.

Features

• Run Gemma language models locally on Android
• 100% offline inference – no internet required
• Fast on-device AI responses
• Build AI chatbots and assistants
• Ask questions about local documents (RAG support coming soon)
• Simple DroidScript API for easy integration

Example Use Cases

• Personal AI assistants
• Offline chatbots
• Document Q&A
• Note and knowledge assistants
• Educational apps
• Coding assistants
• Productivity tools
• AI-powered utilities

GemmaLLM is powered by Google's on-device AI technology, allowing your apps to keep user data on the device while reducing latency and eliminating API costs.

We're looking forward to seeing what the DroidScript community builds with it! Feel free to share your projects, feedback, and feature requests here and in Discord.

Drive link for the plugin:
https://drive.google.com/drive/folders/19SkVseKPN9SITN0QocgYCLaa82Z-tf48?usp=drive_link

The folder also contains a sample Gemma3-1B-it model downloaded from HuggingFace:
https://huggingface.co/litert-community/Gemma3-1B-IT/tree/main

You can also download .task model from HuggingFace.

[Remboz Zarzitoo]

I've tried this plugin, but the screen keep showing load model. How long it will take time to load?

I use model in your google drive and place it inside Download folder. Before load model I validate it if file is exist .

[Jumar]

Put the model inside your app or in your app's scope storage.

[Jumar]

For low-end devices, it takes around 3 or 4 seconds to load.

[Mikhael]

I downloaded the Gemma3-1B-IT_multi-prefill-seq_f32_ekv1280.task model

Size 4.01 GB.

But I can't load it into DroidScript through the built-in file manager. DroidScript throws an error

The error is likely due to the model being too large. Can large size models be used?

пятница, 24 июля 2026 г. в 12:48:14 UTC+5, Jumar:

[Jumar]

Try this smaller models first:
https://huggingface.co/litert-community/Gemma3-1B-IT/blob/main/Gemma3-1B-IT_multi-prefill-seq_q4_block128_ekv1280.task

[Mikhael]

After about 2 minutes of waiting I get an error





When I try to load a model from an external USB device or from Download folder, the loading message show endlessly (as in the message above)

пятница, 24 июля 2026 г. в 15:56:06 UTC+5, Mikhael:

[Mikhael]

Small model loads fine

[Jumar]

Try copying through adb or connecting your phone to your laptop. If you are on Mac you can use this MacOS app: https://android-file-transfer.macupdate.com/

You can use it to navigate to Android/data/* folders.

[Jumar]

I highly encouraged everyone to post in this thread if you find .task models form huggingface, that can be loaded successfuly in GemmaLLM plugin. Thanks

1. https://huggingface.co/litert-community/Gemma3-1B-IT/blob/main/Gemma3-1B-IT_multi-prefill-seq_q4_block128_ekv1280.task