The Future of Local "Agentic" AI: FunctionGemma on Android
The most interesting part of this recent experimental release isn't the chatbot—it's a 270M parameter model called FunctionGemma.
It’s specifically designed for "Mobile Actions" (local function calling). This means you can control Android OS features (flashlights, intents, system settings) using natural language entirely offline.
What makes this a "must-click" for developers:
- No API Keys: You aren't paying OpenAI or Google for tokens. The model lives in your phone's storage.
- The .litertlm Format: It uses a specialized lightweight runtime that targets the mobile CPU/GPU directly.
- Open Source Recipes: The dataset and "recipes" have been released on GitHub so you can fine-tune the 270M model for your own custom app functions.
It’s essentially a glimpse into how "Agentic" AI will work on-device without the latency or privacy concerns of a cloud round-trip.
Full setup guide and technical breakdown: https://blog.harislab.tech/3/now-you-can-control-your-phone-with-offline-ai-no-apis-no-cloud-no-latency
Top comments (4)
The minimum requirements for running AI Locally On your Android Phone is 4GB Ram With Minimum Android 12
You can smoothly run Gemma3-1B-IT AI Model on low end devices
But you can also run Gemma-3n-E2B-it AI Model Which is heavier than the Gemma3-1B-IT AI Model But Gemma-3n-E2B-it AI Model Support Image Analyzing Audio Scribe AI Chat And Prompt Lab
I tried on my phone and it works well but heavy models cause ui crashing and freezing issue on low end devices