Building Offline AI Agents in Your Mobile App
summary
Shows how full offline agents run on modern phones and browsers using Google’s Gemma family (including Gemma 3 and the tiny Gemma free nano), MediaPipe for on-device LLM inference, and the flutter_gemma plugin to hook models into Flutter UIs. Claims models as small as ~1B params quantize to ~500 MB and Gemma free runs on a single GPU; function calling + local embeddings (using on-device RAG with Embedding Gemma and a local vector store) enables agents that change UI, read images, translate menus and provide live captions entirely offline.
AI-written summary · about
chapters (16)
- 0:00Offline AI agents intro
- 1:56Why offline AI matters
- 7:17AI in mobile evolution
- 10:12Generative AI explained
- 13:09Open vs proprietary AI
- 17:30On-device AI with MediaPipe
- 21:29Flutter Gemma plugin
- 23:37Finding pre-trained models
- 25:27Fine-tuning models
- 27:12Multimodality on device
- 30:04Agentic behavior
- 32:56Choosing the right model
- 35:09RAG without cloud
- 38:32AI transparency
- 40:43Offline AI apps examples
- 47:06Q&A
related talks
21mFrom 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents
10mBuilding intelligent agents with Gemma 3
21mAdd Generative AI capabilities to your Android app
30mGenAI on Android with Gemini Nano and Firebase
22mClient side Web AI Agents: Building smarter user experiences for future agentic ...
05h 56mFrom the main stage to the terminal
31mBringing and Running AI Agents in the Browser
29mClient side Web AI Agents Building smarter user experiences for a future age