Create agent skills for on-device generative AI
summary
Live demo shows on‑device LLMs using Google AI Edge and LiteRT‑LM APIs to compile, quantize, and run models (Gemini 2D) inside the AI Edge gallery app. The workshop builds a Task planner and timer skill as a directory with metadata, a skill.markup instruction, a deterministic script that parses JSON and issues function calls returning task_id + minutes, and an HTML WebView; timers render and function on the device, proving the end‑to‑end flow. LiteRT‑LM APIs AI Edge gallery app
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:55Google AI Edge stack
- 3:43Lite TLM APIs deep dive
- 5:50Agent skills benefits
- 7:14Agent skill architecture
- 8:39Live demo: task planner skill
- 10:25Workshop setup
- 11:55Skill creation in Colab
- 13:42Workshop recap
related talks
01h 34mHands-on AI workshop: Multimodal agents
01h 21mTLMs: Tiny LLMs and Agents on Edge Devices with LiteRT-LM
22mAI-Powered Creative Experiences
44mHands-on-lab: It's all about going hands-on with Generative AI!
42mOn-Device Machine Learning with Google AI Edge
21mUnlocking Client-Side AI in Your Web App
21mFrom 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents
41mBuild intelligent Android apps with Google's AI