Building AI agents no longer means the cloud is mandatory. Models can now run directly in browsers and on mobile devices — no internet required, no per-request costs, full data privacy.

This talk explores on-device AI (Edge AI) as an emerging architectural pattern. We'll look at lightweight open models like Gemma, Llama, and DeepSeek that serve as the "brain" for offline-capable agents.

But this isn't just about simple inference. We'll cover full agent capabilities on-device:

On-device Function Calling: Teaching the model to interact with local APIs — contacts, calendar, sensors — without cloud roundtrips.

On-device RAG: Querying local data (documents, notes, emails) for context-aware answers that never leave the device.

Hybrid patterns: When to run on-device, when to use cloud, and how to combine both for the best of each — privacy and offline capability from edge, power and scale from cloud.

We'll discuss architecture decisions, practical implementation steps, and honest trade-offs: model size vs capability, performance vs privacy, and when each approach makes sense