Many mobile AI features assume constant connectivity. This episode starts from a different assumption: real users lose network access all the time.

Using a health app example, the video shows what happens when a user opens an app underground, without signal, and asks a question. With a cloud-based API, the request never completes. With on-device AI, the response starts streaming immediately because the model runs on the phone itself.

From there, the episode explains why Callstack built React Native AI, what problems it is meant to solve, and how it fits into real-world mobile constraints around privacy, offline usage, reliability, and cost.

When cloud AI fails in mobile environments

The opening scenario focuses on a common mobile case: a user on a morning commute with no network access. The transcript highlights the dependency chain behind cloud AI calls, including DNS resolution, HTTP handshakes, and server queues. When connectivity drops, the app shows a loading spinner and never produces an answer.

This section establishes the baseline problem React Native AI is designed to address.

Running inference directly on the device

The video contrasts cloud calls with local inference. With on-device AI, responses stream instantly because there is no network involved. The phone becomes the server, and inference happens where the user interaction occurs.

This moment introduces the core idea of on-device LLMs and why they fundamentally change how AI features behave in React Native apps.

Privacy and compliance by default

The next section shifts the same health app scenario to sensitive data. The transcript explains how cloud AI sends user input to third-party servers, making compliance the developer’s responsibility.

With React Native AI configured for local inference, inputs never leave the device. The model processes data in the user’s hand, removing third-party data transfer and reducing compliance risk by design.

Offline usage and resilience during outages

The video returns to offline scenarios such as subways, flights, and large service outages. Cloud LLMs stop responding when connectivity drops, while local models continue working.

This section highlights that on-device AI is not only about speed or privacy, but about ensuring that AI features remain available regardless of network conditions.

Free inference after installation

A key moment in the episode explains the cost model of on-device AI. Once the app and model are downloaded, every interaction is free for both developers and users.

There is no per-token billing and no usage-based invoices. This changes how AI features can be designed and scaled in mobile apps.

React Native AI architecture overview

The final part of the episode explains how React Native AI is structured. The library supports built-in models shipped with the operating system as well as third-party models tailored to specific needs.

The transcript introduces the two supported engines, Apple Foundation Models via @react-native-ai/apple and MLC LLM via @react-native-ai/mlc, and explains why the JavaScript layer is built on top of the Vercel AI SDK using a provider-based approach.

This section sets the foundation for the rest of the series, which explores each of these pieces in more detail.