Uploaded March 2026 | Updated September 2026, 1 hour ago
In this episode of React Native AI Unpacked, Mike shows what on-device AI can do beyond generating and streaming text.
First, we look at tool calling with the AI SDK. Tool calling lets the model invoke functions and APIs you provide, so it can fetch real-time data and take actions. The demo starts with a getCurrentTime tool, then moves to a calendar example that reads an agenda and highlights a practical detail: tool results often need time zone localization before you render them.
Next, we cover structured outputs. Instead of freeform text, the model returns typed JSON shaped data, which is predictable and directly usable in application logic. The playground demo includes objects like a user profile, enums like RGB, booleans, arrays, and a quick look at how parameters like temperature and topK affect results.
If you want to build offline-first, production-grade AI features in React Native, this episode is the bridge between simple chat UI and app-level automation.
**Links:**
- react-native-ai repository: github.com/callstackincubator/ai
- Vercel AI SDK: https://ai-sdk.dev/
Star the repo, share feedback, and try these patterns in a real app.
## Chapters
00:00 Intro and what changes beyond text
00:39 Tool calling overview
01:11 Tool calling demo with getCurrentTime
01:52 Inspecting the tool implementation
02:32 Calendar and agenda example
02:59 Time zone localization note
03:15 Letting the model choose tools
03:52 Structured outputs overview
04:20 Structured output playground demo
04:43 User profile object (name, age, active)
04:56 Enum output (RGB)
05:01 Boolean and array outputs
05:10 Shopping list ingredients example
05:18 Code walkthrough and parameters
05:43 Array demo and fruit list
05:59 Structured outputs recap and use cases
06:16 Wrap up and call to action
In this episode of React Native AI Unpacked, Mike shows what on-device AI can do beyond generating and streaming text.
First, we look at tool calling with the AI SDK. Tool calling lets the model invoke functions and APIs you provide, so it can fetch real-time data and take actions. The demo starts with a getCurrentTime tool, then moves to a calendar example that reads an agenda and highlights a practical detail: tool results often need time zone localization before you render them.
Next, we cover structured outputs. Instead of freeform text, the model returns typed JSON shaped data, which is predictable and directly usable in application logic. The playground demo includes objects like a user profile, enums like RGB, booleans, arrays, and a quick look at how parameters like temperature and topK affect results.
If you want to build offline-first, production-grade AI features in React Native, this episode is the bridge between simple chat UI and app-level automation.
**Links:**
- react-native-ai repository: github.com/callstackincubator/ai
- Vercel AI SDK: https://ai-sdk.dev/
Star the repo, share feedback, and try these patterns in a real app.
## Chapters
00:00 Intro and what changes beyond text
00:39 Tool calling overview
01:11 Tool calling demo with getCurrentTime
01:52 Inspecting the tool implementation
02:32 Calendar and agenda example
02:59 Time zone localization note
03:15 Letting the model choose tools
03:52 Structured outputs overview
04:20 Structured output playground demo
04:43 User profile object (name, age, active)
04:56 Enum output (RGB)
05:01 Boolean and array outputs
05:10 Shopping list ingredients example
05:18 Code walkthrough and parameters
05:43 Array demo and fruit list
05:59 Structured outputs recap and use cases
06:16 Wrap up and call to action
