# ROHAN AI TESTING, Release Stories This is Awesome ## A NEW ONE AGAIN _qwdswqed · 2026-04-04_ Adds LLM-controlled dynamic UI, tool calling, streaming support, and unifies AI providers across platforms. ### What changed LLM can now dynamically control the UI through @react-native-ai/json-ui integration with enhanced tool call visualization capabilities. Updated MLC-LLM compatibility for latest versions and fixed model availability detection on Android. Enhanced user interface and visual consistency for the Android demo application. Updated MLC-LLM to latest version with fixes for Android demo application compatibility. Documentation now includes a comprehensive compatibility table for different AI SDK versions. New profiler plugin for monitoring and analyzing AI SDK performance. Fixed issue where non-text chat deltas were incorrectly processed before streaming. Implemented tool calling support for streaming responses with improved integration. Llama provider now supports function calling and tool invocation. Refactored to use shared abstractions across different AI providers for better consistency. Emit text-start event before completion to prevent loss of initial character in streamed responses. Updated implementation to conform to AI SDK v3 specification and patterns. Implemented re-ranking functionality for Llama provider to improve search result ordering. Enhanced Llama provider with embeddings, speech processing, and multimodal capabilities. Refactored Llama provider to follow consistent patterns used in Apple and MLC implementations. Example app now only displays Llama reasoning output when it contains meaningful content. Corrected emission of first token in Llama streaming responses. New documentation page explaining required polyfills for AI SDK usage. New @react-native-ai/llama package providing LLM capabilities through Llama integration. Unified demo app implementation to share common code between iOS and Android platforms. Implemented full MLC-LLM support for Android platform with script automation. Upgraded to Expo SDK 54 and removed beta tag from AI SDK dependency. Enhanced setup script to automatically configure MLC for Android development. Implemented function calling capabilities allowing AI to invoke external tools and functions. Made text generation functions more universal and provider-agnostic. Corrected issues preventing proper streaming of AI responses. Updated API to use more descriptive naming for Llama-specific model download checks. Fix example app on iOS Ensured react-native-worklets dependency is properly included. R… ### Why it matters Adds LLM-controlled dynamic UI, tool calling, streaming support, and unifies AI providers across platforms. ### New Features - Add dynamic UI controlled by LLM with JSON UI and tool call visualization (#196), LLM can now dynamically control the UI through @react-native-ai/json-ui integration with enhanced tool call visualization capabilities. - Upgrade MLC-LLM and fix Android demo app (#190), Updated MLC-LLM to latest version with fixes for Android demo application compatibility. - Add AI SDK version compatibility table (#187), Documentation now includes a comprehensive compatibility table for different AI SDK versions. - Add Rozenite AI SDK profiler plugin (#186), New profiler plugin for monitoring and analyzing AI SDK performance. - Add tool calling for streaming (#184), Implemented tool calling support for streaming responses with improved integration. - Support tool calling with Llama (#183), Llama provider now supports function calling and tool invocation. - Unify AI providers with shared abstractions (#179), Refactored to use shared abstractions across different AI providers for better consistency. - Upgrade to AI SDK v3 specification (#177), Updated implementation to conform to AI SDK v3 specification and patterns. - Add re-ranking support for Llama (#176), Implemented re-ranking functionality for Llama provider to improve search result ordering. - Extend Llama provider with embeddings, speech, and multimodal support (#174), Enhanced Llama provider with embeddings, speech processing, and multimodal capabilities. - Unify Llama provider implementation with Apple and MLC patterns (#170), Refactored Llama provider to follow consistent patterns used in Apple and MLC implementations. - Add AI SDK polyfills documentation (#164), New documentation page explaining required polyfills for AI SDK usage. - Create Llama React Native AI provider (#161), New @react-native-ai/llama package providing LLM capabilities through Llama integration. - Refactor demo app for cross-platform code sharing (#163), Unified demo app implementation to share common code between iOS and Android platforms. - Add Android support for MLC (#134), Implemented full MLC-LLM support for Android platform with script automation. - Update to Expo SDK 54 and AI SDK v3 (#132), Upgraded to Expo SDK 54 and removed beta tag from AI SDK dependency. - Add Android support in MLC setup script (#130), Enhanced setup script to automatically configure MLC for Android development. - Add tool calling support (#120), Implemented function calling capabilities allowing AI to invoke external tools and functions. - Refactor generateText and generateStream for universal usage (#121), Made text generation functions more universal and provider-agnostic. ### Bug Fixes - Support newest MLC on iOS and fix availability check on Android (#192), Updated MLC-LLM compatibility for latest versions and fixed model availability detection on Android. - Improve demo app UI on Android (#191), Enhanced user interface and visual consistency for the Android demo application. - Fix MLC chat deltas streaming (#147), Fixed issue where non-text chat deltas were incorrectly processed before streaming. - Fix first character being dropped in streaming (#178), Emit text-start event before completion to prevent loss of initial character in streamed responses. - Fix example app reasoning display (#166), Example app now only displays Llama reasoning output when it contains meaningful content. - Fix Llama streaming response first token (#167), Corrected emission of first token in Llama streaming responses. - Fix streaming functionality, Corrected issues preventing proper streaming of AI responses. - Fix example app on iOS - Add missing react-native-worklets package (#157), Ensured react-native-worklets dependency is properly included. - Fix Android freeze with MLC model handling (#153), Resolved freezing issues on Android when handling specific model variants. - Fix buffer type compatibility (#142), Corrected type definitions for mergeBuffersToFloat32Array to accept library's buffer types. - Fix polyfill initialization RangeError (#141), Properly initialized polyfill to prevent RangeError from @ungap/structured-clone. ### Maintenance - Rename isModelDownloaded to isLlamaModelDownloaded (#194), Updated API to use more descriptive naming for Llama-specific model download checks. - Documentation and maintenance updates, Various documentation improvements including simulator guides, README updates, and theme enhancements.