From 313e2e51d1a1a22fa5e3d0f8b36b3321d1e5bcc6 Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Tue, 10 Feb 2026 20:53:52 +0100
Subject: [PATCH 1/7] chore: add reference files for all functionalities

---
 .../references/core-utilities.md              | 327 ++++++++++++++++++
 .../references/reference-audio.md             | 311 +++++++++++++++++
 .../references/reference-cv-2.md              | 190 ++++++++++
 .../references/reference-cv.md                | 215 ++++++++++++
 .../references/reference-llms.md              | 259 ++++++++++++++
 .../references/reference-models.md            | 265 ++++++++++++++
 .../references/reference-nlp.md               | 169 +++++++++
 .../references/reference-ocr.md               | 171 +++++++++
 8 files changed, 1907 insertions(+)
 create mode 100644 skills/react-native-executorch/references/core-utilities.md
 create mode 100644 skills/react-native-executorch/references/reference-audio.md
 create mode 100644 skills/react-native-executorch/references/reference-cv-2.md
 create mode 100644 skills/react-native-executorch/references/reference-cv.md
 create mode 100644 skills/react-native-executorch/references/reference-llms.md
 create mode 100644 skills/react-native-executorch/references/reference-models.md
 create mode 100644 skills/react-native-executorch/references/reference-nlp.md
 create mode 100644 skills/react-native-executorch/references/reference-ocr.md

diff --git a/skills/react-native-executorch/references/core-utilities.md b/skills/react-native-executorch/references/core-utilities.md
new file mode 100644
index 000000000..9b5a0c18b
--- /dev/null
+++ b/skills/react-native-executorch/references/core-utilities.md
@@ -0,0 +1,327 @@
+---
+title: RN Executorch core Utilities
+description: Reference for using core RN Executorch utils - low-level ExecuTorch bindings, resource management, and error handling.
+---
+
+# useExecutorchModule
+
+**Purpose:** Low-level bindings to ExecuTorch Module API for custom model integration.
+
+**Use cases:** Custom models without dedicated hooks, advanced model control, experimental models, research applications.
+
+**Important:** Use dedicated hooks (useLLM, useClassification, etc.) when available. This hook is for custom models where no pre-built solution exists.
+
+## Basic Usage
+
+```typescript
+import { useExecutorchModule } from 'react-native-executorch';
+
+const executorchModule = useExecutorchModule({
+  modelSource: require('../assets/models/model.pte'),
+});
+```
+
+## Understanding TensorPtr
+
+A `TensorPtr` is the JavaScript representation of a tensor passed to the model:
+
+```typescript
+interface TensorPtr {
+  dataPtr: ArrayBuffer | TypedArray; // Raw data buffer
+  sizes: number[]; // Tensor shape [batch, channels, height, width]
+  scalarType: ScalarType; // Data type (FLOAT, INT, etc.)
+}
+```
+
+## Example usage
+
+```typescript
+import {
+  useExecutorchModule,
+  ScalarType,
+  STYLE_TRANSFER_CANDY,
+} from 'react-native-executorch';
+
+const executorchModule = useExecutorchModule({
+  modelSource: STYLE_TRANSFER_CANDY,
+});
+
+const runInference = async () => {
+  // Prepare input tensor (example: 640x640 RGB image)
+  const inputTensor = {
+    dataPtr: new Float32Array(1 * 3 * 640 * 640),
+    sizes: [1, 3, 640, 640],
+    scalarType: ScalarType.FLOAT,
+  };
+
+  try {
+    // Perform the forward operation and receive the stylized image output.
+    const output = await executorchModule.forward([inputTensor]);
+    // Interpret the output ArrayBuffer
+    // foo(output[0].dataPtr);
+  } catch (error) {
+    // Log any errors that occur during the forward pass.
+    console.error('Error during model execution:', error);
+  }
+};
+```
+
+## Troubleshooting
+
+**Preprocessing required:** You must handle all preprocessing (normalization, resizing, color space conversion) yourself.
+**Postprocessing required:** Output interpretation is your responsibility based on your model's architecture.
+**Shape matching:** Input tensor shapes must exactly match your model's expected input dimensions.
+**Use dedicated hooks:** If a hook exists for your use case, use it instead for automatic pre/post-processing.
+
+## Additional references
+
+- [useExecutorchModule docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/executorch-bindings/useExecutorchModule)
+- [useExecutorchModule API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useExecutorchModule)
+- [ExecuTorch Module API](https://pytorch.org/executorch/stable/extension-module.html)
+- [Typescript API implementation of useExecutorchModule](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/executorch-bindings/ExecutorchModule)
+
+---
+
+# ResourceFetcher
+
+**Purpose:** Manage model and resource downloads with pause/resume capabilities.
+
+**Use cases:** Download management, storage cleanup, progress tracking, offline-first apps.
+
+## Basic Usage
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+// Download multiple resources with progress tracking
+const downloadModels = async () => {
+  try {
+    const uris = await ResourceFetcher.fetch(
+      (progress) =>
+        console.log(`Download progress: ${(progress * 100).toFixed(1)}%`),
+      'https://example.com/llama3_2.pte',
+      'https://example.com/qwen3.pte'
+    );
+
+    if (uris) {
+      console.log('Downloaded files:', uris);
+    } else {
+      console.log('Download was paused or cancelled');
+    }
+  } catch (error) {
+    console.error('Download failed:', error);
+  }
+};
+```
+
+## Pause and Resume Downloads
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+const uris = ResourceFetcher.fetch(
+  (progress) => console.log('Total progress:', progress),
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+).then((uris) => {
+  console.log('URI resolved as: ', uris); // since we pause the fetch, uris is resolved to null
+});
+
+await ResourceFetcher.pauseFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+
+const resolvedUris = await ResourceFetcher.resumeFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+```
+
+## Cancel Downloads
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+const uris = ResourceFetcher.fetch(
+  (progress) => console.log('Total progress:', progress),
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+).then((uris) => {
+  console.log('URI resolved as: ', uris); // since we cancel the fetch, uris is resolved to null
+});
+
+await ResourceFetcher.cancelFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+```
+
+## Manage Downloaded Resources
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+// List all downloaded files
+const listFiles = async () => {
+  const files = await ResourceFetcher.listDownloadedFiles();
+  console.log('All downloaded files:', files);
+
+  const models = await ResourceFetcher.listDownloadedModels();
+  console.log('Model files:', models);
+};
+
+// Clean up old resources
+const cleanup = async () => {
+  const oldModelUrl = 'https://example.com/old_model.pte';
+
+  await ResourceFetcher.deleteResources(oldModelUrl);
+  console.log('Old model deleted');
+};
+```
+
+## Resource Types
+
+Resources can be:
+
+- Remote URLs (https://)
+- Local file paths (file://)
+- Asset references (require())
+- JSON objects
+
+## Troubleshooting
+
+**Resume vs re-fetch:** Use `resumeFetching()` for faster resume. Calling `fetch()` again works but is slower.
+**Progress callback:** Progress is reported as 0-1 for all downloads combined.
+**Null return:** If `fetch()` returns `null`, download was paused or cancelled.
+**Network errors:** Implement retry logic with exponential backoff for reliability.
+**Storage location:** Downloaded files are stored in application's document directory under `react-native-executorch/`
+
+## Additional references
+
+- [ResourceFetcher full reference docs](https://docs.swmansion.com/react-native-executorch/docs/utilities/resource-fetcher)
+- [Loading Models guide](https://docs.swmansion.com/react-native-executorch/docs/fundamentals/loading-models)
+
+---
+
+# Error Handling
+
+**Purpose:** Comprehensive error handling with typed error codes.
+
+**Use cases:** Debugging, production error recovery, user feedback, logging and monitoring.
+
+## Basic Error Handling
+
+```typescript
+import {
+  LLMModule,
+  LLAMA3_2_1B_QLORA,
+  RnExecutorchError,
+  RnExecutorchErrorCode,
+} from 'react-native-executorch';
+
+const llm = new LLMModule({
+  tokenCallback: (token) => console.log(token),
+  messageHistoryCallback: (messages) => console.log(messages),
+});
+
+try {
+  await llm.load(LLAMA3_2_1B_QLORA, (progress) => console.log(progress));
+  await llm.sendMessage('Hello!');
+} catch (err) {
+  if (err instanceof RnExecutorchError) {
+    console.error(`Error code: ${err.code}`);
+    console.error(`Error message: ${err.message}`);
+  } else {
+    throw err;
+  }
+}
+```
+
+## Handling Specific Error Types
+
+```typescript
+import {
+  RnExecutorchError,
+  RnExecutorchErrorCode,
+} from 'react-native-executorch';
+
+const handleModelError = async (llm, message: string) => {
+  try {
+    await llm.sendMessage(message);
+  } catch (err) {
+    if (err instanceof RnExecutorchError) {
+      switch (err.code) {
+        case RnExecutorchErrorCode.ModuleNotLoaded:
+          console.error('Model not loaded. Loading now...');
+          await llm.load(LLAMA3_2_1B_QLORA);
+          // Retry the message
+          await llm.sendMessage(message);
+          break;
+
+        case RnExecutorchErrorCode.ModelGenerating:
+          console.error('Model busy. Waiting...');
+          // Wait and retry, or queue the message
+          break;
+
+        case RnExecutorchErrorCode.InvalidConfig:
+          console.error('Invalid configuration:', err.message);
+          // Reset to default config
+          await llm.configure({ topp: 0.9, temperature: 0.7 });
+          break;
+
+        default:
+          console.error('Unexpected error:', err.message);
+          throw err;
+      }
+    }
+  }
+};
+```
+
+## Error Categories
+
+**Module State Errors**
+
+- `ModuleNotLoaded` - Model not loaded yet
+- `ModelGenerating` - Model already processing
+
+**Configuration Errors**
+
+- `InvalidConfig` - Invalid parameters
+- `InvalidUserInput` - Bad input data
+- `InvalidModelSource` - Wrong model source type
+- `WrongDimensions` - Incorrect tensor shape
+
+**File Operations**
+
+- `FileReadFailed` - Can't read file
+- `FileWriteFailed` - Can't write file
+
+**Download & Resources**
+
+- `DownloadInterrupted` - Download didn't complete
+- `ResourceFetcherDownloadFailed` - Network/server error
+- `ResourceFetcherDownloadInProgress` - Already downloading
+- `ResourceFetcherAlreadyPaused` - Already paused
+- `ResourceFetcherNotActive` - No active download
+
+**Runtime Errors**
+
+- `MemoryAllocationFailed` - Out of memory
+- `NotSupported` - Operation not supported
+- `InvalidProgram` - Invalid model file
+
+For complete error reference, see the [Error Handling documentation](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling).
+
+## Troubleshooting
+
+**Always check instance:** Use `instanceof RnExecutorchError` before accessing `.code`.
+**Log error codes:** Include error codes in logs for easier debugging.
+**Retry logic:** Implement exponential backoff for network and resource errors.
+**User feedback:** Translate error codes into user-friendly messages.
+
+## Additional references
+
+- [Error Handling docs](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling)
+- [Complete error code list](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling#reference)
diff --git a/skills/react-native-executorch/references/reference-audio.md b/skills/react-native-executorch/references/reference-audio.md
new file mode 100644
index 000000000..50bb60c13
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-audio.md
@@ -0,0 +1,311 @@
+---
+title: Audio related models usage
+description: Reference for using Speech to Text, Text to Speech and Voice Activity Detection models.
+---
+
+# useSpeechToText
+
+**Purpose:** Convert spoken audio to text (transcription).
+**Use cases:** Voice assistants, transcription apps, voice commands, accessibility features.
+
+## Basic Usage
+
+```typescript
+import { useSpeechToText, WHISPER_TINY_EN } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+
+const model = useSpeechToText({ model: WHISPER_TINY_EN });
+
+// Process audio file
+const audioContext = new AudioContext({ sampleRate: 16000 });
+const decodedAudio = await audioContext.decodeAudioDataSource(audioUri);
+const waveform = decodedAudio.getChannelData(0);
+
+const transcription = await model.transcribe(waveform);
+console.log(transcription);
+```
+
+## Multilingual Transcription
+
+```typescript
+import { WHISPER_TINY } from 'react-native-executorch';
+
+const model = useSpeechToText({ model: WHISPER_TINY });
+
+// Specify language
+const transcription = await model.transcribe(spanishAudio, {
+  language: 'es',
+});
+```
+
+## Streaming Transcription
+
+```typescript
+import { AudioRecorder, AudioManager } from 'react-native-audio-api';
+
+const recorder = new AudioRecorder({
+  sampleRate: 16000,
+  bufferLengthInSamples: 1600,
+});
+
+// Start streaming
+recorder.onAudioReady(({ buffer }) => {
+  model.streamInsert(buffer.getChannelData(0));
+});
+recorder.start();
+
+await model.stream();
+
+// Access results
+console.log(model.committedTranscription);
+console.log(model.nonCommittedTranscription);
+
+// Stop streaming
+recorder.stop();
+model.streamStop();
+```
+
+## Troubleshooting
+
+**Audio must be 16kHz:** Ensure proper sample rate before processing
+**Streaming algorithm:** Use whisper-streaming for longer audio (handles 30s chunks automatically)
+
+## Additional references
+
+- [useSpeechToText docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useSpeechToText)
+- [HuggingFace STT collection](https://huggingface.co/collections/software-mansion/speech-to-text)
+- [Available models](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---speech-to-text)
+- [useSpeechToText API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useSpeechToText)
+- [Typescript API implementation of useSpeechToText](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/SpeechToTextModule)
+
+---
+
+# useTextToSpeech
+
+**Purpose:** Convert text to natural-sounding speech (TTS).
+**Use cases:** Voice assistants, audiobooks, accessibility tools, voice navigation.
+
+## Basic Usage
+
+```typescript
+import {
+  useTextToSpeech,
+  KOKORO_MEDIUM,
+  KOKORO_VOICE_AF_HEART,
+} from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+
+const model = useTextToSpeech({
+  model: KOKORO_MEDIUM,
+  voice: KOKORO_VOICE_AF_HEART,
+});
+
+const audioContext = new AudioContext({ sampleRate: 24000 });
+
+const handleSpeech = async (text: string) => {
+  const speed = 1.0;
+  const waveform = await model.forward(text, speed);
+
+  const audioBuffer = audioContext.createBuffer(1, waveform.length, 24000);
+  audioBuffer.getChannelData(0).set(waveform);
+
+  const source = audioContext.createBufferSource();
+  source.buffer = audioBuffer;
+  source.connect(audioContext.destination);
+  source.start();
+};
+```
+
+## Streaming TTS
+
+```typescript
+// Stream chunks for lower latency
+await tts.stream({
+  text: 'Long text to be streamed chunk by chunk...',
+  speed: 1.0,
+  onNext: async (chunk) => {
+    return new Promise((resolve) => {
+      const buffer = ctx.createBuffer(1, chunk.length, 24000);
+      buffer.getChannelData(0).set(chunk);
+
+      const source = ctx.createBufferSource();
+      source.buffer = buffer;
+      source.connect(ctx.destination);
+      source.onEnded = () => resolve();
+      source.start();
+    });
+  },
+});
+```
+
+## Available Models & Voices
+
+**Model:** Kokoro (English only)
+
+For all available models check out [this exported HuggingFace models collection](https://huggingface.co/software-mansion/react-native-executorch-kokoro).
+
+**Available Voices:**
+
+- `KOKORO_VOICE_AF_HEART` - Female, heart
+- `KOKORO_VOICE_AF_SKY` - Female, sky
+- `KOKORO_VOICE_AF_BELLA` - Female, bella
+- `KOKORO_VOICE_AF_NICOLE` - Female, nicole
+- `KOKORO_VOICE_AF_SARAH` - Female, sarah
+- `KOKORO_VOICE_AM_ADAM` - Male, adam
+- `KOKORO_VOICE_AM_MICHAEL` - Male, michael
+- `KOKORO_VOICE_BF_EMMA` - British Female, emma
+- `KOKORO_VOICE_BF_ISABELLA` - British Female, isabella
+- `KOKORO_VOICE_BM_GEORGE` - British Male, george
+- `KOKORO_VOICE_BM_LEWIS` - British Male, lewis
+
+## Troubleshooting
+
+**Streaming vs Forward:** Use `stream()` for long texts to reduce time-to-first-audio
+
+## Additional references
+
+- [useTextToSpeech docs - reference and examples](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToSpeech)
+- [Supported Voices](https://docs.swmansion.com/react-native-executorch/docs/api-reference#tts-supported-voices)
+- [useTextToSpeech API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToSpeech)
+- [HuggingFace TTS collection](https://huggingface.co/collections/software-mansion/text-to-speech)
+- [Typescript API implementation of useTextToSpeech hook](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TextToSpeechModule)
+
+---
+
+# useVAD
+
+**Purpose:** Detect speech segments in audio (Voice Activity Detection).
+
+**Use cases:** Audio preprocessing, removing silence, speech segmentation, smart recording.
+
+## Basic Usage
+
+```typescript
+import { useVAD, FSMN_VAD } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+import * as FileSystem from 'expo-file-system';
+
+const model = useVAD({
+  model: FSMN_VAD,
+});
+
+const { uri } = await FileSystem.downloadAsync(
+  'https://some-audio-url.com/file.mp3',
+  FileSystem.cacheDirectory + 'audio_file'
+);
+
+const audioContext = new AudioContext({ sampleRate: 16000 });
+const decodedAudioData = await audioContext.decodeAudioDataSource(uri);
+const audioBuffer = decodedAudioData.getChannelData(0);
+
+try {
+  // NOTE: to obtain segments in seconds, you need to divide
+  // start / end of the segment by the sampling rate (16k)
+
+  const speechSegments = await model.forward(audioBuffer);
+  console.log(speechSegments);
+} catch (error) {
+  console.error('Error during running VAD model', error);
+}
+```
+
+## Example usage
+
+```tsx
+import React from 'react';
+import { Button, Text, SafeAreaView } from 'react-native';
+import { useVAD, FSMN_VAD } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+import * as FileSystem from 'expo-file-system';
+
+export default function App() {
+  const model = useVAD({
+    model: FSMN_VAD,
+  });
+
+  const audioURL = 'https://some-audio-url.com/file.mp3';
+
+  const handleAudio = async () => {
+    if (!model) {
+      console.error('VAD model is not loaded yet.');
+      return;
+    }
+
+    console.log('Processing URL:', audioURL);
+
+    try {
+      const { uri } = await FileSystem.downloadAsync(
+        audioURL,
+        FileSystem.cacheDirectory + 'vad_example.tmp'
+      );
+
+      const audioContext = new AudioContext({ sampleRate: 16000 });
+      const originalDecodedBuffer =
+        await audioContext.decodeAudioDataSource(uri);
+      const originalChannelData = originalDecodedBuffer.getChannelData(0);
+
+      const segments = await model.forward(originalChannelData);
+      if (segments.length === 0) {
+        console.log('No speech segments were found.');
+        return;
+      }
+      console.log(`Found ${segments.length} speech segments.`);
+
+      const totalLength = segments.reduce(
+        (sum, seg) => sum + (seg.end - seg.start),
+        0
+      );
+      const newAudioBuffer = audioContext.createBuffer(
+        1, // Mono
+        totalLength,
+        originalDecodedBuffer.sampleRate
+      );
+      const newChannelData = newAudioBuffer.getChannelData(0);
+
+      let offset = 0;
+      for (const segment of segments) {
+        const slice = originalChannelData.subarray(segment.start, segment.end);
+        newChannelData.set(slice, offset);
+        offset += slice.length;
+      }
+
+      //  Play the processed audio
+      const source = audioContext.createBufferSource();
+      source.buffer = newAudioBuffer;
+      source.connect(audioContext.destination);
+      source.start();
+    } catch (error) {
+      console.error('Error processing audio data:', error);
+    }
+  };
+
+  return (
+    <SafeAreaView>
+      <Text>
+        Press the button to process and play speech from a sample file.
+      </Text>
+      <Button onPress={handleAudio} title="Run VAD Example" />
+    </SafeAreaView>
+  );
+}
+```
+
+## Available Models
+
+**Model:** [FSMN-VAD](https://huggingface.co/software-mansion/react-native-executorch-fsmn-vad)
+
+For all available models check out exported models in [this HuggingFace VAD models collection](https://huggingface.co/collections/software-mansion/voice-activity-detection).
+
+## Troubleshooting
+
+**Audio must be 16kHz:** Ensure proper sample rate  
+**Timestamps are indices:** Divide by sample rate (16000) to get seconds  
+**Returns array of segments:** `[{ start: number, end: number }]`
+
+## Additional references
+
+- [useVAD docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVAD)
+- [HuggingFace VAD collection](https://huggingface.co/collections/software-mansion/voice-activity-detection)
+- [useVAD API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVAD)
+- [available VAD model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---voice-activity-detection)
+- [Typescript API implementation of useVad hook](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/VADModule)
diff --git a/skills/react-native-executorch/references/reference-cv-2.md b/skills/react-native-executorch/references/reference-cv-2.md
new file mode 100644
index 000000000..1c5b07d16
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-cv-2.md
@@ -0,0 +1,190 @@
+---
+title: Computer Vision models usage (continued)
+description: Reference for using Style Transfer, Text to Image, Image Embeddings.
+---
+
+# useStyleTransfer
+
+**Purpose:** Apply artistic styles from one image to the content of another image.
+
+**Use cases:** Photo editing, artistic filters, creative content generation, style transformation apps.
+
+## Basic Usage
+
+```typescript
+import {
+  useStyleTransfer,
+  STYLE_TRANSFER_CANDY,
+} from 'react-native-executorch';
+
+const model = useStyleTransfer({ model: STYLE_TRANSFER_CANDY });
+
+const imageUri = 'file:///Users/.../photo.png';
+
+try {
+  const generatedImageUrl = await model.forward(imageUri);
+  console.log('Styled image:', generatedImageUrl);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Available Models
+
+**Model constants:**
+
+- `STYLE_TRANSFER_CANDY` - Candy artistic style
+- `STYLE_TRANSFER_MOSAIC` - Mosaic artistic style
+- `STYLE_TRANSFER_UDNIE` - Udnie artistic style
+- `STYLE_TRANSFER_RAIN_PRINCESS` - Rain princess artistic style
+
+For the latest available models reference exported models in [HuggingFace Style Transfer collection](https://huggingface.co/collections/software-mansion/style-transfer)
+
+## Troubleshooting
+
+**Image storage:** Both input images from external sources and generated images are stored in the application's temporary directory.
+**Processing time:** Style transfer can be computationally intensive. Expect processing to take several seconds depending on image size and device capabilities.
+
+## Additional references
+
+- [useStyleTransfer docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useStyleTransfer)
+- [HuggingFace Style Transfer collection](https://huggingface.co/collections/software-mansion/style-transfer)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---style-transfer)
+- [useStyleTransfer API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useStyleTransfer)
+
+---
+
+# useTextToImage
+
+**Purpose:** Generate images from text descriptions using on-device Stable Diffusion.
+
+**Use cases:** AI art generation, creative content creation, concept visualization, design prototyping.
+
+## Basic Usage
+
+```typescript
+import { useTextToImage, BK_SDM_TINY_VPRED_256 } from 'react-native-executorch';
+
+const model = useTextToImage({ model: BK_SDM_TINY_VPRED_256 });
+
+const input = 'a castle';
+
+try {
+  const image = await model.generate(input);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Example Usage with Parameters
+
+```tsx
+import { useTextToImage, BK_SDM_TINY_VPRED_256 } from 'react-native-executorch';
+
+function App() {
+  const model = useTextToImage({ model: BK_SDM_TINY_VPRED_256 });
+
+  //...
+  const input = 'a medieval castle by the sea shore';
+
+  const imageSize = 256;
+  const numSteps = 25;
+
+  try {
+    image = await model.generate(input, imageSize, numSteps);
+  } catch (error) {
+    console.error(error);
+  }
+  //...
+
+  return <Image source={{ uri: `data:image/png;base64,${image}` }} />;
+}
+```
+
+**Model constants:** `BK_SDM_TINY_VPRED_256`
+
+For the latest available models reference exported models in [HuggingFace Text to Image collection](https://huggingface.co/collections/software-mansion/text-to-image)
+
+## Troubleshooting
+
+**Memory requirements:** Larger image sizes require significantly more memory. Use 256x256 for lower-end devices, 512x512 for high-end devices.
+**Image size constraint:** Image size must be a multiple of 32 (e.g., 256, 320, 384, 512) due to U-Net and VAE decoder architecture.
+**Generation time:** Expect 20-60 seconds per image depending on device, image size, and number of steps.
+**Prompt engineering:** More descriptive prompts yield better results. Include style descriptors like "digital art", "photorealistic", "watercolor" etc.
+
+## Additional references
+
+- [useTextToImage docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useTextToImage)
+- [HuggingFace Text to Image collection](https://huggingface.co/collections/software-mansion/text-to-image)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---image-generation)
+- [useTextToImage API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToImage)
+- [Typescript API implementation of Text to Image](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/TextToImageModule)
+
+---
+
+# useImageEmbeddings
+
+**Purpose:** Convert images into numerical vectors for similarity comparison and image search.
+
+**Use cases:** Image similarity search, duplicate detection, visual search, clustering, content-based retrieval.
+
+## Basic Usage
+
+```typescript
+import {
+  useImageEmbeddings,
+  CLIP_VIT_BASE_PATCH32_IMAGE,
+} from 'react-native-executorch';
+
+const model = useImageEmbeddings({ model: CLIP_VIT_BASE_PATCH32_IMAGE });
+
+try {
+  const imageEmbedding = await model.forward('https://url-to-image.jpg');
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Computing Image Similarity
+
+```typescript
+const dotProduct = (a: Float32Array, b: Float32Array) =>
+  a.reduce((sum, val, i) => sum + val * b[i], 0);
+
+const cosineSimilarity = (a: Float32Array, b: Float32Array) => {
+  const dot = dotProduct(a, b);
+  const normA = Math.sqrt(dotProduct(a, a));
+  const normB = Math.sqrt(dotProduct(b, b));
+  return dot / (normA * normB);
+};
+
+try {
+  // we assume you've provided catImage and dogImage
+  const catImageEmbedding = await model.forward(catImage);
+  const dogImageEmbedding = await model.forward(dogImage);
+
+  const similarity = cosineSimilarity(catImageEmbedding, dogImageEmbedding);
+
+  console.log(`Cosine similarity: ${similarity}`);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Available Models
+
+**Model constants:** `CLIP_VIT_BASE_PATCH32_IMAGE`
+
+For the latest available models reference exported models in [HuggingFace Image Embeddings collection](https://huggingface.co/collections/software-mansion/image-embeddings)
+
+## Troubleshooting
+
+**Normalized vectors:** Returned embeddings are already normalized, so cosine similarity can be computed as a simple dot product.
+**Image resizing:** Images are automatically resized to 224×224. Resizing large images can add processing time.
+
+## Additional references
+
+- [useImageEmbeddings docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useImageEmbeddings)
+- [HuggingFace Image Embeddings collection](https://huggingface.co/collections/software-mansion/image-embeddings)
+- [useImageEmbeddings API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useImageEmbeddings)
+- [Typescript API implementation of useImageEmbeddings](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ImageEmbeddingsModule)
diff --git a/skills/react-native-executorch/references/reference-cv.md b/skills/react-native-executorch/references/reference-cv.md
new file mode 100644
index 000000000..fb07b0ccf
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-cv.md
@@ -0,0 +1,215 @@
+---
+title: Computer Vision models usage
+description: Reference for using Image Classification, Image Segmentation, and Object Detection models.
+---
+
+# useClassification
+
+**Purpose:** Classify images into predefined categories.
+
+**Use cases:** Image recognition apps, content filtering, product categorization, accessibility features.
+
+## Basic Usage
+
+```typescript
+import { useClassification, EFFICIENTNET_V2_S } from 'react-native-executorch';
+
+const model = useClassification({ model: EFFICIENTNET_V2_S });
+
+const imageUri = 'file:///Users/.../cute_puppy.png';
+
+try {
+  const classesWithProbabilities = await model.forward(imageUri);
+  console.log(classesWithProbabilities);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Processing Results
+
+```typescript
+import { useClassification, EFFICIENTNET_V2_S } from 'react-native-executorch';
+
+function App() {
+  const model = useClassification({ model: EFFICIENTNET_V2_S });
+
+  // ...
+  const imageUri = 'file:///Users/.../cute_puppy.png';
+
+  try {
+    const classesWithProbabilities = await model.forward(imageUri);
+
+    // Extract three classes with the highest probabilities
+    const topThreeClasses = Object.entries(classesWithProbabilities)
+      .sort(([, a], [, b]) => b - a)
+      .slice(0, 3)
+      .map(([label, score]) => ({ label, score }));
+  } catch (error) {
+    console.error(error);
+  }
+  // ...
+}
+```
+
+## Available Models
+
+**Model constant:** `EFFICIENTNET_V2_S`
+
+For the latest available models reference exported models in [HuggingFace Classification collection](https://huggingface.co/collections/software-mansion/classification)
+
+## Troubleshooting
+
+**Confidence interpretation:** The class with the highest probability is typically assigned, but multiple high probabilities may indicate model uncertainty.
+**Image formats:** Accepts remote URLs, local file URIs, and base64-encoded images. Remote images are stored in the app's temporary directory.
+
+## Additional references
+
+- [useClassification docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useClassification)
+- [HuggingFace Classification collection](https://huggingface.co/collections/software-mansion/classification)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---classification)
+- [useClassification API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useClassification)
+- [Typescript API implementation of segmentation](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ClassificationModule)
+
+---
+
+# useImageSegmentation
+
+**Purpose:** Classify each pixel in an image to detect objects on a per-pixel basis.
+
+**Use cases:** Object detection, image editing, scene understanding.
+
+## Basic Usage
+
+```typescript
+import {
+  useImageSegmentation,
+  DEEPLAB_V3_RESNET50,
+  DeeplabLabel,
+} from 'react-native-executorch';
+
+const model = useImageSegmentation({ model: DEEPLAB_V3_RESNET50 });
+
+const imageUri = 'file:///Users/.../cute_cat.png';
+
+try {
+  const outputDict = await model.forward(imageUri);
+  console.log(outputDict[DeeplabLabel.ARGMAX]);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Advanced Usage with Classes of Interest
+
+```typescript
+const model = useImageSegmentation({ model: DEEPLAB_V3_RESNET50 });
+
+const handleSegmentation = async (imageUri: string) => {
+  try {
+    const outputDict = await model.forward(
+      imageUri,
+      [DeeplabLabel.CAT, DeeplabLabel.DOG, DeeplabLabel.PERSON],
+      true
+    );
+
+    const argmaxArray = outputDict[DeeplabLabel.ARGMAX];
+
+    const catProbabilities = outputDict[DeeplabLabel.CAT];
+    const dogProbabilities = outputDict[DeeplabLabel.DOG];
+    const personProbabilities = outputDict[DeeplabLabel.PERSON];
+    // ...
+  } catch (error) {
+    console.error(error);
+  }
+};
+```
+
+## Available Models
+
+**Model constant:** `DEEPLAB_V3_RESNET50`
+
+For the latest available models check out exported models in [this HuggingFace Segmentation collection](https://huggingface.co/collections/software-mansion/image-segmentation)
+
+## Troubleshooting
+
+**Performance:** Setting `resize=true` significantly increases processing time. Use `resize=false` for better performance when you don't need original image dimensions.
+**Memory usage:** Resize increases memory usage, especially with high-resolution images.
+**Pixel mapping:** When `resize=false`, pixel indices map to a 224x224 grid. When `resize=true`, indices map to original image dimensions.
+
+## Additional references
+
+- [useImageSegmentation docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useImageSegmentation)
+- [useImageSegmentation API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useImageSegmentation)
+- [HuggingFace Segmentation collection](https://huggingface.co/collections/software-mansion/image-segmentation)
+- [Typescript API implementation of segmentation](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ImageSegmentationModule)
+
+---
+
+# useObjectDetection
+
+**Purpose:** Identify and locate objects within images by providing bounding boxes, labels, and confidence scores.
+
+**Use cases:** Object detection.
+
+## Basic Usage
+
+```typescript
+import {
+  useObjectDetection,
+  SSDLITE_320_MOBILENET_V3_LARGE,
+} from 'react-native-executorch';
+
+const model = useObjectDetection({ model: SSDLITE_320_MOBILENET_V3_LARGE });
+
+try {
+  const detections = await model.forward('https://url-to-image.jpg');
+
+  for (const detection of detections) {
+    console.log('Bounding box:', detection.bbox);
+    console.log('Label:', detection.label);
+    console.log('Confidence:', detection.score);
+  }
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Detection Object Structure
+
+Each detection returned by `forward` has the following structure:
+
+```typescript
+interface Bbox {
+  x1: number; // Bottom-left x coordinate
+  y1: number; // Bottom-left y coordinate
+  x2: number; // Top-right x coordinate
+  y2: number; // Top-right y coordinate
+}
+
+interface Detection {
+  bbox: Bbox;
+  label: keyof typeof CocoLabels; // Object class name
+  score: number; // Confidence score (0-1)
+}
+```
+
+## Available Models
+
+**Model constant:** `SSDLITE_320_MOBILENET_V3_LARGE`
+
+For the latest available models reference exported models in [HuggingFace Object Detection collection](https://huggingface.co/collections/software-mansion/object-detection)
+
+## Troubleshooting
+
+**Multiple detections:** The model may detect the same object multiple times with slightly different bounding boxes. Consider implementing non-maximum suppression (NMS) if needed.
+**Confidence thresholds:** Adjust the confidence threshold based on your use case. Higher thresholds (>0.7) reduce false positives but may miss objects.
+**Coordinate system:** Bounding box coordinates are in pixel space relative to the input image dimensions.
+
+## Additional references
+
+- [useObjectDetection docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useObjectDetection)
+- [HuggingFace Object Detection collection](https://huggingface.co/collections/software-mansion/object-detection)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---object-detection)
+- [useObjectDetection API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useObjectDetection)
+- [Typescript API implementation of Object Detection](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ObjectDetectionModule)
diff --git a/skills/react-native-executorch/references/reference-llms.md b/skills/react-native-executorch/references/reference-llms.md
new file mode 100644
index 000000000..1497765fc
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-llms.md
@@ -0,0 +1,259 @@
+---
+title: LLMs usage
+description: Reference for using Large Language Models in React Native Executorch.
+---
+
+# useLLM
+
+**Purpose:** Run Large Language Models (LLMs) on-device for text generation, chat, tool calling, and structured outputs.
+
+**Use cases:** AI assistants, text generation, function calling, structured data extraction.
+
+## Basic Usage
+
+```typescript
+import { useLLM, LLAMA3_2_1B } from 'react-native-executorch';
+
+const llm = useLLM({ model: LLAMA3_2_1B });
+```
+
+## Functional Mode (Stateless)
+
+```tsx
+const llm = useLLM({ model: LLAMA3_2_1B });
+
+const handleGenerate = async () => {
+  const chat: Message[] = [
+    { role: 'system', content: 'You are a helpful assistant' },
+    { role: 'user', content: 'Hi!' },
+    { role: 'assistant', content: 'Hi!, how can I help you?' },
+    { role: 'user', content: 'What is the meaning of life?' },
+  ];
+
+  // Chat completion - returns the generated response
+  const response = await llm.generate(chat);
+  console.log('Complete response:', response);
+};
+
+return (
+  <View>
+    <Button onPress={handleGenerate} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Managed Mode (Stateful)
+
+```tsx
+// Configure the model
+useEffect(() => {
+  llm.configure({
+    chatConfig: {
+      systemPrompt: 'You are a helpful assistant',
+      contextWindowLength: 10,
+    },
+    generationConfig: {
+      temperature: 0.7,
+      topp: 0.9,
+    },
+  });
+}, []);
+
+// Send messages
+llm.sendMessage('Hello!');
+
+// Access conversation history
+console.log(llm.messageHistory);
+```
+
+## Tool Calling
+
+```
+const TOOL_DEFINITIONS: LLMTool[] = [
+  {
+    name: 'get_weather',
+    description: 'Get/check weather in given location.',
+    parameters: {
+      type: 'dict',
+      properties: {
+        location: {
+          type: 'string',
+          description: 'Location where user wants to check weather',
+        },
+      },
+      required: ['location'],
+    },
+  },
+];
+
+const llm = useLLM({ model: HAMMER2_1_1_5B });
+
+const handleGenerate = () => {
+  const chat: Message[] = [
+    {
+      role: 'system',
+      content: `You are a helpful assistant. Current time and date: ${new Date().toString()}`,
+    },
+    {
+      role: 'user',
+      content: `Hi, what's the weather like in Cracow right now?`,
+    },
+  ];
+
+  // Chat completion
+  llm.generate(chat, TOOL_DEFINITIONS);
+};
+
+export const executeTool: (call: ToolCall) => Promise<string | null> = async (
+  call
+) => {
+  switch (call.toolName) {
+    case 'get_weather':
+      return await get_weather(call);
+    default:
+      console.error(`Wrong function! We don't handle it!`);
+      return null;
+  }
+};
+
+const { configure } = llm;
+useEffect(() => {
+  configure({
+    chatConfig: {
+      systemPrompt: `${DEFAULT_SYSTEM_PROMPT} Current time and date: ${new Date().toString()}`,
+    },
+    toolsConfig: {
+      tools: TOOL_DEFINITIONS,
+      executeToolCallback: executeTool,
+      displayToolCalls: true,
+    },
+  });
+}, [configure]);
+
+return (
+  <View>
+    <Button onPress={handleGenerate} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Structured Output
+
+```tsx
+import { Schema } from 'jsonschema';
+
+const responseSchema: Schema = {
+  properties: {
+    username: {
+      type: 'string',
+      description: 'Name of user, that is asking a question.',
+    },
+    question: {
+      type: 'string',
+      description: 'Question that user asks.',
+    },
+    bid: {
+      type: 'number',
+      description: 'Amount of money, that user offers.',
+    },
+    currency: {
+      type: 'string',
+      description: 'Currency of offer.',
+    },
+  },
+  required: ['username', 'bid'],
+  type: 'object',
+};
+
+// alternatively use Zod
+import * as z from 'zod/v4';
+const responseSchemaWithZod = z.object({
+  username: z
+    .string()
+    .meta({ description: 'Name of user, that is asking a question.' }),
+  question: z.optional(
+    z.string().meta({ description: 'Question that user asks.' })
+  ),
+  bid: z.number().meta({ description: 'Amount of money, that user offers.' }),
+  currency: z.optional(z.string().meta({ description: 'Currency of offer.' })),
+});
+
+const llm = useLLM({ model: QWEN3_4B_QUANTIZED });
+
+useEffect(() => {
+  const formattingInstructions = getStructuredOutputPrompt(responseSchema);
+  // alternatively pass schema defined with Zod
+  //  const formattingInstructions = getStructuredOutputPrompt(responseSchemaWithZod);
+
+  // Some extra prompting to improve quality of response.
+  const prompt = `Your goal is to parse user's messages and return them in JSON format. Don't respond to user. Simply return JSON with user's question parsed. \n${formattingInstructions}\n /no_think`;
+
+  llm.configure({
+    chatConfig: {
+      systemPrompt: prompt,
+    },
+  });
+}, []);
+
+useEffect(() => {
+  const lastMessage = llm.messageHistory.at(-1);
+  if (!llm.isGenerating && lastMessage?.role === 'assistant') {
+    try {
+      const formattedOutput = fixAndValidateStructuredOutput(
+        lastMessage.content,
+        responseSchemaWithZod
+      );
+      // Zod will allow you to correctly type output
+      const formattedOutputWithZod = fixAndValidateStructuredOutput(
+        lastMessage.content,
+        responseSchema
+      );
+      console.log('Formatted output:', formattedOutput, formattedOutputWithZod);
+    } catch (e) {
+      console.log(
+        "Error parsing output and/or output doesn't match required schema!",
+        e
+      );
+    }
+  }
+}, [llm.messageHistory, llm.isGenerating]);
+
+const send = () => {
+  const message = `I'm John. Is this product damaged? I can give you $100 for this.`;
+  llm.sendMessage(message);
+};
+
+return (
+  <View>
+    <Button onPress={send} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Interrupting Generation
+
+```tsx
+// Stop generating
+llm.interrupt();
+
+// Check if generating
+{
+  llm.isGenerating && <Button onPress={llm.interrupt} title="Stop" />;
+}
+```
+
+## Troubleshooting
+
+**Memory issues:** Use quantized models on lower-end devices - suggest smaller models as lower-end devices might not be able to fit LLMs into memory.
+**Crash on unmount:** Always call `llm.interrupt()` and wait for `isGenerating === false` before unmounting.
+**Reasoning mode:** Model-specific feature (e.g., Qwen 3 uses `/no_think` suffix to disable) - search for a way to disable the reasoning.
+
+## Additional references
+
+- [useLLM docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useLLM)
+- [HuggingFace repository with exported models](https://huggingface.co/collections/software-mansion/llm)
+- [Available LLM model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---lmm)
+- [API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useLLM)
diff --git a/skills/react-native-executorch/references/reference-models.md b/skills/react-native-executorch/references/reference-models.md
new file mode 100644
index 000000000..c8ff0fcab
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-models.md
@@ -0,0 +1,265 @@
+---
+title: Available models and loading models.
+description: RN Executorch provides models for a variety of AI tasks - Image classification, Object detection, Style transfer, OCR, Image embeddings, Image segmentation, Text to image, LLMs, Text embeddings, Speech to text, Text to speech, and Voice activity detection. This reference lists all supported models and provides a quick reference on available ways of loading a model.
+---
+
+# Available models RN Executorch
+
+It is generally recommended to use models provided by the authors of the library, which are available at [HuggingFace repository](https://huggingface.co/software-mansion/collections).
+
+---
+
+## LLMs (Large Language Models)
+
+For a full list of available LLM models reference [HuggingFace LLMs collection](https://huggingface.co/collections/software-mansion/llm).
+
+### LLAMA 3.2
+
+- **LLAMA3_2_3B** - Llama 3.2 3B original (bf16)
+- **LLAMA3_2_3B_QLORA** - Llama 3.2 3B with QLoRA quantization
+- **LLAMA3_2_3B_SPINQUANT** - Llama 3.2 3B with SpinQuant
+- **LLAMA3_2_1B** - Llama 3.2 1B original (bf16)
+- **LLAMA3_2_1B_QLORA** - Llama 3.2 1B with QLoRA quantization
+- **LLAMA3_2_1B_SPINQUANT** - Llama 3.2 1B with SpinQuant
+
+[HuggingFace Llama 3.2 Model](https://huggingface.co/software-mansion/react-native-executorch-llama-3.2)
+
+### QWEN 3
+
+- **QWEN3_0_6B** - Qwen 3 0.6B original (bf16)
+- **QWEN3_0_6B_QUANTIZED** - Qwen 3 0.6B quantized (8da4w)
+- **QWEN3_1_7B** - Qwen 3 1.7B original (bf16)
+- **QWEN3_1_7B_QUANTIZED** - Qwen 3 1.7B quantized (8da4w)
+- **QWEN3_4B** - Qwen 3 4B original (bf16)
+- **QWEN3_4B_QUANTIZED** - Qwen 3 4B quantized (8da4w)
+
+[HuggingFace Qwen 3 Model](https://huggingface.co/software-mansion/react-native-executorch-qwen-3)
+
+### HAMMER 2.1
+
+- **HAMMER2_1_0_5B** - Hammer 2.1 0.5B original (bf16)
+- **HAMMER2_1_0_5B_QUANTIZED** - Hammer 2.1 0.5B quantized (8da4w)
+- **HAMMER2_1_1_5B** - Hammer 2.1 1.5B original (bf16)
+- **HAMMER2_1_1_5B_QUANTIZED** - Hammer 2.1 1.5B quantized (8da4w)
+- **HAMMER2_1_3B** - Hammer 2.1 3B original (bf16)
+- **HAMMER2_1_3B_QUANTIZED** - Hammer 2.1 3B quantized (8da4w)
+
+[HuggingFace Hammer 2.1 model](https://huggingface.co/software-mansion/react-native-executorch-hammer-2.1)
+
+### SMOLLM2
+
+- **SMOLLM2_1_135M** - SmolLM2 135M original (bf16)
+- **SMOLLM2_1_135M_QUANTIZED** - SmolLM2 135M quantized (8da4w)
+- **SMOLLM2_1_360M** - SmolLM2 360M original (bf16)
+- **SMOLLM2_1_360M_QUANTIZED** - SmolLM2 360M quantized (8da4w)
+- **SMOLLM2_1_1_7B** - SmolLM2 1.7B original (bf16)
+- **SMOLLM2_1_1_7B_QUANTIZED** - SmolLM2 1.7B quantized (8da4w)
+
+[HuggingFace SmoLlm 2 model](https://huggingface.co/software-mansion/react-native-executorch-smolLm-2)
+
+### QWEN 2.5
+
+- **QWEN2_5_0_5B** - Qwen 2.5 0.5B original (bf16)
+- **QWEN2_5_0_5B_QUANTIZED** - Qwen 2.5 0.5B quantized (8da4w)
+- **QWEN2_5_1_5B** - Qwen 2.5 1.5B original (bf16)
+- **QWEN2_5_1_5B_QUANTIZED** - Qwen 2.5 1.5B quantized (8da4w)
+- **QWEN2_5_3B** - Qwen 2.5 3B original (bf16)
+- **QWEN2_5_3B_QUANTIZED** - Qwen 2.5 3B quantized (8da4w)
+
+[HuggingFace Qwen 2.5 Model](https://huggingface.co/software-mansion/react-native-executorch-qwen-2.5)
+
+### PHI 4
+
+- **PHI_4_MINI_4B** - Phi 4 Mini 4B original (bf16)
+- **PHI_4_MINI_4B_QUANTIZED** - Phi 4 Mini 4B quantized (8da4w)
+
+[HuggingFace PHI 4 Mini Model](https://huggingface.co/software-mansion/react-native-executorch-phi-4-mini)
+
+---
+
+## Image classification
+
+- **EFFICIENTNET_V2_S** - [EfficientNet V2 S](https://huggingface.co/software-mansion/react-native-executorch-efficientnet-v2-s) (CoreML for iOS, XNNPACK for Android)
+
+For a list of all available Image Classification models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/classification).
+
+---
+
+## Object detection
+
+- **SSDLITE_320_MOBILENET_V3_LARGE** - [SSDLite 320 with MobileNet V3 Large](https://huggingface.co/software-mansion/react-native-executorch-ssdlite320-mobilenet-v3-large)
+
+For a list of all available Object Detection models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/object-detection).
+
+---
+
+## Style transfer
+
+- **STYLE_TRANSFER_CANDY** - [Candy style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-candy)
+- **STYLE_TRANSFER_MOSAIC** - [Mosaic style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-mosaic)
+- **STYLE_TRANSFER_RAIN_PRINCESS** - [Rain Princess style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-rain-princess)
+- **STYLE_TRANSFER_UDNIE** - [Udnie style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-udnie)
+
+For a list of all available Style Transfer models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/style-transfer).
+
+---
+
+## OCR
+
+- **DETECTOR_CRAFT** - [CRAFT text detector](https://huggingface.co/software-mansion/react-native-executorch-detector-craft) - detects text regions in images
+- **RECOGNIZER_CRNN_EN** - [CRNN text recognizer](https://huggingface.co/software-mansion/react-native-executorch-recognizer-crnn.en) - recognizes English text
+
+For a list of all available OCR models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/ocr).
+
+---
+
+## Image embeddings
+
+- **CLIP_VIT_BASE_PATCH32_IMAGE** - [CLIP ViT Base Patch32](https://huggingface.co/software-mansion/react-native-executorch-clip-vit-base-patch32) for image embeddings
+
+For a list of all available Image embeddings models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/image-embeddings).
+
+---
+
+## Image segmentation
+
+- **DEEPLAB_V3_RESNET50** - [DeepLab V3](https://huggingface.co/software-mansion/react-native-executorch-deeplab-v3) with ResNet50 backbone
+
+For a list of all available Image Segmentation models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/image-segmentation).
+
+---
+
+## Text to image
+
+- **BK_SDM_TINY_VPRED_256** - BK-SDM Tiny V-Pred (256x256 resolution)
+
+For a list of all available Text to Image models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-to-image).
+
+---
+
+## Text embeddings
+
+- **ALL_MINILM_L6_V2** - [All-MiniLM-L6-v2](https://huggingface.co/software-mansion/react-native-executorch-all-MiniLM-L6-v2) text embeddings
+- **ALL_MPNET_BASE_V2** - [All-MPNet-Base-v2](https://huggingface.co/software-mansion/react-native-executorch-all-mpnet-base-v2) text embeddings
+- **MULTI_QA_MINILM_L6_COS_V1** - [Multi-QA MiniLM-L6](https://huggingface.co/software-mansion/react-native-executorch-multi-qa-MiniLM-L6-cos-v1) cosine similarity
+- **MULTI_QA_MPNET_BASE_DOT_V1** - Multi-QA MPNet Base dot product
+- **CLIP_VIT_BASE_PATCH32_TEXT** - [CLIP ViT Base Patch32](https://huggingface.co/software-mansion/react-native-executorch-clip-vit-base-patch32) for text embeddings
+
+For a list of all available Text embeddings models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-embeddings).
+
+---
+
+## Speech to text
+
+For a list of all available Speech to Text models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/speech-to-text).
+
+### Whisper Models (English only)
+
+- **WHISPER_TINY_EN** - [Whisper Tiny](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny.en)
+- **WHISPER_TINY_EN_QUANTIZED** - [Whisper Tiny English-only (quantized)](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny-quantized.en)
+- **WHISPER_BASE_EN** - [Whisper Base](https://huggingface.co/software-mansion/react-native-executorch-whisper-base.en)
+- **WHISPER_SMALL_EN** - [Whisper Small](https://huggingface.co/software-mansion/react-native-executorch-whisper-small.en)
+- **WHISPER_MEDIUM_EN** - [Whisper Medium](https://huggingface.co/software-mansion/react-native-executorch-whisper-medium.en)
+
+### Whisper Models (Multilingual)
+
+- **WHISPER_TINY** - [Whisper Tiny multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny)
+- **WHISPER_BASE** - [Whisper Base multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-base)
+- **WHISPER_SMALL** - [Whisper Small multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-small)
+- **WHISPER_MEDIUM** - [Whisper Medium multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-medium)
+
+### Other models
+
+- **MOONSHINE_TINY** - [Moonshine Tiny](https://huggingface.co/software-mansion/react-native-executorch-moonshine-tiny)
+
+---
+
+## Text to speech
+
+- **KOKORO_SMALL** - [Kokoro TTS](https://huggingface.co/software-mansion/react-native-executorch-kokoro) - Text-to-Speech model
+- **KOKORO_MEDIUM** - [Kokoro TTS](https://huggingface.co/software-mansion/react-native-executorch-kokoro) - Text-to-Speech model
+
+For a list of all available Text to Speech models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-to-speech).
+
+---
+
+## Voice activity detection
+
+- **FSMN_VAD** - [FSMN Voice Activity Detection](https://huggingface.co/software-mansion/react-native-executorch-fsmn-vad)
+
+For a list of all available VAD models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/voice-activity-detection).
+
+---
+
+# Loading models
+
+## Using predefined constants
+
+**When to use:** This is the recommended approach for all models that have [exported constants in the library](https://github.com/software-mansion/react-native-executorch/blob/main/packages/react-native-executorch/src/constants/modelUrls.ts). It provides type safety, automatic URL management, and ensures you're using the correct model configuration with all required files.
+
+**Benefits:** No need to manually specify URLs, automatic updates when model versions change, and guaranteed compatibility with the library version you're using.
+
+```typescript
+import {
+  useLLM,
+  LLAMA3_2_1B,
+  QWEN3_1_7B_QUANTIZED,
+} from 'react-native-executorch';
+
+const llama = useLLM(LLAMA3_2_1B);
+const qwen = useLLM(QWEN3_1_7B_QUANTIZED);
+```
+
+---
+
+## From React Native assets folder
+
+**When to use:** Best for small models (< 512MB) that you want to bundle with your app for offline use from the first launch. This ensures the model is immediately available without any download, but increases your app's installation size.
+
+**Trade-offs:** Larger app bundle size, but instant availability and guaranteed offline functionality. Good for demo apps or when your target users have limited internet connectivity.
+
+```typescript
+useExecutorchModule({
+  modelSource: require('../assets/llama3_2.pte'),
+});
+```
+
+---
+
+## From remote URL
+
+**When to use:** Ideal for large models (> 512MB) or when you want to keep your app's download size small. The model downloads on first use and is cached locally for subsequent uses. Perfect for production apps where initial app size matters.
+
+**Trade-offs:** Requires internet connection on first use, longer initial loading time, but keeps your app bundle small and allows for easy model updates without app resubmission.
+
+```typescript
+useExecutorchModule({
+  modelSource: 'https://.../llama3_2.pte',
+});
+```
+
+---
+
+## From local file system
+
+**When to use:** When you want to give users full control over model management, allow custom model imports, or enable advanced users to use their own fine-tuned models. Also useful for development and testing with local model files.
+
+**Trade-offs:** Requires implementing your own download/file management UI, but provides maximum flexibility and user control.
+
+```typescript
+useExecutorchModule({
+  modelSource: 'file:///var/mobile/.../llama3_2.pte',
+});
+```
+
+---
+
+# References
+
+- **Documentation:** https://docs.swmansion.com/react-native-executorch/docs/fundamentals/loading-models
+- **HuggingFace Repository:** https://huggingface.co/software-mansion
+- **Model Constants:** https://github.com/software-mansion/react-native-executorch/blob/main/packages/react-native-executorch/src/constants/modelUrls.ts
+- **All Collections:** https://huggingface.co/software-mansion/collections
+- **Model sizes reference:** https://docs.swmansion.com/react-native-executorch/docs/benchmarks/model-size
+- **Model memory usage reference:**: https://docs.swmansion.com/react-native-executorch/docs/benchmarks/memory-usage
+- **Model inference time benchmarks:** https://docs.swmansion.com/react-native-executorch/docs/benchmarks/inference-time
diff --git a/skills/react-native-executorch/references/reference-nlp.md b/skills/react-native-executorch/references/reference-nlp.md
new file mode 100644
index 000000000..2052686bc
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-nlp.md
@@ -0,0 +1,169 @@
+---
+title: Text Embeddings and Tokenizer
+description: Reference for using Text Embeddings and Tokenizer in React Native Executorch.
+---
+
+# useTextEmbeddings
+
+**Purpose:** Convert text into numerical vectors for semantic search, similarity, and clustering.
+
+**Use cases:** Semantic search, document similarity, text classification, clustering, RAG systems.
+
+## Basic Usage
+
+```typescript
+import { useTextEmbeddings, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+const model = useTextEmbeddings({ model: ALL_MINILM_L6_V2 });
+
+try {
+  const embedding = await model.forward('Hello World!');
+  console.log(embedding);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Example (computing similarity)
+
+```typescript
+const dotProduct = (a: number[], b: number[]) =>
+  a.reduce((sum, val, i) => sum + val * b[i], 0);
+
+const cosineSimilarity = (a: number[], b: number[]) => {
+  const dot = dotProduct(a, b);
+  const normA = Math.sqrt(dotProduct(a, a));
+  const normB = Math.sqrt(dotProduct(b, b));
+  return dot / (normA * normB);
+};
+
+function App() {
+  const model = useTextEmbeddings({ model: ALL_MINILM_L6_V2 });
+
+  // ...
+
+  try {
+    const helloWorldEmbedding = await model.forward('Hello World!');
+    const goodMorningEmbedding = await model.forward('Good Morning!');
+
+    const similarity = cosineSimilarity(
+      helloWorldEmbedding,
+      goodMorningEmbedding
+    );
+
+    console.log(`Cosine similarity: ${similarity}`);
+  } catch (error) {
+    console.error(error);
+  }
+
+  // ...
+}
+```
+
+## Available Models
+
+| Model                      | Max Tokens | Dimensions | Use Case                 |
+| -------------------------- | ---------- | ---------- | ------------------------ |
+| all-MiniLM-L6-v2           | 254        | 384        | General purpose          |
+| all-mpnet-base-v2          | 382        | 768        | Higher quality, slower   |
+| multi-qa-MiniLM-L6-cos-v1  | 509        | 384        | Q&A / semantic search    |
+| multi-qa-mpnet-base-dot-v1 | 510        | 768        | Q&A / semantic search    |
+| clip-vit-base-patch32-text | 74         | 512        | Match with images (CLIP) |
+
+**Model constants:** `ALL_MINILM_L6_V2`, `ALL_MPNET_BASE_V2`, `MULTI_QA_MINILM_L6_COS_V1`, `MULTI_QA_MPNET_BASE_DOT_V1`, `CLIP_VIT_BASE_PATCH32_TEXT`
+
+**Max Tokens** - The maximum number of tokens that can be processed by the model. If the input text exceeds this limit, it will be truncated.
+
+**Embedding Dimensions** - The size of the output embedding vector. This is the number of dimensions in the vector representation of the input text.
+
+For the latest available models reference exported models in [HuggingFace Text Embeddings collection](https://huggingface.co/collections/software-mansion/text-embeddings)
+
+## Troubleshooting
+
+**Normalized vectors:** For the supported models, the returned embedding vector is normalized, meaning that its length is equal to 1. This allows for easier comparison of vectors using cosine similarity, just calculate the dot product of two vectors to get the cosine similarity score.
+**Token limits:** Text exceeding max tokens will be truncated.
+
+## Additional references
+
+- [useTextEmbeddings docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/natural-language-processing/useTextEmbeddings)
+- [HuggingFace Text Embeddings collection](https://huggingface.co/collections/software-mansion/text-embeddings)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---text-embeddings)
+- [useTextEmbeddings API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextEmbeddings)
+- [Typescript API implementation of useTextEmbeddings](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TextEmbeddingsModule)
+
+---
+
+# useTokenizer
+
+**Purpose:** Convert text to tokens and vice versa (used internally by LLMs and embeddings).
+**Use cases:** Token counting, understanding model limits, debugging, custom preprocessing.
+
+## Basic Usage
+
+```typescript
+import { useTokenizer, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+const tokenizer = useTokenizer({ tokenizer: ALL_MINILM_L6_V2 });
+
+const text = 'Hello, world!';
+
+try {
+  // Tokenize the text
+  const tokens = await tokenizer.encode(text);
+  console.log('Tokens:', tokens);
+
+  // Decode the tokens back to text
+  const decodedText = await tokenizer.decode(tokens);
+  console.log('Decoded text:', decodedText);
+} catch (error) {
+  console.error('Error tokenizing text:', error);
+}
+```
+
+## Example usage
+
+```typescript
+import { useTokenizer, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+function App() {
+  const tokenizer = useTokenizer({ tokenizer: ALL_MINILM_L6_V2 });
+
+  // ...
+
+  try {
+    const text = 'Hello, world!';
+
+    const vocabSize = await tokenizer.getVocabSize();
+    console.log('Vocabulary size:', vocabSize);
+
+    const tokens = await tokenizer.encode(text);
+    console.log('Token IDs:', tokens);
+
+    const decoded = await tokenizer.decode(tokens);
+    console.log('Decoded text:', decoded);
+
+    const tokenId = await tokenizer.tokenToId('hello');
+    console.log('Token ID for "Hello":', tokenId);
+
+    const token = await tokenizer.idToToken(tokenId);
+    console.log('Token for ID:', token);
+  } catch (error) {
+    console.error(error);
+  }
+
+  // ...
+}
+```
+
+## Troubleshooting
+
+**Uses HuggingFace Tokenizers:** Full compatibility with HF ecosystem - we are using [Hugging Face Tokenizers](https://huggingface.co/docs/tokenizers/index) under the hood, ensuring compatibility with the Hugging Face ecosystem.
+
+**Mostly internal:** You usually don't need this directly - LLM/embedding hooks handle it.
+
+## Additional references
+
+- [useTokenizer docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/natural-language-processing/useTokenizer)
+- [useTokenizer API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTokenizer)
+- [HuggingFace Tokenizers documentation](https://huggingface.co/docs/tokenizers/index)
+- [Typescript API implementation of useTokenizer](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TokenizerModule)
diff --git a/skills/react-native-executorch/references/reference-ocr.md b/skills/react-native-executorch/references/reference-ocr.md
new file mode 100644
index 000000000..818478e57
--- /dev/null
+++ b/skills/react-native-executorch/references/reference-ocr.md
@@ -0,0 +1,171 @@
+---
+title: OCR usage
+description: Reference for using OCR and Vertical OCR.
+---
+
+# useOCR
+
+**Purpose:** Detect and recognize horizontal text within images (Optical Character Recognition).
+
+**Use cases:** Document scanning, receipt processing, business card reading, text extraction from photos.
+
+## Basic Usage
+
+```typescript
+import { useOCR, OCR_ENGLISH } from 'react-native-executorch';
+
+function App() {
+  const model = useOCR({ model: OCR_ENGLISH });
+
+  // ...
+  for (const ocrDetection of await model.forward('https://url-to-image.jpg')) {
+    console.log('Bounding box: ', ocrDetection.bbox);
+    console.log('Bounding label: ', ocrDetection.text);
+    console.log('Bounding score: ', ocrDetection.score);
+  }
+  // ...
+}
+```
+
+## Understanding the Detection Object
+
+```typescript
+interface Point {
+  x: number;
+  y: number;
+}
+
+interface OCRDetection {
+  bbox: Point[]; // 4 corner points of bounding box
+  text: string; // Recognized text
+  score: number; // Confidence score (0-1)
+}
+```
+
+The `bbox` array contains four points representing the corners of the detected text region, allowing for rotated or skewed text detection.
+
+## Language Support
+
+Different recognizer models support different alphabets and languages:
+
+```typescript
+import {
+  useOCR,
+  OCR_ENGLISH,
+  RECOGNIZER_LATIN_CRNN,
+  RECOGNIZER_CYRILLIC_CRNN,
+  DETECTOR_CRAFT,
+} from 'react-native-executorch';
+
+// For English (uses Latin alphabet)
+const englishOCR = useOCR({ model: OCR_ENGLISH });
+
+// For custom language configuration
+const customOCR = useOCR({
+  model: {
+    detectorSource: DETECTOR_CRAFT,
+    recognizerSource: RECOGNIZER_CYRILLIC_CRNN,
+    language: 'ru', // Russian
+  },
+});
+```
+
+**Important:** The recognizer model must match the alphabet of your target language. For example, use `RECOGNIZER_LATIN_CRNN` for English, Polish, German, etc., and `RECOGNIZER_CYRILLIC_CRNN` for Russian, Ukrainian, etc.
+
+## Available Models
+
+For all supported alphabets and languages, see [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets) and exported models in [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr).
+
+## Troubleshooting
+
+**Text orientation:** This hook is designed for horizontal text. For vertical text (e.g., traditional Chinese/Japanese), use `useVerticalOCR`.
+**Language/alphabet matching:** Ensure the recognizer model matches your target language's alphabet.
+**Image quality:** Higher resolution and better contrast improve recognition accuracy.
+
+## Additional references
+
+- [useOCR docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useOCR)
+- [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr)
+- [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets)
+- [useOCR API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useOCR)
+- [Typescript API implementation of useOCR](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/OCRModule)
+
+---
+
+# useVerticalOCR
+
+**Purpose:** Detect and recognize vertical text within images (experimental).
+
+**Use cases:** Traditional Chinese/Japanese text recognition, vertical signage, manga/comic text extraction.
+
+## Basic Usage
+
+```typescript
+import { useVerticalOCR, OCR_ENGLISH } from 'react-native-executorch';
+
+function App() {
+  const model = useVerticalOCR({
+    model: OCR_ENGLISH,
+    independentCharacters: true,
+  });
+
+  // ...
+  for (const ocrDetection of await model.forward('https://url-to-image.jpg')) {
+    console.log('Bounding box: ', ocrDetection.bbox);
+    console.log('Bounding label: ', ocrDetection.text);
+    console.log('Bounding score: ', ocrDetection.score);
+  }
+  // ...
+}
+```
+
+## Character vs Word Mode
+
+The `independentCharacters` parameter controls how text is processed:
+
+```typescript
+// Character mode - each character detected separately (better for CJK)
+const charMode = useVerticalOCR({
+  model: OCR_CHINESE,
+  independentCharacters: true,
+});
+
+// Word mode - characters grouped into words (better for Latin alphabets)
+const wordMode = useVerticalOCR({
+  model: OCR_ENGLISH,
+  independentCharacters: false,
+});
+```
+
+## Understanding the Detection Object
+
+```typescript
+interface Point {
+  x: number;
+  y: number;
+}
+
+interface OCRDetection {
+  bbox: Point[]; // 4 corner points of bounding box
+  text: string; // Recognized text
+  score: number; // Confidence score (0-1)
+}
+```
+
+## Available Models
+
+For all supported alphabets and languages, see [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets) and exported models in [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr).
+
+## Troubleshooting
+
+**Experimental status:** This hook is experimental and may have limitations with certain text layouts or languages.
+**Character vs word mode:** Use `independentCharacters: true` for CJK languages, `false` for Latin alphabets.
+**Alphabet matching:** Ensure the recognizer matches your target language's alphabet.
+
+## Additional references
+
+- [useVerticalOCR docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useVerticalOCR)
+- [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr)
+- [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets)
+- [useVerticalOCR API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVerticalOCR)
+- [Typescript API implementation of useVerticalOCR](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/VerticalOCRModule)

From ad0228499f266f5f18949aedb8dc69de75030205 Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Tue, 10 Feb 2026 21:00:13 +0100
Subject: [PATCH 2/7] chore: add language to code block

---
 skills/react-native-executorch/references/reference-llms.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/skills/react-native-executorch/references/reference-llms.md b/skills/react-native-executorch/references/reference-llms.md
index 1497765fc..36e41c686 100644
--- a/skills/react-native-executorch/references/reference-llms.md
+++ b/skills/react-native-executorch/references/reference-llms.md
@@ -69,7 +69,7 @@ console.log(llm.messageHistory);
 
 ## Tool Calling
 
-```
+```tsx
 const TOOL_DEFINITIONS: LLMTool[] = [
   {
     name: 'get_weather',

From 973e3661ab0e29822334794e29f602b78f9a21a8 Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Tue, 10 Feb 2026 22:56:09 +0100
Subject: [PATCH 3/7] chore: add main SKILL.md file

---
 .cspell-wordlist.txt                    |   2 +
 skills/react-native-executorch/SKILL.md | 660 ++++++++++++++++++++++++
 2 files changed, 662 insertions(+)
 create mode 100644 skills/react-native-executorch/SKILL.md

diff --git a/.cspell-wordlist.txt b/.cspell-wordlist.txt
index 283af1481..2a6a720bd 100644
--- a/.cspell-wordlist.txt
+++ b/.cspell-wordlist.txt
@@ -6,6 +6,8 @@ torch
 huggingface
 bbox
 bboxes
+chatbots
+llms
 deeplab
 unsqueeze
 qlora
diff --git a/skills/react-native-executorch/SKILL.md b/skills/react-native-executorch/SKILL.md
new file mode 100644
index 000000000..b9f0f60ba
--- /dev/null
+++ b/skills/react-native-executorch/SKILL.md
@@ -0,0 +1,660 @@
+---
+name: react-native-executorch
+description: Build on-device AI into React Native apps using ExecuTorch. Provides hooks for LLMs, computer vision, OCR, audio processing, and embeddings without cloud dependencies. Use when building AI features into mobile apps - AI chatbots, image recognition, speech processing, or text search.
+---
+
+## When to Use This Skill
+
+Use this skill when you need to:
+
+- **Build AI features directly into mobile apps** without cloud infrastructure
+- **Deploy LLMs locally** for text generation, chat, or function calling
+- **Add computer vision** (image classification, object detection, OCR)
+- **Process audio** (speech-to-text, text-to-speech, voice activity detection)
+- **Implement semantic search** with text embeddings
+- **Ensure privacy** by keeping all AI processing on-device
+- **Reduce latency** by eliminating cloud API calls
+- **Work offline** once models are downloaded
+
+## Overview
+
+React Native Executorch is a library that enables on-device AI model execution in React Native applications. It provides hooks and utilities for running machine learning models directly on mobile devices without requiring cloud infrastructure or internet connectivity (after initial model download).
+
+## Key Use Cases
+
+### Use Case 1: Mobile Chatbot/Assistant
+
+**Trigger:** User asks to build a chat interface, create a conversational AI, or add an AI assistant to their app
+
+**Steps:**
+
+1. Choose appropriate LLM based on device memory constraints
+2. Load model using ExecuTorch hooks
+3. Implement message handling and conversation history
+4. Optionally add system prompts, tool calling, or structured output
+
+**Result:** Functional chat interface with on-device AI responding without cloud dependency
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md)
+
+---
+
+### Use Case 2: Image Recognition & Tagging
+
+**Trigger:** User needs to classify images, detect objects, or recognize content in photos
+
+**Steps:**
+
+1. Select vision model (classification, detection, or segmentation)
+2. Load model for image processing task
+3. Pass image URI and process results
+4. Display detections or classifications in app UI
+
+**Result:** App that understands image content without sending data to servers
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md)
+
+---
+
+### Use Case 3: Document/Receipt Scanning
+
+**Trigger:** User wants to extract text from photos (receipts, documents, business cards)
+
+**Steps:**
+
+1. Choose OCR model matching target language
+2. Load appropriate recognizer for alphabet/language
+3. Capture or load image
+4. Extract text regions with bounding boxes
+5. Post-process results for application
+
+**Result:** OCR-enabled app that reads text directly from device camera
+
+**Reference:** [./references/reference-ocr.md](./references/reference-ocr.md)
+
+---
+
+### Use Case 4: Voice Interface
+
+**Trigger:** User wants to add voice commands, transcription, or voice output to app
+
+**Steps:**
+
+- **For voice input:** Capture audio at correct sample rate → transcribe with STT model
+- **For voice output:** Generate speech from text → play through audio context
+- Handle audio format/sample rate conversion
+
+**Result:** App with hands-free voice interaction
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+---
+
+### Use Case 5: Semantic Search
+
+**Trigger:** User needs intelligent search, similarity matching, or content recommendations
+
+**Steps:**
+
+1. Load text or image embeddings model
+2. Generate embeddings for searchable content
+3. Compute similarity scores between queries and content
+4. Rank and return results
+
+**Result:** Smart search that understands meaning, not just keywords
+
+**Reference:** [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Core Capabilities by Category
+
+### Large Language Models (LLMs)
+
+Run text generation, chat, function calling, and structured output generation locally on-device.
+
+**Supported features:**
+
+- Text generation and chat completions
+- Function/tool calling
+- Structured output with JSON schema validation
+- Streaming responses
+- Multiple model families (Llama 3.2, Qwen 3, Hammer 2.1, SmolLM2, Phi 4)
+
+**Reference:** See [./references/reference-llms.md](./references/reference-llms.md)
+
+---
+
+### Computer Vision
+
+Perform image understanding and manipulation tasks entirely on-device.
+
+**Supported tasks:**
+
+- **Image Classification** - Categorize images into predefined classes
+- **Object Detection** - Locate and identify objects with bounding boxes
+- **Image Segmentation** - Pixel-level classification
+- **Style Transfer** - Apply artistic styles to images
+- **Text-to-Image** - Generate images from text descriptions
+- **Image Embeddings** - Convert images to numerical vectors for similarity/search
+
+**Reference:** See [./references/reference-cv.md](./references/reference-cv.md) and [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+---
+
+### Optical Character Recognition (OCR)
+
+Extract and recognize text from images with support for multiple languages and text orientations.
+
+**Supported features:**
+
+- Text detection in images
+- Text recognition across different alphabets
+- Horizontal text (standard documents, receipts)
+- Vertical text support (experimental, for CJK languages)
+- Multi-language support with language-specific recognizers
+
+**Reference:** See [./references/reference-ocr.md](./references/reference-ocr.md)
+
+---
+
+### Audio Processing
+
+Convert between speech and text, and detect speech activity in audio.
+
+**Supported tasks:**
+
+- **Speech-to-Text** - Transcribe audio to text (supports English and multilingual)
+- **Text-to-Speech** - Generate natural-sounding speech from text
+- **Voice Activity Detection** - Detect speech segments in audio
+
+**Reference:** See [./references/reference-audio.md](./references/reference-audio.md)
+
+---
+
+### Natural Language Processing
+
+Convert text to numerical representations for semantic understanding and search.
+
+**Supported tasks:**
+
+- **Text Embeddings** - Convert text to vectors for similarity/search
+- **Tokenization** - Convert text to tokens and vice versa
+
+**Reference:** See [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Getting Started by Use Case
+
+### I want to build a chatbot or AI assistant
+
+Use `useLLM` hook with one of the available language models.
+
+**What to do:**
+
+1. Choose a model from available LLM options (consider device memory constraints)
+2. Use the `useLLM` hook to load the model
+3. Send messages and receive responses
+4. Optionally configure system prompts, generation parameters, and tools
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - LLMs section
+
+---
+
+### I want to enable function/tool calling in my LLM
+
+Use `useLLM` with tool definitions to allow the model to call predefined functions.
+
+**What to do:**
+
+1. Define tools with name, description, and parameter schema
+2. Configure the LLM with tool definitions
+3. Implement callbacks to execute tools when the model requests them
+4. Parse tool results and pass them back to the model
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md) - Tool Calling section
+
+---
+
+### I want structured data extraction from text
+
+Use `useLLM` with structured output generation using JSON schema validation.
+
+**What to do:**
+
+1. Define a schema (JSON Schema or Zod) for desired output format
+2. Configure the LLM with the schema
+3. Generate responses and validate against the schema
+4. Use the validated structured data in your app
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md) - Structured Output section
+
+---
+
+### I want to classify or recognize objects in images
+
+Use `useClassification` for simple categorization or `useObjectDetection` for locating specific objects.
+
+**What to do:**
+
+1. Choose appropriate computer vision model based on task
+2. Load the model with the appropriate hook
+3. Pass image URI (local, remote, or base64)
+4. Process results (classifications, detections with bounding boxes)
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Classification and Object Detection sections
+
+---
+
+### I want to extract text from images
+
+Use `useOCR` for horizontal text or `useVerticalOCR` for vertical text (experimental).
+
+**What to do:**
+
+1. Choose appropriate OCR model and recognizer matching your target language
+2. Load the model with `useOCR` or `useVerticalOCR` hook
+3. Pass image URI
+4. Extract detected text regions with bounding boxes and confidence scores
+5. Process results based on your application needs
+
+**Reference:** [./references/reference-ocr.md](./references/reference-ocr.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - OCR section
+
+---
+
+### I want to convert speech to text or text to speech
+
+Use `useSpeechToText` for transcription or `useTextToSpeech` for voice synthesis.
+
+**What to do:**
+
+- **For Speech-to-Text:** Capture or load audio, ensure 16kHz sample rate, transcribe
+- **For Text-to-Speech:** Prepare text, specify voice parameters, generate audio waveform, play using audio context
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Speech to Text and Text to Speech sections
+
+---
+
+### I want to find similar images or texts
+
+Use `useImageEmbeddings` for images or `useTextEmbeddings` for text.
+
+**What to do:**
+
+1. Load appropriate embeddings model
+2. Generate embeddings for your content
+3. Compute similarity metrics (cosine similarity, dot product)
+4. Use similarity scores for search, clustering, or deduplication
+
+**Reference:**
+
+- Text: [./references/reference-nlp.md](./references/reference-nlp.md)
+- Images: [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+---
+
+### I want to apply artistic filters to photos
+
+Use `useStyleTransfer` to apply predefined artistic styles to images.
+
+**What to do:**
+
+1. Choose from available artistic styles (Candy, Mosaic, Udnie, Rain Princess)
+2. Load the style transfer model
+3. Pass image URI
+4. Retrieve and use the stylized image
+
+**Reference:** [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Style Transfer section
+
+---
+
+### I want to generate images from text
+
+Use `useTextToImage` to create images based on text descriptions.
+
+**What to do:**
+
+1. Load the text-to-image model
+2. Provide text description (prompt)
+3. Optionally specify image size and number of generation steps
+4. Receive generated image (may take 20-60 seconds depending on device)
+
+**Reference:** [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Text to Image section
+
+---
+
+## Understanding Model Loading
+
+Before using any AI model, you need to load it. Models can be loaded from three sources:
+
+**1. Bundled with app (assets folder)**
+
+- Best for small models (< 512MB)
+- Available immediately without download
+- Increases app installation size
+
+**2. Remote URL (downloaded on first use)**
+
+- Best for large models (> 512MB)
+- Downloaded once and cached locally
+- Keeps app size small
+- Requires internet on first use
+
+**3. Local file system**
+
+- Maximum flexibility for user-managed models
+- Requires custom download/file management UI
+
+**Model selection strategy:**
+
+1. Small models (< 512MB) → Bundle with app or download from URL
+2. Large models (> 512MB) → Download from URL on first use with progress tracking
+3. Quantized models → Preferred for lower-end devices to save memory
+
+**Reference:** [./references/reference-models.md](./references/reference-models.md) - Loading Models section
+
+---
+
+## Device Constraints and Model Selection
+
+Not all models work on all devices. Consider these constraints:
+
+**Memory limitations:**
+
+- Low-end devices: Use smaller models (135M-1.7B parameters) and quantized variants
+- High-end devices: Can run larger models (3B-4B parameters)
+
+**Processing power:**
+
+- Lower-end devices: Expect longer inference times
+- Audio processing requires specific sample rates (16kHz for STT, 24kHz for TTS output)
+
+**Storage:**
+
+- Large models require significant disk space
+- Implement cleanup mechanisms to remove unused models
+- Monitor total downloaded model size
+
+**Guidance:**
+
+- Always check model memory requirements before recommending models
+- Prefer quantized model variants on lower-end devices
+- Show download progress for models > 512MB
+- Test on target devices before release
+
+**Reference:** [./references/reference-models.md](./references/reference-models.md)
+
+---
+
+## Important Technical Requirements
+
+### Audio Processing
+
+Audio must be in correct sample rate for processing:
+
+- **Speech-to-Text input:** 16kHz sample rate
+- **Text-to-Speech output:** 24kHz sample rate
+- Always decode/resample audio to correct rate before processing
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+### Image Processing
+
+Images can be provided as:
+
+- Remote URLs (http/https) - automatically cached
+- Local file URIs (file://)
+- Base64-encoded strings
+
+Image preprocessing (resizing, normalization) is handled automatically by most hooks.
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md) and [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+### Text Tokens
+
+Text embeddings and LLMs have maximum token limits. Text exceeding these limits will be truncated. Use `useTokenizer` to count tokens before processing.
+
+**Reference:** [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Core Utilities and Error Handling
+
+The library provides core utilities for managing models and handling errors:
+
+**ResourceFetcher:** Manage model downloads with pause/resume capabilities, storage cleanup, and progress tracking.
+
+**Error Handling:** Use `RnExecutorchError` and error codes for robust error handling and user feedback.
+
+**useExecutorchModule:** Low-level API for custom models not covered by dedicated hooks.
+
+**Reference:** [./references/core-utilities.md](./references/core-utilities.md)
+
+---
+
+## Common Troubleshooting
+
+**Model not loading:** Check model source URL/path validity and sufficient device storage
+
+**Out of memory errors:** Switch to smaller model or quantized variant
+
+**Poor LLM quality:** Adjust temperature/top-p parameters or improve system prompt
+
+**Audio issues:** Verify correct sample rate (16kHz for STT, 24kHz output for TTS)
+
+**Download failures:** Implement retry logic and check network connectivity
+
+**Reference:** [./references/core-utilities.md](./references/core-utilities.md) for error handling details, or specific reference file for your use case
+
+---
+
+## Quick Reference by Hook
+
+| Hook                   | Purpose                                   | Reference                                             |
+| ---------------------- | ----------------------------------------- | ----------------------------------------------------- |
+| `useLLM`               | Text generation, chat, function calling   | [reference-llms.md](./references/reference-llms.md)   |
+| `useClassification`    | Image categorization                      | [reference-cv.md](./references/reference-cv.md)       |
+| `useObjectDetection`   | Object localization                       | [reference-cv.md](./references/reference-cv.md)       |
+| `useImageSegmentation` | Pixel-level classification                | [reference-cv.md](./references/reference-cv.md)       |
+| `useStyleTransfer`     | Artistic image filters                    | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useTextToImage`       | Image generation                          | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useImageEmbeddings`   | Image similarity/search                   | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useOCR`               | Text recognition (horizontal)             | [reference-ocr.md](./references/reference-ocr.md)     |
+| `useVerticalOCR`       | Text recognition (vertical, experimental) | [reference-ocr.md](./references/reference-ocr.md)     |
+| `useSpeechToText`      | Audio transcription                       | [reference-audio.md](./references/reference-audio.md) |
+| `useTextToSpeech`      | Voice synthesis                           | [reference-audio.md](./references/reference-audio.md) |
+| `useVAD`               | Voice activity detection                  | [reference-audio.md](./references/reference-audio.md) |
+| `useTextEmbeddings`    | Text similarity/search                    | [reference-nlp.md](./references/reference-nlp.md)     |
+| `useTokenizer`         | Text to tokens conversion                 | [reference-nlp.md](./references/reference-nlp.md)     |
+| `useExecutorchModule`  | Custom model inference (advanced)         | [core-utilities.md](./references/core-utilities.md)   |
+
+---
+
+## Quick Checklist for Implementation
+
+Use this when building AI features with ExecuTorch:
+
+**Planning Phase**
+
+- Identified what AI task you need (chat, vision, audio, search)
+- Considered device memory constraints and target devices
+- Chose appropriate model from available options
+- Determined if cloud backup fallback is needed
+
+**Development Phase**
+
+- Selected correct hook for your task
+- Configured model loading (bundled, remote URL, or local)
+- Implemented proper error handling
+- Added loading states for model operations
+- Tested audio sample rates (if audio task)
+- Set up resource management for large models
+
+**Testing Phase**
+
+- Tested on target minimum device
+- Verified offline functionality works
+- Checked memory usage doesn't exceed device limits
+- Tested error handling (network, memory, invalid inputs)
+- Measured inference time for acceptable UX
+
+**Deployment Phase**
+
+- Model bundling strategy decided (size/download tradeoff)
+- Download progress UI implemented (if remote models)
+- Version management plan for model updates
+- User feedback mechanism for quality issues
+
+---
+
+## Reference Files Overview
+
+**[reference-llms.md](./references/reference-llms.md)**
+
+- Complete LLM hook documentation
+- Functional vs Managed modes
+- Tool calling implementation
+- Structured output generation
+
+**[reference-cv.md](./references/reference-cv.md)**
+
+- Image classification, detection, and segmentation
+- Basic computer vision tasks
+
+**[reference-cv-2.md](./references/reference-cv-2.md)**
+
+- Advanced vision tasks: style transfer, text-to-image, embeddings
+- Image similarity and search
+
+**[reference-ocr.md](./references/reference-ocr.md)**
+
+- Horizontal and vertical text recognition
+- Multi-language support
+- OCR model selection
+
+**[reference-audio.md](./references/reference-audio.md)**
+
+- Speech-to-text transcription
+- Text-to-speech voice synthesis
+- Voice activity detection
+- Audio sample rate requirements
+
+**[reference-nlp.md](./references/reference-nlp.md)**
+
+- Text embeddings for semantic search
+- Tokenization utilities
+- Token limits and model compatibility
+
+**[reference-models.md](./references/reference-models.md)**
+
+- Complete list of available models
+- Model loading strategies
+- Model selection guidelines
+- Device memory/performance considerations
+
+**[core-utilities.md](./references/core-utilities.md)**
+
+- ResourceFetcher for download management
+- Error handling with RnExecutorchError
+- Low-level useExecutorchModule API
+- Error codes reference
+
+---
+
+## Troubleshooting Guide
+
+**Model not loading or crashing**
+
+- Check model source (URL valid, file exists)
+- Verify device has sufficient free storage and memory
+- Try bundling smaller models first
+- Check error codes with `RnExecutorchError`
+
+**Out of memory errors**
+
+- Switch to quantized model variant (smaller file size)
+- Use smaller parameter model (135M instead of 1.7B)
+- Close other apps to free device memory
+- Implement model unloading when not in use
+
+**Poor quality results from LLM**
+
+- Adjust generation parameters (temperature, top-p)
+- Improve system prompt
+- Try larger model if device supports it
+- Check input preprocessing
+
+**Audio not processing**
+
+- Verify sample rate is 16kHz for STT, 24kHz output for TTS
+- Check audio format compatibility
+- Ensure audio buffer has data before processing
+- Validate microphone permissions
+
+**Slow inference speed**
+
+- Expected on lower-end devices (especially larger models)
+- Show loading indicator to user
+- Consider preprocessing optimization
+- Profile on actual target device
+
+---
+
+## Best Practices
+
+**Model Selection**
+
+- Match model size to device capabilities
+- Use quantized variants for memory-constrained devices
+- Test on minimum target device before release
+- Keep models updated via download mechanism
+
+**Error Handling**
+
+- Always wrap AI operations in try-catch
+- Provide user-friendly error messages
+- Implement fallback behavior (cloud API, simplified UX)
+- Log errors for debugging
+
+**User Experience**
+
+- Show loading states during model operations
+- Display download progress for large models
+- Ensure app remains responsive during inference
+- Consider offline-first design
+
+**Resource Management**
+
+- Unload unused models to free memory
+- Implement cleanup for old cached models
+- Show storage impact of AI features
+- Monitor battery usage of continuous processing
+
+**Performance Optimization**
+
+- Batch requests when possible
+- Preload models during idle time
+- Profile actual device performance before launch
+- Use appropriate model size for each task
+
+---
+
+## External Resources
+
+- **Official Documentation:** https://docs.swmansion.com/react-native-executorch
+- **HuggingFace Models:** https://huggingface.co/software-mansion/collections
+- **GitHub Repository:** https://github.com/software-mansion/react-native-executorch
+- **API Reference:** https://docs.swmansion.com/react-native-executorch/docs/api-reference

From d70c3f3735c278042cb522f0529fa99934af0705 Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Wed, 11 Feb 2026 18:10:38 +0100
Subject: [PATCH 4/7] chore: apply review suggestions

---
 skills/react-native-executorch/SKILL.md | 49 +++++++++++++------------
 1 file changed, 25 insertions(+), 24 deletions(-)

diff --git a/skills/react-native-executorch/SKILL.md b/skills/react-native-executorch/SKILL.md
index b9f0f60ba..3d1768095 100644
--- a/skills/react-native-executorch/SKILL.md
+++ b/skills/react-native-executorch/SKILL.md
@@ -1,6 +1,6 @@
 ---
 name: react-native-executorch
-description: Build on-device AI into React Native apps using ExecuTorch. Provides hooks for LLMs, computer vision, OCR, audio processing, and embeddings without cloud dependencies. Use when building AI features into mobile apps - AI chatbots, image recognition, speech processing, or text search.
+description: Integrate on-device AI into React Native apps using React Native ExecuTorch, which provides APIs for LLMs, computer vision, OCR, audio processing, and embeddings without cloud dependencies, as well as a variety of pre-exported models for common use cases. Use when user asks to build AI features into mobile apps - AI chatbots, image classification, object detection, style transfer, OCR, document parsing, speech processing, or semantic search - all running locally without cloud dependencies. Use when user mentions offline support, privacy, latency or cost concerns in AI-based applications.
 ---
 
 ## When to Use This Skill
@@ -18,7 +18,7 @@ Use this skill when you need to:
 
 ## Overview
 
-React Native Executorch is a library that enables on-device AI model execution in React Native applications. It provides hooks and utilities for running machine learning models directly on mobile devices without requiring cloud infrastructure or internet connectivity (after initial model download).
+[React Native Executorch](https://github.com/software-mansion/react-native-executorch) is a library developed by [Software Mansion](https://swmansion.com/) that enables on-device AI model execution in React Native applications. It provides APIs for running machine learning models directly on mobile devices without requiring cloud infrastructure or internet connectivity (after initial model download). React Native Executorch provides APIs for LLMs, computer vision, OCR, audio processing and embeddings without cloud dependencies, as well as a variety of pre-exported models for common use cases. React Native Executorch is a way of bringing ExecuTorch into the React Native world.
 
 ## Key Use Cases
 
@@ -39,7 +39,7 @@ React Native Executorch is a library that enables on-device AI model execution i
 
 ---
 
-### Use Case 2: Image Recognition & Tagging
+### Use Case 2: Computer Vision
 
 **Trigger:** User needs to classify images, detect objects, or recognize content in photos
 
@@ -47,7 +47,7 @@ React Native Executorch is a library that enables on-device AI model execution i
 
 1. Select vision model (classification, detection, or segmentation)
 2. Load model for image processing task
-3. Pass image URI and process results
+3. Pass image and process results
 4. Display detections or classifications in app UI
 
 **Result:** App that understands image content without sending data to servers
@@ -164,7 +164,7 @@ Convert between speech and text, and detect speech activity in audio.
 
 **Supported tasks:**
 
-- **Speech-to-Text** - Transcribe audio to text (supports English and multilingual)
+- **Speech-to-Text** - Transcribe audio to text (supports multiple languages including English)
 - **Text-to-Speech** - Generate natural-sounding speech from text
 - **Voice Activity Detection** - Detect speech segments in audio
 
@@ -189,12 +189,12 @@ Convert text to numerical representations for semantic understanding and search.
 
 ### I want to build a chatbot or AI assistant
 
-Use `useLLM` hook with one of the available language models.
+Use `useLLM` hook or `LLMModule` with one of the available language models.
 
 **What to do:**
 
 1. Choose a model from available LLM options (consider device memory constraints)
-2. Use the `useLLM` hook to load the model
+2. Use the `useLLM` hook or `LLMModule` to load the model
 3. Send messages and receive responses
 4. Optionally configure system prompts, generation parameters, and tools
 
@@ -206,7 +206,7 @@ Use `useLLM` hook with one of the available language models.
 
 ### I want to enable function/tool calling in my LLM
 
-Use `useLLM` with tool definitions to allow the model to call predefined functions.
+Use `useLLM` hook or `LLMModule` with tool definitions to allow the model to call predefined functions.
 
 **What to do:**
 
@@ -221,7 +221,7 @@ Use `useLLM` with tool definitions to allow the model to call predefined functio
 
 ### I want structured data extraction from text
 
-Use `useLLM` with structured output generation using JSON schema validation.
+Use `useLLM` hook or `LLMModule` with structured output generation using JSON schema validation.
 
 **What to do:**
 
@@ -236,13 +236,13 @@ Use `useLLM` with structured output generation using JSON schema validation.
 
 ### I want to classify or recognize objects in images
 
-Use `useClassification` for simple categorization or `useObjectDetection` for locating specific objects.
+Use `useClassification` hook or `ClassificationModule` for simple categorization or use `useObjectDetection` hook or `ObjectDetectionModule` for locating specific objects.
 
 **What to do:**
 
 1. Choose appropriate computer vision model based on task
-2. Load the model with the appropriate hook
-3. Pass image URI (local, remote, or base64)
+2. Load the model with the appropriate hook or module
+3. Pass image (local, remote, or base64)
 4. Process results (classifications, detections with bounding boxes)
 
 **Reference:** [./references/reference-cv.md](./references/reference-cv.md)
@@ -253,13 +253,13 @@ Use `useClassification` for simple categorization or `useObjectDetection` for lo
 
 ### I want to extract text from images
 
-Use `useOCR` for horizontal text or `useVerticalOCR` for vertical text (experimental).
+Use `useOCR` hook or `OCRModule` for horizontal text or use `useVerticalOCR` hook or `VerticalOCRModule` for vertical text (experimental).
 
 **What to do:**
 
 1. Choose appropriate OCR model and recognizer matching your target language
-2. Load the model with `useOCR` or `useVerticalOCR` hook
-3. Pass image URI
+2. Load the model with appropriate hook or module
+3. Pass image
 4. Extract detected text regions with bounding boxes and confidence scores
 5. Process results based on your application needs
 
@@ -271,7 +271,7 @@ Use `useOCR` for horizontal text or `useVerticalOCR` for vertical text (experime
 
 ### I want to convert speech to text or text to speech
 
-Use `useSpeechToText` for transcription or `useTextToSpeech` for voice synthesis.
+Use `useSpeechToText` hook or `SpeechToTextModule` for transcription or use `useTextToSpeech` hook or `TextToSpeechModule` for voice synthesis.
 
 **What to do:**
 
@@ -286,7 +286,7 @@ Use `useSpeechToText` for transcription or `useTextToSpeech` for voice synthesis
 
 ### I want to find similar images or texts
 
-Use `useImageEmbeddings` for images or `useTextEmbeddings` for text.
+Use `useImageEmbeddings` hook or `ImageEmbeddingsModule` for images or `useTextEmbeddings` hook or `TextEmbeddingsModule` for text.
 
 **What to do:**
 
@@ -304,13 +304,13 @@ Use `useImageEmbeddings` for images or `useTextEmbeddings` for text.
 
 ### I want to apply artistic filters to photos
 
-Use `useStyleTransfer` to apply predefined artistic styles to images.
+Use `useStyleTransfer` hook or `StyleTransferModule` to apply predefined artistic styles to images.
 
 **What to do:**
 
 1. Choose from available artistic styles (Candy, Mosaic, Udnie, Rain Princess)
 2. Load the style transfer model
-3. Pass image URI
+3. Pass image
 4. Retrieve and use the stylized image
 
 **Reference:** [./references/reference-cv-2.md](./references/reference-cv-2.md)
@@ -321,7 +321,7 @@ Use `useStyleTransfer` to apply predefined artistic styles to images.
 
 ### I want to generate images from text
 
-Use `useTextToImage` to create images based on text descriptions.
+Use `useTextToImage` hook or `TextToImageModule` to create images based on text descriptions.
 
 **What to do:**
 
@@ -380,7 +380,7 @@ Not all models work on all devices. Consider these constraints:
 **Processing power:**
 
 - Lower-end devices: Expect longer inference times
-- Audio processing requires specific sample rates (16kHz for STT, 24kHz for TTS output)
+- Audio processing requires specific sample rates (16kHz for STT and VAD, 24kHz for TTS output)
 
 **Storage:**
 
@@ -405,7 +405,7 @@ Not all models work on all devices. Consider these constraints:
 
 Audio must be in correct sample rate for processing:
 
-- **Speech-to-Text input:** 16kHz sample rate
+- **Speech-to-Text or VAD input:** 16kHz sample rate
 - **Text-to-Speech output:** 24kHz sample rate
 - Always decode/resample audio to correct rate before processing
 
@@ -453,7 +453,7 @@ The library provides core utilities for managing models and handling errors:
 
 **Poor LLM quality:** Adjust temperature/top-p parameters or improve system prompt
 
-**Audio issues:** Verify correct sample rate (16kHz for STT, 24kHz output for TTS)
+**Audio issues:** Verify correct sample rate (16kHz for STT and VAD, 24kHz output for TTS)
 
 **Download failures:** Implement retry logic and check network connectivity
 
@@ -599,7 +599,7 @@ Use this when building AI features with ExecuTorch:
 
 **Audio not processing**
 
-- Verify sample rate is 16kHz for STT, 24kHz output for TTS
+- Verify sample rate is 16kHz for STT and VAD, 24kHz output for TTS
 - Check audio format compatibility
 - Ensure audio buffer has data before processing
 - Validate microphone permissions
@@ -658,3 +658,4 @@ Use this when building AI features with ExecuTorch:
 - **HuggingFace Models:** https://huggingface.co/software-mansion/collections
 - **GitHub Repository:** https://github.com/software-mansion/react-native-executorch
 - **API Reference:** https://docs.swmansion.com/react-native-executorch/docs/api-reference
+- **Software Mansion:** https://swmansion.com/

From 1a14d6586b60dcdf47a6bb28304dfbe6962416ef Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Mon, 16 Feb 2026 12:02:06 +0100
Subject: [PATCH 5/7] add .canary (internal) version of claude skill

---
 .../.canary/react-native-executorch/SKILL.md  | 663 ++++++++++++++++++
 .../references/core-utilities.md              | 327 +++++++++
 .../references/reference-audio.md             | 433 ++++++++++++
 .../references/reference-cv-2.md              | 190 +++++
 .../references/reference-cv.md                | 215 ++++++
 .../references/reference-llms.md              | 259 +++++++
 .../references/reference-models.md            | 265 +++++++
 .../references/reference-nlp.md               | 169 +++++
 .../references/reference-ocr.md               | 171 +++++
 9 files changed, 2692 insertions(+)
 create mode 100644 skills/.canary/react-native-executorch/SKILL.md
 create mode 100644 skills/.canary/react-native-executorch/references/core-utilities.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-audio.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-cv-2.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-cv.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-llms.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-models.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-nlp.md
 create mode 100644 skills/.canary/react-native-executorch/references/reference-ocr.md

diff --git a/skills/.canary/react-native-executorch/SKILL.md b/skills/.canary/react-native-executorch/SKILL.md
new file mode 100644
index 000000000..8a424eb82
--- /dev/null
+++ b/skills/.canary/react-native-executorch/SKILL.md
@@ -0,0 +1,663 @@
+---
+name: react-native-executorch
+description: Integrate on-device AI into React Native apps using React Native ExecuTorch, which provides APIs for LLMs, computer vision, OCR, audio processing, and embeddings without cloud dependencies, as well as a variety of pre-exported models for common use cases. Use when user asks to build AI features into mobile apps - AI chatbots, image classification, object detection, style transfer, OCR, document parsing, speech processing, or semantic search - all running locally without cloud dependencies. Use when user mentions offline support, privacy, latency or cost concerns in AI-based applications.
+metadata:
+  internal: true
+---
+
+## When to Use This Skill
+
+Use this skill when you need to:
+
+- **Build AI features directly into mobile apps** without cloud infrastructure
+- **Deploy LLMs locally** for text generation, chat, or function calling
+- **Add computer vision** (image classification, object detection, OCR)
+- **Process audio** (speech-to-text, text-to-speech, voice activity detection)
+- **Implement semantic search** with text embeddings
+- **Ensure privacy** by keeping all AI processing on-device
+- **Reduce latency** by eliminating cloud API calls
+- **Work offline** once models are downloaded
+
+## Overview
+
+[React Native Executorch](https://github.com/software-mansion/react-native-executorch) is a library developed by [Software Mansion](https://swmansion.com/) that enables on-device AI model execution in React Native applications. It provides APIs for running machine learning models directly on mobile devices without requiring cloud infrastructure or internet connectivity (after initial model download). React Native Executorch provides APIs for LLMs, computer vision, OCR, audio processing and embeddings without cloud dependencies, as well as a variety of pre-exported models for common use cases. React Native Executorch is a way of bringing ExecuTorch into the React Native world.
+
+## Key Use Cases
+
+### Use Case 1: Mobile Chatbot/Assistant
+
+**Trigger:** User asks to build a chat interface, create a conversational AI, or add an AI assistant to their app
+
+**Steps:**
+
+1. Choose appropriate LLM based on device memory constraints
+2. Load model using ExecuTorch hooks
+3. Implement message handling and conversation history
+4. Optionally add system prompts, tool calling, or structured output
+
+**Result:** Functional chat interface with on-device AI responding without cloud dependency
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md)
+
+---
+
+### Use Case 2: Computer Vision
+
+**Trigger:** User needs to classify images, detect objects, or recognize content in photos
+
+**Steps:**
+
+1. Select vision model (classification, detection, or segmentation)
+2. Load model for image processing task
+3. Pass image and process results
+4. Display detections or classifications in app UI
+
+**Result:** App that understands image content without sending data to servers
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md)
+
+---
+
+### Use Case 3: Document/Receipt Scanning
+
+**Trigger:** User wants to extract text from photos (receipts, documents, business cards)
+
+**Steps:**
+
+1. Choose OCR model matching target language
+2. Load appropriate recognizer for alphabet/language
+3. Capture or load image
+4. Extract text regions with bounding boxes
+5. Post-process results for application
+
+**Result:** OCR-enabled app that reads text directly from device camera
+
+**Reference:** [./references/reference-ocr.md](./references/reference-ocr.md)
+
+---
+
+### Use Case 4: Voice Interface
+
+**Trigger:** User wants to add voice commands, transcription, or voice output to app
+
+**Steps:**
+
+- **For voice input:** Capture audio at correct sample rate → transcribe with STT model
+- **For voice output:** Generate speech from text → play through audio context
+- Handle audio format/sample rate conversion
+
+**Result:** App with hands-free voice interaction
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+---
+
+### Use Case 5: Semantic Search
+
+**Trigger:** User needs intelligent search, similarity matching, or content recommendations
+
+**Steps:**
+
+1. Load text or image embeddings model
+2. Generate embeddings for searchable content
+3. Compute similarity scores between queries and content
+4. Rank and return results
+
+**Result:** Smart search that understands meaning, not just keywords
+
+**Reference:** [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Core Capabilities by Category
+
+### Large Language Models (LLMs)
+
+Run text generation, chat, function calling, and structured output generation locally on-device.
+
+**Supported features:**
+
+- Text generation and chat completions
+- Function/tool calling
+- Structured output with JSON schema validation
+- Streaming responses
+- Multiple model families (Llama 3.2, Qwen 3, Hammer 2.1, SmolLM2, Phi 4)
+
+**Reference:** See [./references/reference-llms.md](./references/reference-llms.md)
+
+---
+
+### Computer Vision
+
+Perform image understanding and manipulation tasks entirely on-device.
+
+**Supported tasks:**
+
+- **Image Classification** - Categorize images into predefined classes
+- **Object Detection** - Locate and identify objects with bounding boxes
+- **Image Segmentation** - Pixel-level classification
+- **Style Transfer** - Apply artistic styles to images
+- **Text-to-Image** - Generate images from text descriptions
+- **Image Embeddings** - Convert images to numerical vectors for similarity/search
+
+**Reference:** See [./references/reference-cv.md](./references/reference-cv.md) and [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+---
+
+### Optical Character Recognition (OCR)
+
+Extract and recognize text from images with support for multiple languages and text orientations.
+
+**Supported features:**
+
+- Text detection in images
+- Text recognition across different alphabets
+- Horizontal text (standard documents, receipts)
+- Vertical text support (experimental, for CJK languages)
+- Multi-language support with language-specific recognizers
+
+**Reference:** See [./references/reference-ocr.md](./references/reference-ocr.md)
+
+---
+
+### Audio Processing
+
+Convert between speech and text, and detect speech activity in audio.
+
+**Supported tasks:**
+
+- **Speech-to-Text** - Transcribe audio to text (supports multiple languages including English)
+- **Text-to-Speech** - Generate natural-sounding speech from text
+- **Voice Activity Detection** - Detect speech segments in audio
+
+**Reference:** See [./references/reference-audio.md](./references/reference-audio.md)
+
+---
+
+### Natural Language Processing
+
+Convert text to numerical representations for semantic understanding and search.
+
+**Supported tasks:**
+
+- **Text Embeddings** - Convert text to vectors for similarity/search
+- **Tokenization** - Convert text to tokens and vice versa
+
+**Reference:** See [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Getting Started by Use Case
+
+### I want to build a chatbot or AI assistant
+
+Use `useLLM` hook or `LLMModule` with one of the available language models.
+
+**What to do:**
+
+1. Choose a model from available LLM options (consider device memory constraints)
+2. Use the `useLLM` hook or `LLMModule` to load the model
+3. Send messages and receive responses
+4. Optionally configure system prompts, generation parameters, and tools
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - LLMs section
+
+---
+
+### I want to enable function/tool calling in my LLM
+
+Use `useLLM` hook or `LLMModule` with tool definitions to allow the model to call predefined functions.
+
+**What to do:**
+
+1. Define tools with name, description, and parameter schema
+2. Configure the LLM with tool definitions
+3. Implement callbacks to execute tools when the model requests them
+4. Parse tool results and pass them back to the model
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md) - Tool Calling section
+
+---
+
+### I want structured data extraction from text
+
+Use `useLLM` hook or `LLMModule` with structured output generation using JSON schema validation.
+
+**What to do:**
+
+1. Define a schema (JSON Schema or Zod) for desired output format
+2. Configure the LLM with the schema
+3. Generate responses and validate against the schema
+4. Use the validated structured data in your app
+
+**Reference:** [./references/reference-llms.md](./references/reference-llms.md) - Structured Output section
+
+---
+
+### I want to classify or recognize objects in images
+
+Use `useClassification` hook or `ClassificationModule` for simple categorization or use `useObjectDetection` hook or `ObjectDetectionModule` for locating specific objects.
+
+**What to do:**
+
+1. Choose appropriate computer vision model based on task
+2. Load the model with the appropriate hook or module
+3. Pass image (local, remote, or base64)
+4. Process results (classifications, detections with bounding boxes)
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Classification and Object Detection sections
+
+---
+
+### I want to extract text from images
+
+Use `useOCR` hook or `OCRModule` for horizontal text or use `useVerticalOCR` hook or `VerticalOCRModule` for vertical text (experimental).
+
+**What to do:**
+
+1. Choose appropriate OCR model and recognizer matching your target language
+2. Load the model with appropriate hook or module
+3. Pass image
+4. Extract detected text regions with bounding boxes and confidence scores
+5. Process results based on your application needs
+
+**Reference:** [./references/reference-ocr.md](./references/reference-ocr.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - OCR section
+
+---
+
+### I want to convert speech to text or text to speech
+
+Use `useSpeechToText` hook or `SpeechToTextModule` for transcription or use `useTextToSpeech` hook or `TextToSpeechModule` for voice synthesis.
+
+**What to do:**
+
+- **For Speech-to-Text:** Capture or load audio, ensure 16kHz sample rate, transcribe
+- **For Text-to-Speech:** Prepare text, specify voice parameters, generate audio waveform, play using audio context
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Speech to Text and Text to Speech sections
+
+---
+
+### I want to find similar images or texts
+
+Use `useImageEmbeddings` hook or `ImageEmbeddingsModule` for images or `useTextEmbeddings` hook or `TextEmbeddingsModule` for text.
+
+**What to do:**
+
+1. Load appropriate embeddings model
+2. Generate embeddings for your content
+3. Compute similarity metrics (cosine similarity, dot product)
+4. Use similarity scores for search, clustering, or deduplication
+
+**Reference:**
+
+- Text: [./references/reference-nlp.md](./references/reference-nlp.md)
+- Images: [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+---
+
+### I want to apply artistic filters to photos
+
+Use `useStyleTransfer` hook or `StyleTransferModule` to apply predefined artistic styles to images.
+
+**What to do:**
+
+1. Choose from available artistic styles (Candy, Mosaic, Udnie, Rain Princess)
+2. Load the style transfer model
+3. Pass image
+4. Retrieve and use the stylized image
+
+**Reference:** [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Style Transfer section
+
+---
+
+### I want to generate images from text
+
+Use `useTextToImage` hook or `TextToImageModule` to create images based on text descriptions.
+
+**What to do:**
+
+1. Load the text-to-image model
+2. Provide text description (prompt)
+3. Optionally specify image size and number of generation steps
+4. Receive generated image (may take 20-60 seconds depending on device)
+
+**Reference:** [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+**Model options:** [./references/reference-models.md](./references/reference-models.md) - Text to Image section
+
+---
+
+## Understanding Model Loading
+
+Before using any AI model, you need to load it. Models can be loaded from three sources:
+
+**1. Bundled with app (assets folder)**
+
+- Best for small models (< 512MB)
+- Available immediately without download
+- Increases app installation size
+
+**2. Remote URL (downloaded on first use)**
+
+- Best for large models (> 512MB)
+- Downloaded once and cached locally
+- Keeps app size small
+- Requires internet on first use
+
+**3. Local file system**
+
+- Maximum flexibility for user-managed models
+- Requires custom download/file management UI
+
+**Model selection strategy:**
+
+1. Small models (< 512MB) → Bundle with app or download from URL
+2. Large models (> 512MB) → Download from URL on first use with progress tracking
+3. Quantized models → Preferred for lower-end devices to save memory
+
+**Reference:** [./references/reference-models.md](./references/reference-models.md) - Loading Models section
+
+---
+
+## Device Constraints and Model Selection
+
+Not all models work on all devices. Consider these constraints:
+
+**Memory limitations:**
+
+- Low-end devices: Use smaller models (135M-1.7B parameters) and quantized variants
+- High-end devices: Can run larger models (3B-4B parameters)
+
+**Processing power:**
+
+- Lower-end devices: Expect longer inference times
+- Audio processing requires specific sample rates (16kHz for STT and VAD, 24kHz for TTS output)
+
+**Storage:**
+
+- Large models require significant disk space
+- Implement cleanup mechanisms to remove unused models
+- Monitor total downloaded model size
+
+**Guidance:**
+
+- Always check model memory requirements before recommending models
+- Prefer quantized model variants on lower-end devices
+- Show download progress for models > 512MB
+- Test on target devices before release
+
+**Reference:** [./references/reference-models.md](./references/reference-models.md)
+
+---
+
+## Important Technical Requirements
+
+### Audio Processing
+
+Audio must be in correct sample rate for processing:
+
+- **Speech-to-Text or VAD input:** 16kHz sample rate
+- **Text-to-Speech output:** 24kHz sample rate
+- Always decode/resample audio to correct rate before processing
+
+**Reference:** [./references/reference-audio.md](./references/reference-audio.md)
+
+### Image Processing
+
+Images can be provided as:
+
+- Remote URLs (http/https) - automatically cached
+- Local file URIs (file://)
+- Base64-encoded strings
+
+Image preprocessing (resizing, normalization) is handled automatically by most hooks.
+
+**Reference:** [./references/reference-cv.md](./references/reference-cv.md) and [./references/reference-cv-2.md](./references/reference-cv-2.md)
+
+### Text Tokens
+
+Text embeddings and LLMs have maximum token limits. Text exceeding these limits will be truncated. Use `useTokenizer` to count tokens before processing.
+
+**Reference:** [./references/reference-nlp.md](./references/reference-nlp.md)
+
+---
+
+## Core Utilities and Error Handling
+
+The library provides core utilities for managing models and handling errors:
+
+**ResourceFetcher:** Manage model downloads with pause/resume capabilities, storage cleanup, and progress tracking.
+
+**Error Handling:** Use `RnExecutorchError` and error codes for robust error handling and user feedback.
+
+**useExecutorchModule:** Low-level API for custom models not covered by dedicated hooks.
+
+**Reference:** [./references/core-utilities.md](./references/core-utilities.md)
+
+---
+
+## Common Troubleshooting
+
+**Model not loading:** Check model source URL/path validity and sufficient device storage
+
+**Out of memory errors:** Switch to smaller model or quantized variant
+
+**Poor LLM quality:** Adjust temperature/top-p parameters or improve system prompt
+
+**Audio issues:** Verify correct sample rate (16kHz for STT and VAD, 24kHz output for TTS)
+
+**Download failures:** Implement retry logic and check network connectivity
+
+**Reference:** [./references/core-utilities.md](./references/core-utilities.md) for error handling details, or specific reference file for your use case
+
+---
+
+## Quick Reference by Hook
+
+| Hook                   | Purpose                                   | Reference                                             |
+| ---------------------- | ----------------------------------------- | ----------------------------------------------------- |
+| `useLLM`               | Text generation, chat, function calling   | [reference-llms.md](./references/reference-llms.md)   |
+| `useClassification`    | Image categorization                      | [reference-cv.md](./references/reference-cv.md)       |
+| `useObjectDetection`   | Object localization                       | [reference-cv.md](./references/reference-cv.md)       |
+| `useImageSegmentation` | Pixel-level classification                | [reference-cv.md](./references/reference-cv.md)       |
+| `useStyleTransfer`     | Artistic image filters                    | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useTextToImage`       | Image generation                          | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useImageEmbeddings`   | Image similarity/search                   | [reference-cv-2.md](./references/reference-cv-2.md)   |
+| `useOCR`               | Text recognition (horizontal)             | [reference-ocr.md](./references/reference-ocr.md)     |
+| `useVerticalOCR`       | Text recognition (vertical, experimental) | [reference-ocr.md](./references/reference-ocr.md)     |
+| `useSpeechToText`      | Audio transcription                       | [reference-audio.md](./references/reference-audio.md) |
+| `useTextToSpeech`      | Voice synthesis                           | [reference-audio.md](./references/reference-audio.md) |
+| `useVAD`               | Voice activity detection                  | [reference-audio.md](./references/reference-audio.md) |
+| `useTextEmbeddings`    | Text similarity/search                    | [reference-nlp.md](./references/reference-nlp.md)     |
+| `useTokenizer`         | Text to tokens conversion                 | [reference-nlp.md](./references/reference-nlp.md)     |
+| `useExecutorchModule`  | Custom model inference (advanced)         | [core-utilities.md](./references/core-utilities.md)   |
+
+---
+
+## Quick Checklist for Implementation
+
+Use this when building AI features with ExecuTorch:
+
+**Planning Phase**
+
+- Identified what AI task you need (chat, vision, audio, search)
+- Considered device memory constraints and target devices
+- Chose appropriate model from available options
+- Determined if cloud backup fallback is needed
+
+**Development Phase**
+
+- Selected correct hook for your task
+- Configured model loading (bundled, remote URL, or local)
+- Implemented proper error handling
+- Added loading states for model operations
+- Tested audio sample rates (if audio task)
+- Set up resource management for large models
+
+**Testing Phase**
+
+- Tested on target minimum device
+- Verified offline functionality works
+- Checked memory usage doesn't exceed device limits
+- Tested error handling (network, memory, invalid inputs)
+- Measured inference time for acceptable UX
+
+**Deployment Phase**
+
+- Model bundling strategy decided (size/download tradeoff)
+- Download progress UI implemented (if remote models)
+- Version management plan for model updates
+- User feedback mechanism for quality issues
+
+---
+
+## Reference Files Overview
+
+**[reference-llms.md](./references/reference-llms.md)**
+
+- Complete LLM hook documentation
+- Functional vs Managed modes
+- Tool calling implementation
+- Structured output generation
+
+**[reference-cv.md](./references/reference-cv.md)**
+
+- Image classification, detection, and segmentation
+- Basic computer vision tasks
+
+**[reference-cv-2.md](./references/reference-cv-2.md)**
+
+- Advanced vision tasks: style transfer, text-to-image, embeddings
+- Image similarity and search
+
+**[reference-ocr.md](./references/reference-ocr.md)**
+
+- Horizontal and vertical text recognition
+- Multi-language support
+- OCR model selection
+
+**[reference-audio.md](./references/reference-audio.md)**
+
+- Speech-to-text transcription
+- Text-to-speech voice synthesis
+- Voice activity detection
+- Audio sample rate requirements
+
+**[reference-nlp.md](./references/reference-nlp.md)**
+
+- Text embeddings for semantic search
+- Tokenization utilities
+- Token limits and model compatibility
+
+**[reference-models.md](./references/reference-models.md)**
+
+- Complete list of available models
+- Model loading strategies
+- Model selection guidelines
+- Device memory/performance considerations
+
+**[core-utilities.md](./references/core-utilities.md)**
+
+- ResourceFetcher for download management
+- Error handling with RnExecutorchError
+- Low-level useExecutorchModule API
+- Error codes reference
+
+---
+
+## Troubleshooting Guide
+
+**Model not loading or crashing**
+
+- Check model source (URL valid, file exists)
+- Verify device has sufficient free storage and memory
+- Try bundling smaller models first
+- Check error codes with `RnExecutorchError`
+
+**Out of memory errors**
+
+- Switch to quantized model variant (smaller file size)
+- Use smaller parameter model (135M instead of 1.7B)
+- Close other apps to free device memory
+- Implement model unloading when not in use
+
+**Poor quality results from LLM**
+
+- Adjust generation parameters (temperature, top-p)
+- Improve system prompt
+- Try larger model if device supports it
+- Check input preprocessing
+
+**Audio not processing**
+
+- Verify sample rate is 16kHz for STT and VAD, 24kHz output for TTS
+- Check audio format compatibility
+- Ensure audio buffer has data before processing
+- Validate microphone permissions
+
+**Slow inference speed**
+
+- Expected on lower-end devices (especially larger models)
+- Show loading indicator to user
+- Consider preprocessing optimization
+- Profile on actual target device
+
+---
+
+## Best Practices
+
+**Model Selection**
+
+- Match model size to device capabilities
+- Use quantized variants for memory-constrained devices
+- Test on minimum target device before release
+- Keep models updated via download mechanism
+
+**Error Handling**
+
+- Always wrap AI operations in try-catch
+- Provide user-friendly error messages
+- Implement fallback behavior (cloud API, simplified UX)
+- Log errors for debugging
+
+**User Experience**
+
+- Show loading states during model operations
+- Display download progress for large models
+- Ensure app remains responsive during inference
+- Consider offline-first design
+
+**Resource Management**
+
+- Unload unused models to free memory
+- Implement cleanup for old cached models
+- Show storage impact of AI features
+- Monitor battery usage of continuous processing
+
+**Performance Optimization**
+
+- Batch requests when possible
+- Preload models during idle time
+- Profile actual device performance before launch
+- Use appropriate model size for each task
+
+---
+
+## External Resources
+
+- **Official Documentation:** https://docs.swmansion.com/react-native-executorch
+- **HuggingFace Models:** https://huggingface.co/software-mansion/collections
+- **GitHub Repository:** https://github.com/software-mansion/react-native-executorch
+- **API Reference:** https://docs.swmansion.com/react-native-executorch/docs/api-reference
+- **Software Mansion:** https://swmansion.com/
diff --git a/skills/.canary/react-native-executorch/references/core-utilities.md b/skills/.canary/react-native-executorch/references/core-utilities.md
new file mode 100644
index 000000000..9b5a0c18b
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/core-utilities.md
@@ -0,0 +1,327 @@
+---
+title: RN Executorch core Utilities
+description: Reference for using core RN Executorch utils - low-level ExecuTorch bindings, resource management, and error handling.
+---
+
+# useExecutorchModule
+
+**Purpose:** Low-level bindings to ExecuTorch Module API for custom model integration.
+
+**Use cases:** Custom models without dedicated hooks, advanced model control, experimental models, research applications.
+
+**Important:** Use dedicated hooks (useLLM, useClassification, etc.) when available. This hook is for custom models where no pre-built solution exists.
+
+## Basic Usage
+
+```typescript
+import { useExecutorchModule } from 'react-native-executorch';
+
+const executorchModule = useExecutorchModule({
+  modelSource: require('../assets/models/model.pte'),
+});
+```
+
+## Understanding TensorPtr
+
+A `TensorPtr` is the JavaScript representation of a tensor passed to the model:
+
+```typescript
+interface TensorPtr {
+  dataPtr: ArrayBuffer | TypedArray; // Raw data buffer
+  sizes: number[]; // Tensor shape [batch, channels, height, width]
+  scalarType: ScalarType; // Data type (FLOAT, INT, etc.)
+}
+```
+
+## Example usage
+
+```typescript
+import {
+  useExecutorchModule,
+  ScalarType,
+  STYLE_TRANSFER_CANDY,
+} from 'react-native-executorch';
+
+const executorchModule = useExecutorchModule({
+  modelSource: STYLE_TRANSFER_CANDY,
+});
+
+const runInference = async () => {
+  // Prepare input tensor (example: 640x640 RGB image)
+  const inputTensor = {
+    dataPtr: new Float32Array(1 * 3 * 640 * 640),
+    sizes: [1, 3, 640, 640],
+    scalarType: ScalarType.FLOAT,
+  };
+
+  try {
+    // Perform the forward operation and receive the stylized image output.
+    const output = await executorchModule.forward([inputTensor]);
+    // Interpret the output ArrayBuffer
+    // foo(output[0].dataPtr);
+  } catch (error) {
+    // Log any errors that occur during the forward pass.
+    console.error('Error during model execution:', error);
+  }
+};
+```
+
+## Troubleshooting
+
+**Preprocessing required:** You must handle all preprocessing (normalization, resizing, color space conversion) yourself.
+**Postprocessing required:** Output interpretation is your responsibility based on your model's architecture.
+**Shape matching:** Input tensor shapes must exactly match your model's expected input dimensions.
+**Use dedicated hooks:** If a hook exists for your use case, use it instead for automatic pre/post-processing.
+
+## Additional references
+
+- [useExecutorchModule docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/executorch-bindings/useExecutorchModule)
+- [useExecutorchModule API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useExecutorchModule)
+- [ExecuTorch Module API](https://pytorch.org/executorch/stable/extension-module.html)
+- [Typescript API implementation of useExecutorchModule](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/executorch-bindings/ExecutorchModule)
+
+---
+
+# ResourceFetcher
+
+**Purpose:** Manage model and resource downloads with pause/resume capabilities.
+
+**Use cases:** Download management, storage cleanup, progress tracking, offline-first apps.
+
+## Basic Usage
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+// Download multiple resources with progress tracking
+const downloadModels = async () => {
+  try {
+    const uris = await ResourceFetcher.fetch(
+      (progress) =>
+        console.log(`Download progress: ${(progress * 100).toFixed(1)}%`),
+      'https://example.com/llama3_2.pte',
+      'https://example.com/qwen3.pte'
+    );
+
+    if (uris) {
+      console.log('Downloaded files:', uris);
+    } else {
+      console.log('Download was paused or cancelled');
+    }
+  } catch (error) {
+    console.error('Download failed:', error);
+  }
+};
+```
+
+## Pause and Resume Downloads
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+const uris = ResourceFetcher.fetch(
+  (progress) => console.log('Total progress:', progress),
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+).then((uris) => {
+  console.log('URI resolved as: ', uris); // since we pause the fetch, uris is resolved to null
+});
+
+await ResourceFetcher.pauseFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+
+const resolvedUris = await ResourceFetcher.resumeFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+```
+
+## Cancel Downloads
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+const uris = ResourceFetcher.fetch(
+  (progress) => console.log('Total progress:', progress),
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+).then((uris) => {
+  console.log('URI resolved as: ', uris); // since we cancel the fetch, uris is resolved to null
+});
+
+await ResourceFetcher.cancelFetching(
+  'https://.../llama3_2.pte',
+  'https://.../qwen3.pte'
+);
+```
+
+## Manage Downloaded Resources
+
+```typescript
+import { ResourceFetcher } from 'react-native-executorch';
+
+// List all downloaded files
+const listFiles = async () => {
+  const files = await ResourceFetcher.listDownloadedFiles();
+  console.log('All downloaded files:', files);
+
+  const models = await ResourceFetcher.listDownloadedModels();
+  console.log('Model files:', models);
+};
+
+// Clean up old resources
+const cleanup = async () => {
+  const oldModelUrl = 'https://example.com/old_model.pte';
+
+  await ResourceFetcher.deleteResources(oldModelUrl);
+  console.log('Old model deleted');
+};
+```
+
+## Resource Types
+
+Resources can be:
+
+- Remote URLs (https://)
+- Local file paths (file://)
+- Asset references (require())
+- JSON objects
+
+## Troubleshooting
+
+**Resume vs re-fetch:** Use `resumeFetching()` for faster resume. Calling `fetch()` again works but is slower.
+**Progress callback:** Progress is reported as 0-1 for all downloads combined.
+**Null return:** If `fetch()` returns `null`, download was paused or cancelled.
+**Network errors:** Implement retry logic with exponential backoff for reliability.
+**Storage location:** Downloaded files are stored in application's document directory under `react-native-executorch/`
+
+## Additional references
+
+- [ResourceFetcher full reference docs](https://docs.swmansion.com/react-native-executorch/docs/utilities/resource-fetcher)
+- [Loading Models guide](https://docs.swmansion.com/react-native-executorch/docs/fundamentals/loading-models)
+
+---
+
+# Error Handling
+
+**Purpose:** Comprehensive error handling with typed error codes.
+
+**Use cases:** Debugging, production error recovery, user feedback, logging and monitoring.
+
+## Basic Error Handling
+
+```typescript
+import {
+  LLMModule,
+  LLAMA3_2_1B_QLORA,
+  RnExecutorchError,
+  RnExecutorchErrorCode,
+} from 'react-native-executorch';
+
+const llm = new LLMModule({
+  tokenCallback: (token) => console.log(token),
+  messageHistoryCallback: (messages) => console.log(messages),
+});
+
+try {
+  await llm.load(LLAMA3_2_1B_QLORA, (progress) => console.log(progress));
+  await llm.sendMessage('Hello!');
+} catch (err) {
+  if (err instanceof RnExecutorchError) {
+    console.error(`Error code: ${err.code}`);
+    console.error(`Error message: ${err.message}`);
+  } else {
+    throw err;
+  }
+}
+```
+
+## Handling Specific Error Types
+
+```typescript
+import {
+  RnExecutorchError,
+  RnExecutorchErrorCode,
+} from 'react-native-executorch';
+
+const handleModelError = async (llm, message: string) => {
+  try {
+    await llm.sendMessage(message);
+  } catch (err) {
+    if (err instanceof RnExecutorchError) {
+      switch (err.code) {
+        case RnExecutorchErrorCode.ModuleNotLoaded:
+          console.error('Model not loaded. Loading now...');
+          await llm.load(LLAMA3_2_1B_QLORA);
+          // Retry the message
+          await llm.sendMessage(message);
+          break;
+
+        case RnExecutorchErrorCode.ModelGenerating:
+          console.error('Model busy. Waiting...');
+          // Wait and retry, or queue the message
+          break;
+
+        case RnExecutorchErrorCode.InvalidConfig:
+          console.error('Invalid configuration:', err.message);
+          // Reset to default config
+          await llm.configure({ topp: 0.9, temperature: 0.7 });
+          break;
+
+        default:
+          console.error('Unexpected error:', err.message);
+          throw err;
+      }
+    }
+  }
+};
+```
+
+## Error Categories
+
+**Module State Errors**
+
+- `ModuleNotLoaded` - Model not loaded yet
+- `ModelGenerating` - Model already processing
+
+**Configuration Errors**
+
+- `InvalidConfig` - Invalid parameters
+- `InvalidUserInput` - Bad input data
+- `InvalidModelSource` - Wrong model source type
+- `WrongDimensions` - Incorrect tensor shape
+
+**File Operations**
+
+- `FileReadFailed` - Can't read file
+- `FileWriteFailed` - Can't write file
+
+**Download & Resources**
+
+- `DownloadInterrupted` - Download didn't complete
+- `ResourceFetcherDownloadFailed` - Network/server error
+- `ResourceFetcherDownloadInProgress` - Already downloading
+- `ResourceFetcherAlreadyPaused` - Already paused
+- `ResourceFetcherNotActive` - No active download
+
+**Runtime Errors**
+
+- `MemoryAllocationFailed` - Out of memory
+- `NotSupported` - Operation not supported
+- `InvalidProgram` - Invalid model file
+
+For complete error reference, see the [Error Handling documentation](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling).
+
+## Troubleshooting
+
+**Always check instance:** Use `instanceof RnExecutorchError` before accessing `.code`.
+**Log error codes:** Include error codes in logs for easier debugging.
+**Retry logic:** Implement exponential backoff for network and resource errors.
+**User feedback:** Translate error codes into user-friendly messages.
+
+## Additional references
+
+- [Error Handling docs](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling)
+- [Complete error code list](https://docs.swmansion.com/react-native-executorch/docs/utilities/error-handling#reference)
diff --git a/skills/.canary/react-native-executorch/references/reference-audio.md b/skills/.canary/react-native-executorch/references/reference-audio.md
new file mode 100644
index 000000000..e18eaf51d
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-audio.md
@@ -0,0 +1,433 @@
+---
+title: Audio related models usage
+description: Reference for using Speech to Text, Text to Speech and Voice Activity Detection models.
+---
+
+# useSpeechToText
+
+**Purpose:** Convert spoken audio to text (transcription).
+**Use cases:** Voice assistants, transcription apps, voice commands, accessibility features.
+
+## Basic Usage
+
+```typescript
+import { useSpeechToText, WHISPER_TINY_EN } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+import * as FileSystem from 'expo-file-system';
+
+const model = useSpeechToText({
+  model: WHISPER_TINY_EN,
+});
+
+const { uri } = await FileSystem.downloadAsync(
+  'https://some-audio-url.com/file.mp3',
+  FileSystem.cacheDirectory + 'audio_file'
+);
+
+const audioContext = new AudioContext({ sampleRate: 16000 });
+const decodedAudioData = await audioContext.decodeAudioData(uri);
+const audioBuffer = decodedAudioData.getChannelData(0);
+
+try {
+  const transcription = await model.transcribe(audioBuffer);
+  console.log(transcription.text);
+} catch (error) {
+  console.error('Error during audio transcription', error);
+}
+```
+
+## Multilingual Transcription
+
+```typescript
+import { WHISPER_TINY } from 'react-native-executorch';
+
+const model = useSpeechToText({ model: WHISPER_TINY });
+
+// Specify language
+const transcription = await model.transcribe(spanishAudio, {
+  language: 'es',
+});
+```
+
+## Timestamps & Transcription Stat Data
+
+You can obtain word-level timestamps and other useful parameters from transcription by setting `verbose: true` in the options:
+
+```typescript
+const transcription = await model.transcribe(audioBuffer, { verbose: true });
+// Example result
+// {
+//   task: "transcription",
+//   text: "Example text for a ...",
+//   duration: 9.05,
+//   language: "en",
+//   segments: [
+//     {
+//       start: 0,
+//       end: 5.4,
+//       text: "Example text for",
+//       words: [
+//         {
+//            word: "Example",
+//            start: 0,
+//            end: 1.4
+//         },
+//         ...
+//       ],
+//       tokens: [1, 32, 45, ...],
+//       temperature: 0.0,
+//       avgLogprob: -1.235,
+//       compressionRatio: 1.632
+//     },
+//     ...
+//   ]
+// }
+```
+
+## Streaming Transcription
+
+For audio longer than 30 seconds, use streaming transcription with the whisper-streaming algorithm. This intelligently chunks audio to avoid cutting speech mid-sentence:
+
+```typescript
+import React, { useEffect, useState, useRef } from 'react';
+import { Text, Button, View, SafeAreaView } from 'react-native';
+import { useSpeechToText, WHISPER_TINY_EN } from 'react-native-executorch';
+import { AudioManager, AudioRecorder } from 'react-native-audio-api';
+
+export default function App() {
+  const model = useSpeechToText({
+    model: WHISPER_TINY_EN,
+  });
+
+  const [transcribedText, setTranscribedText] = useState('');
+
+  const isRecordingRef = useRef(false);
+
+  const [recorder] = useState(() => new AudioRecorder());
+
+  useEffect(() => {
+    AudioManager.setAudioSessionOptions({
+      iosCategory: 'playAndRecord',
+      iosMode: 'spokenAudio',
+      iosOptions: ['allowBluetooth', 'defaultToSpeaker'],
+    });
+    AudioManager.requestRecordingPermissions();
+  }, []);
+
+  const handleStartStreamingTranscribe = async () => {
+    isRecordingRef.current = true;
+    setTranscribedText('');
+
+    const sampleRate = 16000;
+
+    recorder.onAudioReady(
+      {
+        sampleRate,
+        bufferLength: 0.1 * sampleRate,
+        channelCount: 1,
+      },
+      (chunk) => {
+        model.streamInsert(chunk.buffer.getChannelData(0));
+      }
+    );
+
+    try {
+      await recorder.start();
+    } catch (e) {
+      console.error('Recorder failed:', e);
+      return;
+    }
+
+    try {
+      let accumulatedCommitted = '';
+
+      const streamIter = model.stream({ verbose: false });
+
+      for await (const { committed, nonCommitted } of streamIter) {
+        if (!isRecordingRef.current) break;
+
+        if (committed.text) {
+          accumulatedCommitted += committed.text;
+        }
+
+        setTranscribedText(accumulatedCommitted + nonCommitted.text);
+      }
+    } catch (error) {
+      console.error('Error during streaming transcription:', error);
+    }
+  };
+
+  const handleStopStreamingTranscribe = () => {
+    isRecordingRef.current = false;
+    recorder.stop();
+    model.streamStop();
+  };
+
+    return (
+    <SafeAreaView>
+      <View style={{ padding: 20 }}>
+        <Text style={{ marginBottom: 20, fontSize: 18 }}>
+          {transcribedText || 'Press start to speak...'}
+        </Text>
+
+        <Button
+          onPress={handleStartStreamingTranscribe}
+          title="Start Streaming"
+          disabled={model.isGenerating}
+        />
+        <View style={{ height: 10 }} />
+        <Button
+          onPress={handleStopStreamingTranscribe}
+          title="Stop Streaming"
+          color="red"
+        />
+      </View>
+    </SafeAreaView>
+  );
+}
+```
+
+## Troubleshooting
+
+**Audio must be 16kHz:** Ensure proper sample rate before processing
+**Streaming algorithm:** Use whisper-streaming for longer audio (handles 30s chunks automatically)
+
+## Additional references
+
+- [useSpeechToText docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useSpeechToText)
+- [HuggingFace STT collection](https://huggingface.co/collections/software-mansion/speech-to-text)
+- [Available models](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---speech-to-text)
+- [useSpeechToText API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useSpeechToText)
+- [Typescript API implementation of useSpeechToText](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/SpeechToTextModule)
+
+---
+
+# useTextToSpeech
+
+**Purpose:** Convert text to natural-sounding speech (TTS).
+**Use cases:** Voice assistants, audiobooks, accessibility tools, voice navigation.
+
+## Basic Usage
+
+```typescript
+import {
+  useTextToSpeech,
+  KOKORO_MEDIUM,
+  KOKORO_VOICE_AF_HEART,
+} from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+
+const model = useTextToSpeech({
+  model: KOKORO_MEDIUM,
+  voice: KOKORO_VOICE_AF_HEART,
+});
+
+const audioContext = new AudioContext({ sampleRate: 24000 });
+
+const handleSpeech = async (text: string) => {
+  const speed = 1.0;
+  const waveform = await model.forward(text, speed);
+
+  const audioBuffer = audioContext.createBuffer(1, waveform.length, 24000);
+  audioBuffer.getChannelData(0).set(waveform);
+
+  const source = audioContext.createBufferSource();
+  source.buffer = audioBuffer;
+  source.connect(audioContext.destination);
+  source.start();
+};
+```
+
+## Streaming TTS
+
+```typescript
+// Stream chunks for lower latency
+await tts.stream({
+  text: 'Long text to be streamed chunk by chunk...',
+  speed: 1.0,
+  onNext: async (chunk) => {
+    return new Promise((resolve) => {
+      const buffer = ctx.createBuffer(1, chunk.length, 24000);
+      buffer.getChannelData(0).set(chunk);
+
+      const source = ctx.createBufferSource();
+      source.buffer = buffer;
+      source.connect(ctx.destination);
+      source.onEnded = () => resolve();
+      source.start();
+    });
+  },
+});
+```
+
+## Available Models & Voices
+
+**Model:** Kokoro (English only)
+
+For all available models check out [this exported HuggingFace models collection](https://huggingface.co/software-mansion/react-native-executorch-kokoro).
+
+**Available Voices:**
+
+- `KOKORO_VOICE_AF_HEART` - Female, heart
+- `KOKORO_VOICE_AF_SKY` - Female, sky
+- `KOKORO_VOICE_AF_BELLA` - Female, bella
+- `KOKORO_VOICE_AF_NICOLE` - Female, nicole
+- `KOKORO_VOICE_AF_SARAH` - Female, sarah
+- `KOKORO_VOICE_AM_ADAM` - Male, adam
+- `KOKORO_VOICE_AM_MICHAEL` - Male, michael
+- `KOKORO_VOICE_BF_EMMA` - British Female, emma
+- `KOKORO_VOICE_BF_ISABELLA` - British Female, isabella
+- `KOKORO_VOICE_BM_GEORGE` - British Male, george
+- `KOKORO_VOICE_BM_LEWIS` - British Male, lewis
+
+## Troubleshooting
+
+**Streaming vs Forward:** Use `stream()` for long texts to reduce time-to-first-audio
+
+## Additional references
+
+- [useTextToSpeech docs - reference and examples](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToSpeech)
+- [Supported Voices](https://docs.swmansion.com/react-native-executorch/docs/api-reference#tts-supported-voices)
+- [useTextToSpeech API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToSpeech)
+- [HuggingFace TTS collection](https://huggingface.co/collections/software-mansion/text-to-speech)
+- [Typescript API implementation of useTextToSpeech hook](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TextToSpeechModule)
+
+---
+
+# useVAD
+
+**Purpose:** Detect speech segments in audio (Voice Activity Detection).
+
+**Use cases:** Audio preprocessing, removing silence, speech segmentation, smart recording.
+
+## Basic Usage
+
+```typescript
+import { useVAD, FSMN_VAD } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+import * as FileSystem from 'expo-file-system';
+
+const model = useVAD({
+  model: FSMN_VAD,
+});
+
+const { uri } = await FileSystem.downloadAsync(
+  'https://some-audio-url.com/file.mp3',
+  FileSystem.cacheDirectory + 'audio_file'
+);
+
+const audioContext = new AudioContext({ sampleRate: 16000 });
+const decodedAudioData = await audioContext.decodeAudioDataSource(uri);
+const audioBuffer = decodedAudioData.getChannelData(0);
+
+try {
+  // NOTE: to obtain segments in seconds, you need to divide
+  // start / end of the segment by the sampling rate (16k)
+
+  const speechSegments = await model.forward(audioBuffer);
+  console.log(speechSegments);
+} catch (error) {
+  console.error('Error during running VAD model', error);
+}
+```
+
+## Example usage
+
+```tsx
+import React from 'react';
+import { Button, Text, SafeAreaView } from 'react-native';
+import { useVAD, FSMN_VAD } from 'react-native-executorch';
+import { AudioContext } from 'react-native-audio-api';
+import * as FileSystem from 'expo-file-system';
+
+export default function App() {
+  const model = useVAD({
+    model: FSMN_VAD,
+  });
+
+  const audioURL = 'https://some-audio-url.com/file.mp3';
+
+  const handleAudio = async () => {
+    if (!model) {
+      console.error('VAD model is not loaded yet.');
+      return;
+    }
+
+    console.log('Processing URL:', audioURL);
+
+    try {
+      const { uri } = await FileSystem.downloadAsync(
+        audioURL,
+        FileSystem.cacheDirectory + 'vad_example.tmp'
+      );
+
+      const audioContext = new AudioContext({ sampleRate: 16000 });
+      const originalDecodedBuffer =
+        await audioContext.decodeAudioDataSource(uri);
+      const originalChannelData = originalDecodedBuffer.getChannelData(0);
+
+      const segments = await model.forward(originalChannelData);
+      if (segments.length === 0) {
+        console.log('No speech segments were found.');
+        return;
+      }
+      console.log(`Found ${segments.length} speech segments.`);
+
+      const totalLength = segments.reduce(
+        (sum, seg) => sum + (seg.end - seg.start),
+        0
+      );
+      const newAudioBuffer = audioContext.createBuffer(
+        1, // Mono
+        totalLength,
+        originalDecodedBuffer.sampleRate
+      );
+      const newChannelData = newAudioBuffer.getChannelData(0);
+
+      let offset = 0;
+      for (const segment of segments) {
+        const slice = originalChannelData.subarray(segment.start, segment.end);
+        newChannelData.set(slice, offset);
+        offset += slice.length;
+      }
+
+      //  Play the processed audio
+      const source = audioContext.createBufferSource();
+      source.buffer = newAudioBuffer;
+      source.connect(audioContext.destination);
+      source.start();
+    } catch (error) {
+      console.error('Error processing audio data:', error);
+    }
+  };
+
+  return (
+    <SafeAreaView>
+      <Text>
+        Press the button to process and play speech from a sample file.
+      </Text>
+      <Button onPress={handleAudio} title="Run VAD Example" />
+    </SafeAreaView>
+  );
+}
+```
+
+## Available Models
+
+**Model:** [FSMN-VAD](https://huggingface.co/software-mansion/react-native-executorch-fsmn-vad)
+
+For all available models check out exported models in [this HuggingFace VAD models collection](https://huggingface.co/collections/software-mansion/voice-activity-detection).
+
+## Troubleshooting
+
+**Audio must be 16kHz:** Ensure proper sample rate  
+**Timestamps are indices:** Divide by sample rate (16000) to get seconds  
+**Returns array of segments:** `[{ start: number, end: number }]`
+
+## Additional references
+
+- [useVAD docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVAD)
+- [HuggingFace VAD collection](https://huggingface.co/collections/software-mansion/voice-activity-detection)
+- [useVAD API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVAD)
+- [available VAD model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---voice-activity-detection)
+- [Typescript API implementation of useVad hook](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/VADModule)
diff --git a/skills/.canary/react-native-executorch/references/reference-cv-2.md b/skills/.canary/react-native-executorch/references/reference-cv-2.md
new file mode 100644
index 000000000..1c5b07d16
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-cv-2.md
@@ -0,0 +1,190 @@
+---
+title: Computer Vision models usage (continued)
+description: Reference for using Style Transfer, Text to Image, Image Embeddings.
+---
+
+# useStyleTransfer
+
+**Purpose:** Apply artistic styles from one image to the content of another image.
+
+**Use cases:** Photo editing, artistic filters, creative content generation, style transformation apps.
+
+## Basic Usage
+
+```typescript
+import {
+  useStyleTransfer,
+  STYLE_TRANSFER_CANDY,
+} from 'react-native-executorch';
+
+const model = useStyleTransfer({ model: STYLE_TRANSFER_CANDY });
+
+const imageUri = 'file:///Users/.../photo.png';
+
+try {
+  const generatedImageUrl = await model.forward(imageUri);
+  console.log('Styled image:', generatedImageUrl);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Available Models
+
+**Model constants:**
+
+- `STYLE_TRANSFER_CANDY` - Candy artistic style
+- `STYLE_TRANSFER_MOSAIC` - Mosaic artistic style
+- `STYLE_TRANSFER_UDNIE` - Udnie artistic style
+- `STYLE_TRANSFER_RAIN_PRINCESS` - Rain princess artistic style
+
+For the latest available models reference exported models in [HuggingFace Style Transfer collection](https://huggingface.co/collections/software-mansion/style-transfer)
+
+## Troubleshooting
+
+**Image storage:** Both input images from external sources and generated images are stored in the application's temporary directory.
+**Processing time:** Style transfer can be computationally intensive. Expect processing to take several seconds depending on image size and device capabilities.
+
+## Additional references
+
+- [useStyleTransfer docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useStyleTransfer)
+- [HuggingFace Style Transfer collection](https://huggingface.co/collections/software-mansion/style-transfer)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---style-transfer)
+- [useStyleTransfer API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useStyleTransfer)
+
+---
+
+# useTextToImage
+
+**Purpose:** Generate images from text descriptions using on-device Stable Diffusion.
+
+**Use cases:** AI art generation, creative content creation, concept visualization, design prototyping.
+
+## Basic Usage
+
+```typescript
+import { useTextToImage, BK_SDM_TINY_VPRED_256 } from 'react-native-executorch';
+
+const model = useTextToImage({ model: BK_SDM_TINY_VPRED_256 });
+
+const input = 'a castle';
+
+try {
+  const image = await model.generate(input);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Example Usage with Parameters
+
+```tsx
+import { useTextToImage, BK_SDM_TINY_VPRED_256 } from 'react-native-executorch';
+
+function App() {
+  const model = useTextToImage({ model: BK_SDM_TINY_VPRED_256 });
+
+  //...
+  const input = 'a medieval castle by the sea shore';
+
+  const imageSize = 256;
+  const numSteps = 25;
+
+  try {
+    image = await model.generate(input, imageSize, numSteps);
+  } catch (error) {
+    console.error(error);
+  }
+  //...
+
+  return <Image source={{ uri: `data:image/png;base64,${image}` }} />;
+}
+```
+
+**Model constants:** `BK_SDM_TINY_VPRED_256`
+
+For the latest available models reference exported models in [HuggingFace Text to Image collection](https://huggingface.co/collections/software-mansion/text-to-image)
+
+## Troubleshooting
+
+**Memory requirements:** Larger image sizes require significantly more memory. Use 256x256 for lower-end devices, 512x512 for high-end devices.
+**Image size constraint:** Image size must be a multiple of 32 (e.g., 256, 320, 384, 512) due to U-Net and VAE decoder architecture.
+**Generation time:** Expect 20-60 seconds per image depending on device, image size, and number of steps.
+**Prompt engineering:** More descriptive prompts yield better results. Include style descriptors like "digital art", "photorealistic", "watercolor" etc.
+
+## Additional references
+
+- [useTextToImage docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useTextToImage)
+- [HuggingFace Text to Image collection](https://huggingface.co/collections/software-mansion/text-to-image)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---image-generation)
+- [useTextToImage API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextToImage)
+- [Typescript API implementation of Text to Image](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/TextToImageModule)
+
+---
+
+# useImageEmbeddings
+
+**Purpose:** Convert images into numerical vectors for similarity comparison and image search.
+
+**Use cases:** Image similarity search, duplicate detection, visual search, clustering, content-based retrieval.
+
+## Basic Usage
+
+```typescript
+import {
+  useImageEmbeddings,
+  CLIP_VIT_BASE_PATCH32_IMAGE,
+} from 'react-native-executorch';
+
+const model = useImageEmbeddings({ model: CLIP_VIT_BASE_PATCH32_IMAGE });
+
+try {
+  const imageEmbedding = await model.forward('https://url-to-image.jpg');
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Computing Image Similarity
+
+```typescript
+const dotProduct = (a: Float32Array, b: Float32Array) =>
+  a.reduce((sum, val, i) => sum + val * b[i], 0);
+
+const cosineSimilarity = (a: Float32Array, b: Float32Array) => {
+  const dot = dotProduct(a, b);
+  const normA = Math.sqrt(dotProduct(a, a));
+  const normB = Math.sqrt(dotProduct(b, b));
+  return dot / (normA * normB);
+};
+
+try {
+  // we assume you've provided catImage and dogImage
+  const catImageEmbedding = await model.forward(catImage);
+  const dogImageEmbedding = await model.forward(dogImage);
+
+  const similarity = cosineSimilarity(catImageEmbedding, dogImageEmbedding);
+
+  console.log(`Cosine similarity: ${similarity}`);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Available Models
+
+**Model constants:** `CLIP_VIT_BASE_PATCH32_IMAGE`
+
+For the latest available models reference exported models in [HuggingFace Image Embeddings collection](https://huggingface.co/collections/software-mansion/image-embeddings)
+
+## Troubleshooting
+
+**Normalized vectors:** Returned embeddings are already normalized, so cosine similarity can be computed as a simple dot product.
+**Image resizing:** Images are automatically resized to 224×224. Resizing large images can add processing time.
+
+## Additional references
+
+- [useImageEmbeddings docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useImageEmbeddings)
+- [HuggingFace Image Embeddings collection](https://huggingface.co/collections/software-mansion/image-embeddings)
+- [useImageEmbeddings API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useImageEmbeddings)
+- [Typescript API implementation of useImageEmbeddings](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ImageEmbeddingsModule)
diff --git a/skills/.canary/react-native-executorch/references/reference-cv.md b/skills/.canary/react-native-executorch/references/reference-cv.md
new file mode 100644
index 000000000..fb07b0ccf
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-cv.md
@@ -0,0 +1,215 @@
+---
+title: Computer Vision models usage
+description: Reference for using Image Classification, Image Segmentation, and Object Detection models.
+---
+
+# useClassification
+
+**Purpose:** Classify images into predefined categories.
+
+**Use cases:** Image recognition apps, content filtering, product categorization, accessibility features.
+
+## Basic Usage
+
+```typescript
+import { useClassification, EFFICIENTNET_V2_S } from 'react-native-executorch';
+
+const model = useClassification({ model: EFFICIENTNET_V2_S });
+
+const imageUri = 'file:///Users/.../cute_puppy.png';
+
+try {
+  const classesWithProbabilities = await model.forward(imageUri);
+  console.log(classesWithProbabilities);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Processing Results
+
+```typescript
+import { useClassification, EFFICIENTNET_V2_S } from 'react-native-executorch';
+
+function App() {
+  const model = useClassification({ model: EFFICIENTNET_V2_S });
+
+  // ...
+  const imageUri = 'file:///Users/.../cute_puppy.png';
+
+  try {
+    const classesWithProbabilities = await model.forward(imageUri);
+
+    // Extract three classes with the highest probabilities
+    const topThreeClasses = Object.entries(classesWithProbabilities)
+      .sort(([, a], [, b]) => b - a)
+      .slice(0, 3)
+      .map(([label, score]) => ({ label, score }));
+  } catch (error) {
+    console.error(error);
+  }
+  // ...
+}
+```
+
+## Available Models
+
+**Model constant:** `EFFICIENTNET_V2_S`
+
+For the latest available models reference exported models in [HuggingFace Classification collection](https://huggingface.co/collections/software-mansion/classification)
+
+## Troubleshooting
+
+**Confidence interpretation:** The class with the highest probability is typically assigned, but multiple high probabilities may indicate model uncertainty.
+**Image formats:** Accepts remote URLs, local file URIs, and base64-encoded images. Remote images are stored in the app's temporary directory.
+
+## Additional references
+
+- [useClassification docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useClassification)
+- [HuggingFace Classification collection](https://huggingface.co/collections/software-mansion/classification)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---classification)
+- [useClassification API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useClassification)
+- [Typescript API implementation of segmentation](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ClassificationModule)
+
+---
+
+# useImageSegmentation
+
+**Purpose:** Classify each pixel in an image to detect objects on a per-pixel basis.
+
+**Use cases:** Object detection, image editing, scene understanding.
+
+## Basic Usage
+
+```typescript
+import {
+  useImageSegmentation,
+  DEEPLAB_V3_RESNET50,
+  DeeplabLabel,
+} from 'react-native-executorch';
+
+const model = useImageSegmentation({ model: DEEPLAB_V3_RESNET50 });
+
+const imageUri = 'file:///Users/.../cute_cat.png';
+
+try {
+  const outputDict = await model.forward(imageUri);
+  console.log(outputDict[DeeplabLabel.ARGMAX]);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Advanced Usage with Classes of Interest
+
+```typescript
+const model = useImageSegmentation({ model: DEEPLAB_V3_RESNET50 });
+
+const handleSegmentation = async (imageUri: string) => {
+  try {
+    const outputDict = await model.forward(
+      imageUri,
+      [DeeplabLabel.CAT, DeeplabLabel.DOG, DeeplabLabel.PERSON],
+      true
+    );
+
+    const argmaxArray = outputDict[DeeplabLabel.ARGMAX];
+
+    const catProbabilities = outputDict[DeeplabLabel.CAT];
+    const dogProbabilities = outputDict[DeeplabLabel.DOG];
+    const personProbabilities = outputDict[DeeplabLabel.PERSON];
+    // ...
+  } catch (error) {
+    console.error(error);
+  }
+};
+```
+
+## Available Models
+
+**Model constant:** `DEEPLAB_V3_RESNET50`
+
+For the latest available models check out exported models in [this HuggingFace Segmentation collection](https://huggingface.co/collections/software-mansion/image-segmentation)
+
+## Troubleshooting
+
+**Performance:** Setting `resize=true` significantly increases processing time. Use `resize=false` for better performance when you don't need original image dimensions.
+**Memory usage:** Resize increases memory usage, especially with high-resolution images.
+**Pixel mapping:** When `resize=false`, pixel indices map to a 224x224 grid. When `resize=true`, indices map to original image dimensions.
+
+## Additional references
+
+- [useImageSegmentation docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useImageSegmentation)
+- [useImageSegmentation API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useImageSegmentation)
+- [HuggingFace Segmentation collection](https://huggingface.co/collections/software-mansion/image-segmentation)
+- [Typescript API implementation of segmentation](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ImageSegmentationModule)
+
+---
+
+# useObjectDetection
+
+**Purpose:** Identify and locate objects within images by providing bounding boxes, labels, and confidence scores.
+
+**Use cases:** Object detection.
+
+## Basic Usage
+
+```typescript
+import {
+  useObjectDetection,
+  SSDLITE_320_MOBILENET_V3_LARGE,
+} from 'react-native-executorch';
+
+const model = useObjectDetection({ model: SSDLITE_320_MOBILENET_V3_LARGE });
+
+try {
+  const detections = await model.forward('https://url-to-image.jpg');
+
+  for (const detection of detections) {
+    console.log('Bounding box:', detection.bbox);
+    console.log('Label:', detection.label);
+    console.log('Confidence:', detection.score);
+  }
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Detection Object Structure
+
+Each detection returned by `forward` has the following structure:
+
+```typescript
+interface Bbox {
+  x1: number; // Bottom-left x coordinate
+  y1: number; // Bottom-left y coordinate
+  x2: number; // Top-right x coordinate
+  y2: number; // Top-right y coordinate
+}
+
+interface Detection {
+  bbox: Bbox;
+  label: keyof typeof CocoLabels; // Object class name
+  score: number; // Confidence score (0-1)
+}
+```
+
+## Available Models
+
+**Model constant:** `SSDLITE_320_MOBILENET_V3_LARGE`
+
+For the latest available models reference exported models in [HuggingFace Object Detection collection](https://huggingface.co/collections/software-mansion/object-detection)
+
+## Troubleshooting
+
+**Multiple detections:** The model may detect the same object multiple times with slightly different bounding boxes. Consider implementing non-maximum suppression (NMS) if needed.
+**Confidence thresholds:** Adjust the confidence threshold based on your use case. Higher thresholds (>0.7) reduce false positives but may miss objects.
+**Coordinate system:** Bounding box coordinates are in pixel space relative to the input image dimensions.
+
+## Additional references
+
+- [useObjectDetection docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useObjectDetection)
+- [HuggingFace Object Detection collection](https://huggingface.co/collections/software-mansion/object-detection)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---object-detection)
+- [useObjectDetection API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useObjectDetection)
+- [Typescript API implementation of Object Detection](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/ObjectDetectionModule)
diff --git a/skills/.canary/react-native-executorch/references/reference-llms.md b/skills/.canary/react-native-executorch/references/reference-llms.md
new file mode 100644
index 000000000..36e41c686
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-llms.md
@@ -0,0 +1,259 @@
+---
+title: LLMs usage
+description: Reference for using Large Language Models in React Native Executorch.
+---
+
+# useLLM
+
+**Purpose:** Run Large Language Models (LLMs) on-device for text generation, chat, tool calling, and structured outputs.
+
+**Use cases:** AI assistants, text generation, function calling, structured data extraction.
+
+## Basic Usage
+
+```typescript
+import { useLLM, LLAMA3_2_1B } from 'react-native-executorch';
+
+const llm = useLLM({ model: LLAMA3_2_1B });
+```
+
+## Functional Mode (Stateless)
+
+```tsx
+const llm = useLLM({ model: LLAMA3_2_1B });
+
+const handleGenerate = async () => {
+  const chat: Message[] = [
+    { role: 'system', content: 'You are a helpful assistant' },
+    { role: 'user', content: 'Hi!' },
+    { role: 'assistant', content: 'Hi!, how can I help you?' },
+    { role: 'user', content: 'What is the meaning of life?' },
+  ];
+
+  // Chat completion - returns the generated response
+  const response = await llm.generate(chat);
+  console.log('Complete response:', response);
+};
+
+return (
+  <View>
+    <Button onPress={handleGenerate} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Managed Mode (Stateful)
+
+```tsx
+// Configure the model
+useEffect(() => {
+  llm.configure({
+    chatConfig: {
+      systemPrompt: 'You are a helpful assistant',
+      contextWindowLength: 10,
+    },
+    generationConfig: {
+      temperature: 0.7,
+      topp: 0.9,
+    },
+  });
+}, []);
+
+// Send messages
+llm.sendMessage('Hello!');
+
+// Access conversation history
+console.log(llm.messageHistory);
+```
+
+## Tool Calling
+
+```tsx
+const TOOL_DEFINITIONS: LLMTool[] = [
+  {
+    name: 'get_weather',
+    description: 'Get/check weather in given location.',
+    parameters: {
+      type: 'dict',
+      properties: {
+        location: {
+          type: 'string',
+          description: 'Location where user wants to check weather',
+        },
+      },
+      required: ['location'],
+    },
+  },
+];
+
+const llm = useLLM({ model: HAMMER2_1_1_5B });
+
+const handleGenerate = () => {
+  const chat: Message[] = [
+    {
+      role: 'system',
+      content: `You are a helpful assistant. Current time and date: ${new Date().toString()}`,
+    },
+    {
+      role: 'user',
+      content: `Hi, what's the weather like in Cracow right now?`,
+    },
+  ];
+
+  // Chat completion
+  llm.generate(chat, TOOL_DEFINITIONS);
+};
+
+export const executeTool: (call: ToolCall) => Promise<string | null> = async (
+  call
+) => {
+  switch (call.toolName) {
+    case 'get_weather':
+      return await get_weather(call);
+    default:
+      console.error(`Wrong function! We don't handle it!`);
+      return null;
+  }
+};
+
+const { configure } = llm;
+useEffect(() => {
+  configure({
+    chatConfig: {
+      systemPrompt: `${DEFAULT_SYSTEM_PROMPT} Current time and date: ${new Date().toString()}`,
+    },
+    toolsConfig: {
+      tools: TOOL_DEFINITIONS,
+      executeToolCallback: executeTool,
+      displayToolCalls: true,
+    },
+  });
+}, [configure]);
+
+return (
+  <View>
+    <Button onPress={handleGenerate} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Structured Output
+
+```tsx
+import { Schema } from 'jsonschema';
+
+const responseSchema: Schema = {
+  properties: {
+    username: {
+      type: 'string',
+      description: 'Name of user, that is asking a question.',
+    },
+    question: {
+      type: 'string',
+      description: 'Question that user asks.',
+    },
+    bid: {
+      type: 'number',
+      description: 'Amount of money, that user offers.',
+    },
+    currency: {
+      type: 'string',
+      description: 'Currency of offer.',
+    },
+  },
+  required: ['username', 'bid'],
+  type: 'object',
+};
+
+// alternatively use Zod
+import * as z from 'zod/v4';
+const responseSchemaWithZod = z.object({
+  username: z
+    .string()
+    .meta({ description: 'Name of user, that is asking a question.' }),
+  question: z.optional(
+    z.string().meta({ description: 'Question that user asks.' })
+  ),
+  bid: z.number().meta({ description: 'Amount of money, that user offers.' }),
+  currency: z.optional(z.string().meta({ description: 'Currency of offer.' })),
+});
+
+const llm = useLLM({ model: QWEN3_4B_QUANTIZED });
+
+useEffect(() => {
+  const formattingInstructions = getStructuredOutputPrompt(responseSchema);
+  // alternatively pass schema defined with Zod
+  //  const formattingInstructions = getStructuredOutputPrompt(responseSchemaWithZod);
+
+  // Some extra prompting to improve quality of response.
+  const prompt = `Your goal is to parse user's messages and return them in JSON format. Don't respond to user. Simply return JSON with user's question parsed. \n${formattingInstructions}\n /no_think`;
+
+  llm.configure({
+    chatConfig: {
+      systemPrompt: prompt,
+    },
+  });
+}, []);
+
+useEffect(() => {
+  const lastMessage = llm.messageHistory.at(-1);
+  if (!llm.isGenerating && lastMessage?.role === 'assistant') {
+    try {
+      const formattedOutput = fixAndValidateStructuredOutput(
+        lastMessage.content,
+        responseSchemaWithZod
+      );
+      // Zod will allow you to correctly type output
+      const formattedOutputWithZod = fixAndValidateStructuredOutput(
+        lastMessage.content,
+        responseSchema
+      );
+      console.log('Formatted output:', formattedOutput, formattedOutputWithZod);
+    } catch (e) {
+      console.log(
+        "Error parsing output and/or output doesn't match required schema!",
+        e
+      );
+    }
+  }
+}, [llm.messageHistory, llm.isGenerating]);
+
+const send = () => {
+  const message = `I'm John. Is this product damaged? I can give you $100 for this.`;
+  llm.sendMessage(message);
+};
+
+return (
+  <View>
+    <Button onPress={send} title="Generate!" />
+    <Text>{llm.response}</Text>
+  </View>
+);
+```
+
+## Interrupting Generation
+
+```tsx
+// Stop generating
+llm.interrupt();
+
+// Check if generating
+{
+  llm.isGenerating && <Button onPress={llm.interrupt} title="Stop" />;
+}
+```
+
+## Troubleshooting
+
+**Memory issues:** Use quantized models on lower-end devices - suggest smaller models as lower-end devices might not be able to fit LLMs into memory.
+**Crash on unmount:** Always call `llm.interrupt()` and wait for `isGenerating === false` before unmounting.
+**Reasoning mode:** Model-specific feature (e.g., Qwen 3 uses `/no_think` suffix to disable) - search for a way to disable the reasoning.
+
+## Additional references
+
+- [useLLM docs](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useLLM)
+- [HuggingFace repository with exported models](https://huggingface.co/collections/software-mansion/llm)
+- [Available LLM model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---lmm)
+- [API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useLLM)
diff --git a/skills/.canary/react-native-executorch/references/reference-models.md b/skills/.canary/react-native-executorch/references/reference-models.md
new file mode 100644
index 000000000..c8ff0fcab
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-models.md
@@ -0,0 +1,265 @@
+---
+title: Available models and loading models.
+description: RN Executorch provides models for a variety of AI tasks - Image classification, Object detection, Style transfer, OCR, Image embeddings, Image segmentation, Text to image, LLMs, Text embeddings, Speech to text, Text to speech, and Voice activity detection. This reference lists all supported models and provides a quick reference on available ways of loading a model.
+---
+
+# Available models RN Executorch
+
+It is generally recommended to use models provided by the authors of the library, which are available at [HuggingFace repository](https://huggingface.co/software-mansion/collections).
+
+---
+
+## LLMs (Large Language Models)
+
+For a full list of available LLM models reference [HuggingFace LLMs collection](https://huggingface.co/collections/software-mansion/llm).
+
+### LLAMA 3.2
+
+- **LLAMA3_2_3B** - Llama 3.2 3B original (bf16)
+- **LLAMA3_2_3B_QLORA** - Llama 3.2 3B with QLoRA quantization
+- **LLAMA3_2_3B_SPINQUANT** - Llama 3.2 3B with SpinQuant
+- **LLAMA3_2_1B** - Llama 3.2 1B original (bf16)
+- **LLAMA3_2_1B_QLORA** - Llama 3.2 1B with QLoRA quantization
+- **LLAMA3_2_1B_SPINQUANT** - Llama 3.2 1B with SpinQuant
+
+[HuggingFace Llama 3.2 Model](https://huggingface.co/software-mansion/react-native-executorch-llama-3.2)
+
+### QWEN 3
+
+- **QWEN3_0_6B** - Qwen 3 0.6B original (bf16)
+- **QWEN3_0_6B_QUANTIZED** - Qwen 3 0.6B quantized (8da4w)
+- **QWEN3_1_7B** - Qwen 3 1.7B original (bf16)
+- **QWEN3_1_7B_QUANTIZED** - Qwen 3 1.7B quantized (8da4w)
+- **QWEN3_4B** - Qwen 3 4B original (bf16)
+- **QWEN3_4B_QUANTIZED** - Qwen 3 4B quantized (8da4w)
+
+[HuggingFace Qwen 3 Model](https://huggingface.co/software-mansion/react-native-executorch-qwen-3)
+
+### HAMMER 2.1
+
+- **HAMMER2_1_0_5B** - Hammer 2.1 0.5B original (bf16)
+- **HAMMER2_1_0_5B_QUANTIZED** - Hammer 2.1 0.5B quantized (8da4w)
+- **HAMMER2_1_1_5B** - Hammer 2.1 1.5B original (bf16)
+- **HAMMER2_1_1_5B_QUANTIZED** - Hammer 2.1 1.5B quantized (8da4w)
+- **HAMMER2_1_3B** - Hammer 2.1 3B original (bf16)
+- **HAMMER2_1_3B_QUANTIZED** - Hammer 2.1 3B quantized (8da4w)
+
+[HuggingFace Hammer 2.1 model](https://huggingface.co/software-mansion/react-native-executorch-hammer-2.1)
+
+### SMOLLM2
+
+- **SMOLLM2_1_135M** - SmolLM2 135M original (bf16)
+- **SMOLLM2_1_135M_QUANTIZED** - SmolLM2 135M quantized (8da4w)
+- **SMOLLM2_1_360M** - SmolLM2 360M original (bf16)
+- **SMOLLM2_1_360M_QUANTIZED** - SmolLM2 360M quantized (8da4w)
+- **SMOLLM2_1_1_7B** - SmolLM2 1.7B original (bf16)
+- **SMOLLM2_1_1_7B_QUANTIZED** - SmolLM2 1.7B quantized (8da4w)
+
+[HuggingFace SmoLlm 2 model](https://huggingface.co/software-mansion/react-native-executorch-smolLm-2)
+
+### QWEN 2.5
+
+- **QWEN2_5_0_5B** - Qwen 2.5 0.5B original (bf16)
+- **QWEN2_5_0_5B_QUANTIZED** - Qwen 2.5 0.5B quantized (8da4w)
+- **QWEN2_5_1_5B** - Qwen 2.5 1.5B original (bf16)
+- **QWEN2_5_1_5B_QUANTIZED** - Qwen 2.5 1.5B quantized (8da4w)
+- **QWEN2_5_3B** - Qwen 2.5 3B original (bf16)
+- **QWEN2_5_3B_QUANTIZED** - Qwen 2.5 3B quantized (8da4w)
+
+[HuggingFace Qwen 2.5 Model](https://huggingface.co/software-mansion/react-native-executorch-qwen-2.5)
+
+### PHI 4
+
+- **PHI_4_MINI_4B** - Phi 4 Mini 4B original (bf16)
+- **PHI_4_MINI_4B_QUANTIZED** - Phi 4 Mini 4B quantized (8da4w)
+
+[HuggingFace PHI 4 Mini Model](https://huggingface.co/software-mansion/react-native-executorch-phi-4-mini)
+
+---
+
+## Image classification
+
+- **EFFICIENTNET_V2_S** - [EfficientNet V2 S](https://huggingface.co/software-mansion/react-native-executorch-efficientnet-v2-s) (CoreML for iOS, XNNPACK for Android)
+
+For a list of all available Image Classification models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/classification).
+
+---
+
+## Object detection
+
+- **SSDLITE_320_MOBILENET_V3_LARGE** - [SSDLite 320 with MobileNet V3 Large](https://huggingface.co/software-mansion/react-native-executorch-ssdlite320-mobilenet-v3-large)
+
+For a list of all available Object Detection models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/object-detection).
+
+---
+
+## Style transfer
+
+- **STYLE_TRANSFER_CANDY** - [Candy style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-candy)
+- **STYLE_TRANSFER_MOSAIC** - [Mosaic style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-mosaic)
+- **STYLE_TRANSFER_RAIN_PRINCESS** - [Rain Princess style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-rain-princess)
+- **STYLE_TRANSFER_UDNIE** - [Udnie style](https://huggingface.co/software-mansion/react-native-executorch-style-transfer-udnie)
+
+For a list of all available Style Transfer models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/style-transfer).
+
+---
+
+## OCR
+
+- **DETECTOR_CRAFT** - [CRAFT text detector](https://huggingface.co/software-mansion/react-native-executorch-detector-craft) - detects text regions in images
+- **RECOGNIZER_CRNN_EN** - [CRNN text recognizer](https://huggingface.co/software-mansion/react-native-executorch-recognizer-crnn.en) - recognizes English text
+
+For a list of all available OCR models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/ocr).
+
+---
+
+## Image embeddings
+
+- **CLIP_VIT_BASE_PATCH32_IMAGE** - [CLIP ViT Base Patch32](https://huggingface.co/software-mansion/react-native-executorch-clip-vit-base-patch32) for image embeddings
+
+For a list of all available Image embeddings models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/image-embeddings).
+
+---
+
+## Image segmentation
+
+- **DEEPLAB_V3_RESNET50** - [DeepLab V3](https://huggingface.co/software-mansion/react-native-executorch-deeplab-v3) with ResNet50 backbone
+
+For a list of all available Image Segmentation models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/image-segmentation).
+
+---
+
+## Text to image
+
+- **BK_SDM_TINY_VPRED_256** - BK-SDM Tiny V-Pred (256x256 resolution)
+
+For a list of all available Text to Image models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-to-image).
+
+---
+
+## Text embeddings
+
+- **ALL_MINILM_L6_V2** - [All-MiniLM-L6-v2](https://huggingface.co/software-mansion/react-native-executorch-all-MiniLM-L6-v2) text embeddings
+- **ALL_MPNET_BASE_V2** - [All-MPNet-Base-v2](https://huggingface.co/software-mansion/react-native-executorch-all-mpnet-base-v2) text embeddings
+- **MULTI_QA_MINILM_L6_COS_V1** - [Multi-QA MiniLM-L6](https://huggingface.co/software-mansion/react-native-executorch-multi-qa-MiniLM-L6-cos-v1) cosine similarity
+- **MULTI_QA_MPNET_BASE_DOT_V1** - Multi-QA MPNet Base dot product
+- **CLIP_VIT_BASE_PATCH32_TEXT** - [CLIP ViT Base Patch32](https://huggingface.co/software-mansion/react-native-executorch-clip-vit-base-patch32) for text embeddings
+
+For a list of all available Text embeddings models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-embeddings).
+
+---
+
+## Speech to text
+
+For a list of all available Speech to Text models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/speech-to-text).
+
+### Whisper Models (English only)
+
+- **WHISPER_TINY_EN** - [Whisper Tiny](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny.en)
+- **WHISPER_TINY_EN_QUANTIZED** - [Whisper Tiny English-only (quantized)](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny-quantized.en)
+- **WHISPER_BASE_EN** - [Whisper Base](https://huggingface.co/software-mansion/react-native-executorch-whisper-base.en)
+- **WHISPER_SMALL_EN** - [Whisper Small](https://huggingface.co/software-mansion/react-native-executorch-whisper-small.en)
+- **WHISPER_MEDIUM_EN** - [Whisper Medium](https://huggingface.co/software-mansion/react-native-executorch-whisper-medium.en)
+
+### Whisper Models (Multilingual)
+
+- **WHISPER_TINY** - [Whisper Tiny multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-tiny)
+- **WHISPER_BASE** - [Whisper Base multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-base)
+- **WHISPER_SMALL** - [Whisper Small multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-small)
+- **WHISPER_MEDIUM** - [Whisper Medium multilingual](https://huggingface.co/software-mansion/react-native-executorch-whisper-medium)
+
+### Other models
+
+- **MOONSHINE_TINY** - [Moonshine Tiny](https://huggingface.co/software-mansion/react-native-executorch-moonshine-tiny)
+
+---
+
+## Text to speech
+
+- **KOKORO_SMALL** - [Kokoro TTS](https://huggingface.co/software-mansion/react-native-executorch-kokoro) - Text-to-Speech model
+- **KOKORO_MEDIUM** - [Kokoro TTS](https://huggingface.co/software-mansion/react-native-executorch-kokoro) - Text-to-Speech model
+
+For a list of all available Text to Speech models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/text-to-speech).
+
+---
+
+## Voice activity detection
+
+- **FSMN_VAD** - [FSMN Voice Activity Detection](https://huggingface.co/software-mansion/react-native-executorch-fsmn-vad)
+
+For a list of all available VAD models reference [this Hugging Face collection](https://huggingface.co/collections/software-mansion/voice-activity-detection).
+
+---
+
+# Loading models
+
+## Using predefined constants
+
+**When to use:** This is the recommended approach for all models that have [exported constants in the library](https://github.com/software-mansion/react-native-executorch/blob/main/packages/react-native-executorch/src/constants/modelUrls.ts). It provides type safety, automatic URL management, and ensures you're using the correct model configuration with all required files.
+
+**Benefits:** No need to manually specify URLs, automatic updates when model versions change, and guaranteed compatibility with the library version you're using.
+
+```typescript
+import {
+  useLLM,
+  LLAMA3_2_1B,
+  QWEN3_1_7B_QUANTIZED,
+} from 'react-native-executorch';
+
+const llama = useLLM(LLAMA3_2_1B);
+const qwen = useLLM(QWEN3_1_7B_QUANTIZED);
+```
+
+---
+
+## From React Native assets folder
+
+**When to use:** Best for small models (< 512MB) that you want to bundle with your app for offline use from the first launch. This ensures the model is immediately available without any download, but increases your app's installation size.
+
+**Trade-offs:** Larger app bundle size, but instant availability and guaranteed offline functionality. Good for demo apps or when your target users have limited internet connectivity.
+
+```typescript
+useExecutorchModule({
+  modelSource: require('../assets/llama3_2.pte'),
+});
+```
+
+---
+
+## From remote URL
+
+**When to use:** Ideal for large models (> 512MB) or when you want to keep your app's download size small. The model downloads on first use and is cached locally for subsequent uses. Perfect for production apps where initial app size matters.
+
+**Trade-offs:** Requires internet connection on first use, longer initial loading time, but keeps your app bundle small and allows for easy model updates without app resubmission.
+
+```typescript
+useExecutorchModule({
+  modelSource: 'https://.../llama3_2.pte',
+});
+```
+
+---
+
+## From local file system
+
+**When to use:** When you want to give users full control over model management, allow custom model imports, or enable advanced users to use their own fine-tuned models. Also useful for development and testing with local model files.
+
+**Trade-offs:** Requires implementing your own download/file management UI, but provides maximum flexibility and user control.
+
+```typescript
+useExecutorchModule({
+  modelSource: 'file:///var/mobile/.../llama3_2.pte',
+});
+```
+
+---
+
+# References
+
+- **Documentation:** https://docs.swmansion.com/react-native-executorch/docs/fundamentals/loading-models
+- **HuggingFace Repository:** https://huggingface.co/software-mansion
+- **Model Constants:** https://github.com/software-mansion/react-native-executorch/blob/main/packages/react-native-executorch/src/constants/modelUrls.ts
+- **All Collections:** https://huggingface.co/software-mansion/collections
+- **Model sizes reference:** https://docs.swmansion.com/react-native-executorch/docs/benchmarks/model-size
+- **Model memory usage reference:**: https://docs.swmansion.com/react-native-executorch/docs/benchmarks/memory-usage
+- **Model inference time benchmarks:** https://docs.swmansion.com/react-native-executorch/docs/benchmarks/inference-time
diff --git a/skills/.canary/react-native-executorch/references/reference-nlp.md b/skills/.canary/react-native-executorch/references/reference-nlp.md
new file mode 100644
index 000000000..2052686bc
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-nlp.md
@@ -0,0 +1,169 @@
+---
+title: Text Embeddings and Tokenizer
+description: Reference for using Text Embeddings and Tokenizer in React Native Executorch.
+---
+
+# useTextEmbeddings
+
+**Purpose:** Convert text into numerical vectors for semantic search, similarity, and clustering.
+
+**Use cases:** Semantic search, document similarity, text classification, clustering, RAG systems.
+
+## Basic Usage
+
+```typescript
+import { useTextEmbeddings, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+const model = useTextEmbeddings({ model: ALL_MINILM_L6_V2 });
+
+try {
+  const embedding = await model.forward('Hello World!');
+  console.log(embedding);
+} catch (error) {
+  console.error(error);
+}
+```
+
+## Example (computing similarity)
+
+```typescript
+const dotProduct = (a: number[], b: number[]) =>
+  a.reduce((sum, val, i) => sum + val * b[i], 0);
+
+const cosineSimilarity = (a: number[], b: number[]) => {
+  const dot = dotProduct(a, b);
+  const normA = Math.sqrt(dotProduct(a, a));
+  const normB = Math.sqrt(dotProduct(b, b));
+  return dot / (normA * normB);
+};
+
+function App() {
+  const model = useTextEmbeddings({ model: ALL_MINILM_L6_V2 });
+
+  // ...
+
+  try {
+    const helloWorldEmbedding = await model.forward('Hello World!');
+    const goodMorningEmbedding = await model.forward('Good Morning!');
+
+    const similarity = cosineSimilarity(
+      helloWorldEmbedding,
+      goodMorningEmbedding
+    );
+
+    console.log(`Cosine similarity: ${similarity}`);
+  } catch (error) {
+    console.error(error);
+  }
+
+  // ...
+}
+```
+
+## Available Models
+
+| Model                      | Max Tokens | Dimensions | Use Case                 |
+| -------------------------- | ---------- | ---------- | ------------------------ |
+| all-MiniLM-L6-v2           | 254        | 384        | General purpose          |
+| all-mpnet-base-v2          | 382        | 768        | Higher quality, slower   |
+| multi-qa-MiniLM-L6-cos-v1  | 509        | 384        | Q&A / semantic search    |
+| multi-qa-mpnet-base-dot-v1 | 510        | 768        | Q&A / semantic search    |
+| clip-vit-base-patch32-text | 74         | 512        | Match with images (CLIP) |
+
+**Model constants:** `ALL_MINILM_L6_V2`, `ALL_MPNET_BASE_V2`, `MULTI_QA_MINILM_L6_COS_V1`, `MULTI_QA_MPNET_BASE_DOT_V1`, `CLIP_VIT_BASE_PATCH32_TEXT`
+
+**Max Tokens** - The maximum number of tokens that can be processed by the model. If the input text exceeds this limit, it will be truncated.
+
+**Embedding Dimensions** - The size of the output embedding vector. This is the number of dimensions in the vector representation of the input text.
+
+For the latest available models reference exported models in [HuggingFace Text Embeddings collection](https://huggingface.co/collections/software-mansion/text-embeddings)
+
+## Troubleshooting
+
+**Normalized vectors:** For the supported models, the returned embedding vector is normalized, meaning that its length is equal to 1. This allows for easier comparison of vectors using cosine similarity, just calculate the dot product of two vectors to get the cosine similarity score.
+**Token limits:** Text exceeding max tokens will be truncated.
+
+## Additional references
+
+- [useTextEmbeddings docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/natural-language-processing/useTextEmbeddings)
+- [HuggingFace Text Embeddings collection](https://huggingface.co/collections/software-mansion/text-embeddings)
+- [Available model constants](https://docs.swmansion.com/react-native-executorch/docs/api-reference#models---text-embeddings)
+- [useTextEmbeddings API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTextEmbeddings)
+- [Typescript API implementation of useTextEmbeddings](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TextEmbeddingsModule)
+
+---
+
+# useTokenizer
+
+**Purpose:** Convert text to tokens and vice versa (used internally by LLMs and embeddings).
+**Use cases:** Token counting, understanding model limits, debugging, custom preprocessing.
+
+## Basic Usage
+
+```typescript
+import { useTokenizer, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+const tokenizer = useTokenizer({ tokenizer: ALL_MINILM_L6_V2 });
+
+const text = 'Hello, world!';
+
+try {
+  // Tokenize the text
+  const tokens = await tokenizer.encode(text);
+  console.log('Tokens:', tokens);
+
+  // Decode the tokens back to text
+  const decodedText = await tokenizer.decode(tokens);
+  console.log('Decoded text:', decodedText);
+} catch (error) {
+  console.error('Error tokenizing text:', error);
+}
+```
+
+## Example usage
+
+```typescript
+import { useTokenizer, ALL_MINILM_L6_V2 } from 'react-native-executorch';
+
+function App() {
+  const tokenizer = useTokenizer({ tokenizer: ALL_MINILM_L6_V2 });
+
+  // ...
+
+  try {
+    const text = 'Hello, world!';
+
+    const vocabSize = await tokenizer.getVocabSize();
+    console.log('Vocabulary size:', vocabSize);
+
+    const tokens = await tokenizer.encode(text);
+    console.log('Token IDs:', tokens);
+
+    const decoded = await tokenizer.decode(tokens);
+    console.log('Decoded text:', decoded);
+
+    const tokenId = await tokenizer.tokenToId('hello');
+    console.log('Token ID for "Hello":', tokenId);
+
+    const token = await tokenizer.idToToken(tokenId);
+    console.log('Token for ID:', token);
+  } catch (error) {
+    console.error(error);
+  }
+
+  // ...
+}
+```
+
+## Troubleshooting
+
+**Uses HuggingFace Tokenizers:** Full compatibility with HF ecosystem - we are using [Hugging Face Tokenizers](https://huggingface.co/docs/tokenizers/index) under the hood, ensuring compatibility with the Hugging Face ecosystem.
+
+**Mostly internal:** You usually don't need this directly - LLM/embedding hooks handle it.
+
+## Additional references
+
+- [useTokenizer docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/natural-language-processing/useTokenizer)
+- [useTokenizer API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useTokenizer)
+- [HuggingFace Tokenizers documentation](https://huggingface.co/docs/tokenizers/index)
+- [Typescript API implementation of useTokenizer](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/natural-language-processing/TokenizerModule)
diff --git a/skills/.canary/react-native-executorch/references/reference-ocr.md b/skills/.canary/react-native-executorch/references/reference-ocr.md
new file mode 100644
index 000000000..818478e57
--- /dev/null
+++ b/skills/.canary/react-native-executorch/references/reference-ocr.md
@@ -0,0 +1,171 @@
+---
+title: OCR usage
+description: Reference for using OCR and Vertical OCR.
+---
+
+# useOCR
+
+**Purpose:** Detect and recognize horizontal text within images (Optical Character Recognition).
+
+**Use cases:** Document scanning, receipt processing, business card reading, text extraction from photos.
+
+## Basic Usage
+
+```typescript
+import { useOCR, OCR_ENGLISH } from 'react-native-executorch';
+
+function App() {
+  const model = useOCR({ model: OCR_ENGLISH });
+
+  // ...
+  for (const ocrDetection of await model.forward('https://url-to-image.jpg')) {
+    console.log('Bounding box: ', ocrDetection.bbox);
+    console.log('Bounding label: ', ocrDetection.text);
+    console.log('Bounding score: ', ocrDetection.score);
+  }
+  // ...
+}
+```
+
+## Understanding the Detection Object
+
+```typescript
+interface Point {
+  x: number;
+  y: number;
+}
+
+interface OCRDetection {
+  bbox: Point[]; // 4 corner points of bounding box
+  text: string; // Recognized text
+  score: number; // Confidence score (0-1)
+}
+```
+
+The `bbox` array contains four points representing the corners of the detected text region, allowing for rotated or skewed text detection.
+
+## Language Support
+
+Different recognizer models support different alphabets and languages:
+
+```typescript
+import {
+  useOCR,
+  OCR_ENGLISH,
+  RECOGNIZER_LATIN_CRNN,
+  RECOGNIZER_CYRILLIC_CRNN,
+  DETECTOR_CRAFT,
+} from 'react-native-executorch';
+
+// For English (uses Latin alphabet)
+const englishOCR = useOCR({ model: OCR_ENGLISH });
+
+// For custom language configuration
+const customOCR = useOCR({
+  model: {
+    detectorSource: DETECTOR_CRAFT,
+    recognizerSource: RECOGNIZER_CYRILLIC_CRNN,
+    language: 'ru', // Russian
+  },
+});
+```
+
+**Important:** The recognizer model must match the alphabet of your target language. For example, use `RECOGNIZER_LATIN_CRNN` for English, Polish, German, etc., and `RECOGNIZER_CYRILLIC_CRNN` for Russian, Ukrainian, etc.
+
+## Available Models
+
+For all supported alphabets and languages, see [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets) and exported models in [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr).
+
+## Troubleshooting
+
+**Text orientation:** This hook is designed for horizontal text. For vertical text (e.g., traditional Chinese/Japanese), use `useVerticalOCR`.
+**Language/alphabet matching:** Ensure the recognizer model matches your target language's alphabet.
+**Image quality:** Higher resolution and better contrast improve recognition accuracy.
+
+## Additional references
+
+- [useOCR docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useOCR)
+- [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr)
+- [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets)
+- [useOCR API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useOCR)
+- [Typescript API implementation of useOCR](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/OCRModule)
+
+---
+
+# useVerticalOCR
+
+**Purpose:** Detect and recognize vertical text within images (experimental).
+
+**Use cases:** Traditional Chinese/Japanese text recognition, vertical signage, manga/comic text extraction.
+
+## Basic Usage
+
+```typescript
+import { useVerticalOCR, OCR_ENGLISH } from 'react-native-executorch';
+
+function App() {
+  const model = useVerticalOCR({
+    model: OCR_ENGLISH,
+    independentCharacters: true,
+  });
+
+  // ...
+  for (const ocrDetection of await model.forward('https://url-to-image.jpg')) {
+    console.log('Bounding box: ', ocrDetection.bbox);
+    console.log('Bounding label: ', ocrDetection.text);
+    console.log('Bounding score: ', ocrDetection.score);
+  }
+  // ...
+}
+```
+
+## Character vs Word Mode
+
+The `independentCharacters` parameter controls how text is processed:
+
+```typescript
+// Character mode - each character detected separately (better for CJK)
+const charMode = useVerticalOCR({
+  model: OCR_CHINESE,
+  independentCharacters: true,
+});
+
+// Word mode - characters grouped into words (better for Latin alphabets)
+const wordMode = useVerticalOCR({
+  model: OCR_ENGLISH,
+  independentCharacters: false,
+});
+```
+
+## Understanding the Detection Object
+
+```typescript
+interface Point {
+  x: number;
+  y: number;
+}
+
+interface OCRDetection {
+  bbox: Point[]; // 4 corner points of bounding box
+  text: string; // Recognized text
+  score: number; // Confidence score (0-1)
+}
+```
+
+## Available Models
+
+For all supported alphabets and languages, see [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets) and exported models in [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr).
+
+## Troubleshooting
+
+**Experimental status:** This hook is experimental and may have limitations with certain text layouts or languages.
+**Character vs word mode:** Use `independentCharacters: true` for CJK languages, `false` for Latin alphabets.
+**Alphabet matching:** Ensure the recognizer matches your target language's alphabet.
+
+## Additional references
+
+- [useVerticalOCR docs](https://docs.swmansion.com/react-native-executorch/docs/hooks/computer-vision/useVerticalOCR)
+- [HuggingFace OCR collection](https://huggingface.co/collections/software-mansion/ocr)
+- [OCR Supported Alphabets](https://docs.swmansion.com/react-native-executorch/docs/api-reference#ocr-supported-alphabets)
+- [useVerticalOCR API reference](https://docs.swmansion.com/react-native-executorch/docs/api-reference/functions/useVerticalOCR)
+- [Typescript API implementation of useVerticalOCR](https://docs.swmansion.com/react-native-executorch/docs/typescript-api/computer-vision/VerticalOCRModule)

From 6b532e47fba9c94d5beee0e422f95326d37e8c80 Mon Sep 17 00:00:00 2001
From: mateuszlampert <mateuszlampert@gmail.com>
Date: Mon, 16 Feb 2026 12:22:49 +0100
Subject: [PATCH 6/7] chore: move canary version

---
 skills/{.canary => canary}/react-native-executorch/SKILL.md | 0
 .../react-native-executorch/references/core-utilities.md    | 0
 .../react-native-executorch/references/reference-audio.md   | 6 +++---
 .../react-native-executorch/references/reference-cv-2.md    | 0
 .../react-native-executorch/references/reference-cv.md      | 0
 .../react-native-executorch/references/reference-llms.md    | 0
 .../react-native-executorch/references/reference-models.md  | 0
 .../react-native-executorch/references/reference-nlp.md     | 0
 .../react-native-executorch/references/reference-ocr.md     | 0
 9 files changed, 3 insertions(+), 3 deletions(-)
 rename skills/{.canary => canary}/react-native-executorch/SKILL.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/core-utilities.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-audio.md (99%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-cv-2.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-cv.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-llms.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-models.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-nlp.md (100%)
 rename skills/{.canary => canary}/react-native-executorch/references/reference-ocr.md (100%)

diff --git a/skills/.canary/react-native-executorch/SKILL.md b/skills/canary/react-native-executorch/SKILL.md
similarity index 100%
rename from skills/.canary/react-native-executorch/SKILL.md
rename to skills/canary/react-native-executorch/SKILL.md
diff --git a/skills/.canary/react-native-executorch/references/core-utilities.md b/skills/canary/react-native-executorch/references/core-utilities.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/core-utilities.md
rename to skills/canary/react-native-executorch/references/core-utilities.md
diff --git a/skills/.canary/react-native-executorch/references/reference-audio.md b/skills/canary/react-native-executorch/references/reference-audio.md
similarity index 99%
rename from skills/.canary/react-native-executorch/references/reference-audio.md
rename to skills/canary/react-native-executorch/references/reference-audio.md
index e18eaf51d..fccdd9e85 100644
--- a/skills/.canary/react-native-executorch/references/reference-audio.md
+++ b/skills/canary/react-native-executorch/references/reference-audio.md
@@ -76,7 +76,7 @@ const transcription = await model.transcribe(audioBuffer, { verbose: true });
 //       ],
 //       tokens: [1, 32, 45, ...],
 //       temperature: 0.0,
-//       avgLogprob: -1.235,
+//       avgLogProb: -1.235,
 //       compressionRatio: 1.632
 //     },
 //     ...
@@ -88,7 +88,7 @@ const transcription = await model.transcribe(audioBuffer, { verbose: true });
 
 For audio longer than 30 seconds, use streaming transcription with the whisper-streaming algorithm. This intelligently chunks audio to avoid cutting speech mid-sentence:
 
-```typescript
+```tsx
 import React, { useEffect, useState, useRef } from 'react';
 import { Text, Button, View, SafeAreaView } from 'react-native';
 import { useSpeechToText, WHISPER_TINY_EN } from 'react-native-executorch';
@@ -163,7 +163,7 @@ export default function App() {
     model.streamStop();
   };
 
-    return (
+  return (
     <SafeAreaView>
       <View style={{ padding: 20 }}>
         <Text style={{ marginBottom: 20, fontSize: 18 }}>
diff --git a/skills/.canary/react-native-executorch/references/reference-cv-2.md b/skills/canary/react-native-executorch/references/reference-cv-2.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-cv-2.md
rename to skills/canary/react-native-executorch/references/reference-cv-2.md
diff --git a/skills/.canary/react-native-executorch/references/reference-cv.md b/skills/canary/react-native-executorch/references/reference-cv.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-cv.md
rename to skills/canary/react-native-executorch/references/reference-cv.md
diff --git a/skills/.canary/react-native-executorch/references/reference-llms.md b/skills/canary/react-native-executorch/references/reference-llms.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-llms.md
rename to skills/canary/react-native-executorch/references/reference-llms.md
diff --git a/skills/.canary/react-native-executorch/references/reference-models.md b/skills/canary/react-native-executorch/references/reference-models.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-models.md
rename to skills/canary/react-native-executorch/references/reference-models.md
diff --git a/skills/.canary/react-native-executorch/references/reference-nlp.md b/skills/canary/react-native-executorch/references/reference-nlp.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-nlp.md
rename to skills/canary/react-native-executorch/references/reference-nlp.md
diff --git a/skills/.canary/react-native-executorch/references/reference-ocr.md b/skills/canary/react-native-executorch/references/reference-ocr.md
similarity index 100%
rename from skills/.canary/react-native-executorch/references/reference-ocr.md
rename to skills/canary/react-native-executorch/references/reference-ocr.md

From a9f0a72f9324ad8841671d2d2ab8dfed34095e8f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Mateusz=20S=C5=82uszniak?= <mateusz.sluszniak@swmansion.com>
Date: Mon, 16 Feb 2026 15:01:53 +0100
Subject: [PATCH 7/7] docs: Update docs about image segmentation

---
 .../03-hooks/02-computer-vision/useImageSegmentation.md     | 6 +++---
 .../02-computer-vision/ImageSegmentationModule.md           | 4 ++--
 .../docs/06-api-reference/functions/useImageSegmentation.md | 2 +-
 3 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/docs/docs/03-hooks/02-computer-vision/useImageSegmentation.md b/docs/docs/03-hooks/02-computer-vision/useImageSegmentation.md
index edcb6de05..c3e7346b9 100644
--- a/docs/docs/03-hooks/02-computer-vision/useImageSegmentation.md
+++ b/docs/docs/03-hooks/02-computer-vision/useImageSegmentation.md
@@ -55,13 +55,13 @@ To run the model, you can use the [`forward`](../../06-api-reference/interfaces/
 
 - The image can be a remote URL, a local file URI, or a base64-encoded image.
 - The [`classesOfInterest`](../../06-api-reference/interfaces/ImageSegmentationType.md#classesofinterest) list contains classes for which to output the full results. By default the list is empty, and only the most probable classes are returned (essentially an arg max for each pixel). Look at [`DeeplabLabel`](../../06-api-reference/enumerations/DeeplabLabel.md) enum for possible classes.
-- The [`resizeToInput`](../../06-api-reference/interfaces/ImageSegmentationType.md#resizeToInput) flag specifies whether the output will be rescaled back to the size of the input image. The default is `true`. The model runs inference on a scaled (probably smaller) version of your image (224x224 for `DEEPLAB_V3_RESNET50`). If you choose to resize, the output will be `number[]` of size `width * height` of your original image.
+- The [`resizeToInput`](../../06-api-reference/interfaces/ImageSegmentationType.md#resizetoinput) flag specifies whether the output will be rescaled back to the size of the input image. The default is `true`. The model runs inference on a scaled (probably smaller) version of your image (224x224 for `DEEPLAB_V3_RESNET50`). If you choose to resize, the output will be `number[]` of size `width * height` of your original image.
 
 :::warning
-Setting `resize` to true will make `forward` slower.
+Setting `resizeToInput` to `false` will make `forward` faster.
 :::
 
-[`forward`](../../06-api-reference/interfaces/ImageSegmentationType.md#forward) returns a promise which can resolve either to an error or a dictionary containing number arrays with size depending on [`resize`](../../06-api-reference/interfaces/ImageSegmentationType.md#resize):
+[`forward`](../../06-api-reference/interfaces/ImageSegmentationType.md#forward) returns a promise which can resolve either to an error or a dictionary containing number arrays with size depending on [`resizeToInput`](../../06-api-reference/interfaces/ImageSegmentationType.md#resizetoinput):
 
 - For the key [`DeeplabLabel.ARGMAX`](../../06-api-reference/enumerations/DeeplabLabel.md#argmax) the array contains for each pixel an integer corresponding to the class with the highest probability.
 - For every other key from [`DeeplabLabel`](../../06-api-reference/enumerations/DeeplabLabel.md), if the label was included in [`classesOfInterest`](../../06-api-reference/interfaces/ImageSegmentationType.md#classesofinterest) the dictionary will contain an array of floats corresponding to the probability of this class for every pixel.
diff --git a/docs/docs/04-typescript-api/02-computer-vision/ImageSegmentationModule.md b/docs/docs/04-typescript-api/02-computer-vision/ImageSegmentationModule.md
index 373da8b9d..2d272b673 100644
--- a/docs/docs/04-typescript-api/02-computer-vision/ImageSegmentationModule.md
+++ b/docs/docs/04-typescript-api/02-computer-vision/ImageSegmentationModule.md
@@ -52,13 +52,13 @@ To run the model, you can use the [`forward`](../../06-api-reference/classes/Ima
 
 - The image can be a remote URL, a local file URI, or a base64-encoded image.
 - The [`classesOfInterest`](../../06-api-reference/classes/ImageSegmentationModule.md#classesofinterest) list contains classes for which to output the full results. By default the list is empty, and only the most probable classes are returned (essentially an arg max for each pixel). Look at [`DeeplabLabel`](../../06-api-reference/enumerations/DeeplabLabel.md) enum for possible classes.
-- The [`resizeToInput`](../../06-api-reference/classes/ImageSegmentationModule.md#resizeToInput) flag specifies whether the output will be rescaled back to the size of the input image. The default is `true`. The model runs inference on a scaled (probably smaller) version of your image (224x224 for the `DEEPLAB_V3_RESNET50`). If you choose to resize, the output will be `number[]` of size `width * height` of your original image.
+- The [`resizeToInput`](../../06-api-reference/classes/ImageSegmentationModule.md#resizetoinput) flag specifies whether the output will be rescaled back to the size of the input image. The default is `true`. The model runs inference on a scaled (probably smaller) version of your image (224x224 for the `DEEPLAB_V3_RESNET50`). If you choose to resize, the output will be `number[]` of size `width * height` of your original image.
 
 :::warning
 Setting `resize` to true will make `forward` slower.
 :::
 
-[`forward`](../../06-api-reference/classes/ImageSegmentationModule.md#forward) returns a promise which can resolve either to an error or a dictionary containing number arrays with size depending on [`resize`](../../06-api-reference/classes/ImageSegmentationModule.md#resize):
+[`forward`](../../06-api-reference/classes/ImageSegmentationModule.md#forward) returns a promise which can resolve either to an error or a dictionary containing number arrays with size depending on [`resizeToInput`](../../06-api-reference/classes/ImageSegmentationModule.md#resizetoinput):
 
 - For the key [`DeeplabLabel.ARGMAX`](../../06-api-reference/enumerations/DeeplabLabel.md#argmax) the array contains for each pixel an integer corresponding to the class with the highest probability.
 - For every other key from [`DeeplabLabel`](../../06-api-reference/enumerations/DeeplabLabel.md), if the label was included in [`classesOfInterest`](../../06-api-reference/classes/ImageSegmentationModule.md#classesofinterest) the dictionary will contain an array of floats corresponding to the probability of this class for every pixel.
diff --git a/docs/docs/06-api-reference/functions/useImageSegmentation.md b/docs/docs/06-api-reference/functions/useImageSegmentation.md
index 6753775cb..5c2d81361 100644
--- a/docs/docs/06-api-reference/functions/useImageSegmentation.md
+++ b/docs/docs/06-api-reference/functions/useImageSegmentation.md
@@ -2,7 +2,7 @@
 
 > **useImageSegmentation**(`ImageSegmentationProps`): [`ImageSegmentationType`](../interfaces/ImageSegmentationType.md)
 
-Defined in: [packages/react-native-executorch/src/hooks/computer_vision/useImageSegmentation.ts:15](https://github.com/software-mansion/react-native-executorch/blob/9e79b9bf2a34159a71071fbfdaed3ddd9393702f/packages/react-native-executorch/src/hooks/computer_vision/useImageSegmentation.ts#L15)
+Defined in: [packages/react-native-executorch/src/hooks/computer_vision/useImageSegmentation.ts:15](https://github.com/software-mansion/react-native-executorch/blob/6b532e47fba9c94d5beee0e422f95326d37e8c80/packages/react-native-executorch/src/hooks/computer_vision/useImageSegmentation.ts#L15)
 
 React hook for managing an Image Segmentation model instance.