expo-ai-kit 0.3.6 → 0.4.0

package/src/index.ts

@@ -7,15 +7,6 @@ import {
   LLMStreamOptions,
   LLMStreamEvent,
   LLMStreamCallback,
-  LLMSummarizeOptions,
-  LLMTranslateOptions,
-  LLMRewriteOptions,
-  LLMExtractKeyPointsOptions,
-  LLMAnswerQuestionOptions,
-  LLMSuggestOptions,
-  LLMSuggestResponse,
-  LLMSmartReplyOptions,
-  LLMAutocompleteOptions,
   BuiltInModel,
   DownloadableModel,
   ModelError,
@@ -24,8 +15,6 @@ import {
 import { MODEL_REGISTRY, getRegistryEntry } from './models';
 
 export * from './types';
-export * from './memory';
-export * from './hooks';
 export * from './models';
 
 const DEFAULT_SYSTEM_PROMPT =
@@ -37,140 +26,9 @@ function generateSessionId(): string {
 }
 
 // ============================================================================
-// Prompt Helper Constants
+// Inference API
 // ============================================================================
 
-const SUMMARIZE_LENGTH_INSTRUCTIONS = {
-  short: 'Keep it very brief, around 1-2 sentences.',
-  medium: 'Provide a moderate summary, around 3-5 sentences.',
-  long: 'Provide a comprehensive summary covering all main points.',
-} as const;
-
-const SUMMARIZE_STYLE_INSTRUCTIONS = {
-  paragraph: 'Write the summary as a flowing paragraph.',
-  bullets: 'Format the summary as bullet points.',
-  tldr: 'Start with "TL;DR:" and give an extremely concise summary in 1 sentence.',
-} as const;
-
-const TRANSLATE_TONE_INSTRUCTIONS = {
-  formal: 'Use formal language and honorifics where appropriate.',
-  informal: 'Use casual, everyday language.',
-  neutral: 'Use standard, neutral language.',
-} as const;
-
-const REWRITE_STYLE_INSTRUCTIONS = {
-  formal:
-    'Rewrite in a formal, professional tone suitable for business communication.',
-  casual: 'Rewrite in a casual, conversational tone.',
-  professional:
-    'Rewrite in a clear, professional tone suitable for work contexts.',
-  friendly: 'Rewrite in a warm, friendly tone.',
-  concise:
-    'Rewrite to be as brief as possible while keeping the meaning intact.',
-  detailed: 'Expand and add more detail and explanation.',
-  simple:
-    'Rewrite using simple words and short sentences, easy for anyone to understand.',
-  academic: 'Rewrite in an academic style suitable for scholarly writing.',
-} as const;
-
-const SUGGEST_TONE_INSTRUCTIONS = {
-  formal: 'Use formal, professional language.',
-  casual: 'Use casual, everyday language.',
-  professional: 'Use clear, professional language suitable for work.',
-  friendly: 'Use warm, friendly language.',
-  neutral: 'Use standard, neutral language.',
-} as const;
-
-const ANSWER_DETAIL_INSTRUCTIONS = {
-  brief: 'Give a brief, direct answer in 1-2 sentences.',
-  medium: 'Provide a clear answer with some explanation.',
-  detailed:
-    'Provide a comprehensive answer with full explanation and relevant details from the context.',
-} as const;
-
-// ============================================================================
-// Prompt Builder Helpers
-// ============================================================================
-
-function buildSummarizePrompt(
-  length: 'short' | 'medium' | 'long',
-  style: 'paragraph' | 'bullets' | 'tldr'
-): string {
-  return `You are a summarization assistant. Summarize the provided text accurately and concisely. ${SUMMARIZE_LENGTH_INSTRUCTIONS[length]} ${SUMMARIZE_STYLE_INSTRUCTIONS[style]} Only output the summary, nothing else.`;
-}
-
-function buildTranslatePrompt(
-  to: string,
-  from: string | undefined,
-  tone: 'formal' | 'informal' | 'neutral'
-): string {
-  const fromClause = from ? `from ${from} ` : '';
-  return `You are a translation assistant. Translate the provided text ${fromClause}to ${to}. ${TRANSLATE_TONE_INSTRUCTIONS[tone]} Only output the translation, nothing else. Do not include any explanations or notes.`;
-}
-
-function buildRewritePrompt(
-  style:
-    | 'formal'
-    | 'casual'
-    | 'professional'
-    | 'friendly'
-    | 'concise'
-    | 'detailed'
-    | 'simple'
-    | 'academic'
-): string {
-  return `You are a writing assistant. ${REWRITE_STYLE_INSTRUCTIONS[style]} Preserve the original meaning. Only output the rewritten text, nothing else.`;
-}
-
-function buildExtractKeyPointsPrompt(maxPoints: number): string {
-  return `You are an analysis assistant. Extract the ${maxPoints} most important key points from the provided text. Format each point as a bullet point starting with "•". Be concise and focus on the most significant information. Only output the bullet points, nothing else.`;
-}
-
-function buildSuggestPrompt(
-  count: number,
-  tone: 'formal' | 'casual' | 'professional' | 'friendly' | 'neutral',
-  context?: string
-): string {
-  const contextClause = context
-    ? ` The user is writing in this context: "${context}".`
-    : '';
-  return `You are a text suggestion assistant. Given the user's partial text, generate exactly ${count} possible continuations or completions.${contextClause} ${SUGGEST_TONE_INSTRUCTIONS[tone]} Output ONLY the suggestions, one per line, numbered like "1. suggestion here". Do not include any other text or explanation.`;
-}
-
-function buildSmartReplyPrompt(
-  count: number,
-  tone: 'formal' | 'casual' | 'professional' | 'friendly' | 'neutral',
-  persona?: string
-): string {
-  const personaClause = persona ? ` You are replying as: ${persona}.` : '';
-  return `You are a smart reply assistant. Given a conversation, generate exactly ${count} short, contextually appropriate reply suggestions that the user could send as their next message.${personaClause} ${SUGGEST_TONE_INSTRUCTIONS[tone]} Each reply should be a complete, ready-to-send message. Output ONLY the replies, one per line, numbered like "1. reply here". Do not include any other text or explanation.`;
-}
-
-function buildAutocompletePrompt(
-  count: number,
-  maxWords: number,
-  context?: string
-): string {
-  const contextClause = context
-    ? ` The user is writing about: "${context}".`
-    : '';
-  return `You are an autocomplete assistant. Given the user's partial text, generate exactly ${count} natural completions of the current sentence or phrase. Each completion should be at most ${maxWords} words and should seamlessly continue from the user's text.${contextClause} Output ONLY the completions, one per line, numbered like "1. completion here". Do not repeat the user's text. Do not include any other text or explanation.`;
-}
-
-function parseSuggestions(raw: string): { text: string }[] {
-  return raw
-    .split('\n')
-    .map((line) => line.replace(/^\d+[\.\)]\s*/, '').trim())
-    .filter((line) => line.length > 0)
-    .map((text) => ({ text }));
-}
-
-function buildAnswerQuestionPrompt(
-  detail: 'brief' | 'medium' | 'detailed'
-): string {
-  return `You are a question-answering assistant. Answer questions based ONLY on the provided context. ${ANSWER_DETAIL_INSTRUCTIONS[detail]} If the answer cannot be found in the context, say so. Do not make up information.`;
-}
-
 /**
  * Check if on-device AI is available on the current device.
  * Returns false on unsupported platforms (web, etc.).
@@ -270,17 +128,6 @@ export async function sendMessage(
  * // Cancel after 5 seconds
  * setTimeout(() => stop(), 5000);
  * ```
- *
- * @example
- * ```ts
- * // React state update pattern
- * const [text, setText] = useState('');
- *
- * streamMessage(
- *   [{ role: 'user', content: 'Hello!' }],
- *   (event) => setText(event.accumulatedText)
- * );
- * ```
  */
 export function streamMessage(
   messages: LLMMessage[],
@@ -353,795 +200,6 @@ export function streamMessage(
   return { promise, stop };
 }
 
-// ============================================================================
-// Prompt Helpers
-// ============================================================================
-
-/**
- * Summarize text content using on-device AI.
- *
- * @param text - The text to summarize
- * @param options - Optional settings for summary style and length
- * @returns Promise with the generated summary
- *
- * @example
- * ```ts
- * // Basic summarization
- * const result = await summarize(longArticle);
- * console.log(result.text);
- * ```
- *
- * @example
- * ```ts
- * // Short bullet-point summary
- * const result = await summarize(longArticle, {
- *   length: 'short',
- *   style: 'bullets'
- * });
- * ```
- *
- * @example
- * ```ts
- * // TL;DR style
- * const result = await summarize(longArticle, {
- *   style: 'tldr'
- * });
- * ```
- */
-export async function summarize(
-  text: string,
-  options?: LLMSummarizeOptions
-): Promise<LLMResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { text: '' };
-  }
-
-  if (!text || text.trim().length === 0) {
-    throw new Error('text cannot be empty');
-  }
-
-  const length = options?.length ?? 'medium';
-  const style = options?.style ?? 'paragraph';
-  const systemPrompt = buildSummarizePrompt(length, style);
-
-  return sendMessage([{ role: 'user', content: text }], { systemPrompt });
-}
-
-/**
- * Summarize text with streaming output.
- *
- * @param text - The text to summarize
- * @param onToken - Callback for each token received
- * @param options - Optional settings for summary style and length
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamSummarize(
- *   longArticle,
- *   (event) => setSummary(event.accumulatedText),
- *   { style: 'bullets' }
- * );
- * await promise;
- * ```
- */
-export function streamSummarize(
-  text: string,
-  onToken: LLMStreamCallback,
-  options?: LLMSummarizeOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!text || text.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('text cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const length = options?.length ?? 'medium';
-  const style = options?.style ?? 'paragraph';
-  const systemPrompt = buildSummarizePrompt(length, style);
-
-  return streamMessage([{ role: 'user', content: text }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Translate text to another language using on-device AI.
- *
- * @param text - The text to translate
- * @param options - Translation options including target language
- * @returns Promise with the translated text
- *
- * @example
- * ```ts
- * // Basic translation
- * const result = await translate('Hello, world!', { to: 'Spanish' });
- * console.log(result.text); // "¡Hola, mundo!"
- * ```
- *
- * @example
- * ```ts
- * // Formal translation with source language
- * const result = await translate('Hey, what\'s up?', {
- *   to: 'French',
- *   from: 'English',
- *   tone: 'formal'
- * });
- * ```
- */
-export async function translate(
-  text: string,
-  options: LLMTranslateOptions
-): Promise<LLMResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { text: '' };
-  }
-
-  if (!text || text.trim().length === 0) {
-    throw new Error('text cannot be empty');
-  }
-
-  const { to, from, tone = 'neutral' } = options;
-  const systemPrompt = buildTranslatePrompt(to, from, tone);
-
-  return sendMessage([{ role: 'user', content: text }], { systemPrompt });
-}
-
-/**
- * Translate text with streaming output.
- *
- * @param text - The text to translate
- * @param onToken - Callback for each token received
- * @param options - Translation options including target language
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamTranslate(
- *   'Hello, world!',
- *   (event) => setTranslation(event.accumulatedText),
- *   { to: 'Japanese' }
- * );
- * await promise;
- * ```
- */
-export function streamTranslate(
-  text: string,
-  onToken: LLMStreamCallback,
-  options: LLMTranslateOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!text || text.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('text cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const { to, from, tone = 'neutral' } = options;
-  const systemPrompt = buildTranslatePrompt(to, from, tone);
-
-  return streamMessage([{ role: 'user', content: text }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Rewrite text in a different style using on-device AI.
- *
- * @param text - The text to rewrite
- * @param options - Rewrite options specifying the target style
- * @returns Promise with the rewritten text
- *
- * @example
- * ```ts
- * // Make text more formal
- * const result = await rewrite('hey can u help me out?', {
- *   style: 'formal'
- * });
- * console.log(result.text); // "Would you be able to assist me?"
- * ```
- *
- * @example
- * ```ts
- * // Simplify complex text
- * const result = await rewrite(technicalText, { style: 'simple' });
- * ```
- */
-export async function rewrite(
-  text: string,
-  options: LLMRewriteOptions
-): Promise<LLMResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { text: '' };
-  }
-
-  if (!text || text.trim().length === 0) {
-    throw new Error('text cannot be empty');
-  }
-
-  const { style } = options;
-  const systemPrompt = buildRewritePrompt(style);
-
-  return sendMessage([{ role: 'user', content: text }], { systemPrompt });
-}
-
-/**
- * Rewrite text with streaming output.
- *
- * @param text - The text to rewrite
- * @param onToken - Callback for each token received
- * @param options - Rewrite options specifying the target style
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamRewrite(
- *   'hey whats up',
- *   (event) => setRewritten(event.accumulatedText),
- *   { style: 'professional' }
- * );
- * await promise;
- * ```
- */
-export function streamRewrite(
-  text: string,
-  onToken: LLMStreamCallback,
-  options: LLMRewriteOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!text || text.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('text cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const { style } = options;
-  const systemPrompt = buildRewritePrompt(style);
-
-  return streamMessage([{ role: 'user', content: text }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Extract key points from text using on-device AI.
- *
- * @param text - The text to extract key points from
- * @param options - Optional settings for extraction
- * @returns Promise with the key points as text
- *
- * @example
- * ```ts
- * // Extract key points from an article
- * const result = await extractKeyPoints(article);
- * console.log(result.text);
- * // "• Point 1\n• Point 2\n• Point 3"
- * ```
- *
- * @example
- * ```ts
- * // Limit to 3 key points
- * const result = await extractKeyPoints(article, { maxPoints: 3 });
- * ```
- */
-export async function extractKeyPoints(
-  text: string,
-  options?: LLMExtractKeyPointsOptions
-): Promise<LLMResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { text: '' };
-  }
-
-  if (!text || text.trim().length === 0) {
-    throw new Error('text cannot be empty');
-  }
-
-  const maxPoints = options?.maxPoints ?? 5;
-  const systemPrompt = buildExtractKeyPointsPrompt(maxPoints);
-
-  return sendMessage([{ role: 'user', content: text }], { systemPrompt });
-}
-
-/**
- * Extract key points with streaming output.
- *
- * @param text - The text to extract key points from
- * @param onToken - Callback for each token received
- * @param options - Optional settings for extraction
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamExtractKeyPoints(
- *   article,
- *   (event) => setKeyPoints(event.accumulatedText),
- *   { maxPoints: 5 }
- * );
- * await promise;
- * ```
- */
-export function streamExtractKeyPoints(
-  text: string,
-  onToken: LLMStreamCallback,
-  options?: LLMExtractKeyPointsOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!text || text.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('text cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const maxPoints = options?.maxPoints ?? 5;
-  const systemPrompt = buildExtractKeyPointsPrompt(maxPoints);
-
-  return streamMessage([{ role: 'user', content: text }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Answer a question based on provided context using on-device AI.
- *
- * @param question - The question to answer
- * @param context - The context/document to base the answer on
- * @param options - Optional settings for the answer
- * @returns Promise with the answer
- *
- * @example
- * ```ts
- * // Answer a question about a document
- * const result = await answerQuestion(
- *   'What is the main topic?',
- *   documentText
- * );
- * console.log(result.text);
- * ```
- *
- * @example
- * ```ts
- * // Get a detailed answer
- * const result = await answerQuestion(
- *   'Explain the methodology',
- *   researchPaper,
- *   { detail: 'detailed' }
- * );
- * ```
- */
-export async function answerQuestion(
-  question: string,
-  context: string,
-  options?: LLMAnswerQuestionOptions
-): Promise<LLMResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { text: '' };
-  }
-
-  if (!question || question.trim().length === 0) {
-    throw new Error('question cannot be empty');
-  }
-
-  if (!context || context.trim().length === 0) {
-    throw new Error('context cannot be empty');
-  }
-
-  const detail = options?.detail ?? 'medium';
-  const systemPrompt = buildAnswerQuestionPrompt(detail);
-  const userContent = `Context:\n${context}\n\nQuestion: ${question}`;
-
-  return sendMessage([{ role: 'user', content: userContent }], { systemPrompt });
-}
-
-/**
- * Answer a question with streaming output.
- *
- * @param question - The question to answer
- * @param context - The context/document to base the answer on
- * @param onToken - Callback for each token received
- * @param options - Optional settings for the answer
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamAnswerQuestion(
- *   'What are the key findings?',
- *   documentText,
- *   (event) => setAnswer(event.accumulatedText),
- *   { detail: 'detailed' }
- * );
- * await promise;
- * ```
- */
-export function streamAnswerQuestion(
-  question: string,
-  context: string,
-  onToken: LLMStreamCallback,
-  options?: LLMAnswerQuestionOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!question || question.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('question cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  if (!context || context.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('context cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const detail = options?.detail ?? 'medium';
-  const systemPrompt = buildAnswerQuestionPrompt(detail);
-  const userContent = `Context:\n${context}\n\nQuestion: ${question}`;
-
-  return streamMessage([{ role: 'user', content: userContent }], onToken, {
-    systemPrompt,
-  });
-}
-
-// ============================================================================
-// Smart Suggestions
-// ============================================================================
-
-/**
- * Generate text suggestions based on partial input using on-device AI.
- *
- * Useful for text completion, writing assistance, and predictive text features.
- *
- * @param partialText - The text the user has typed so far
- * @param options - Optional settings for suggestions
- * @returns Promise with an array of suggestions
- *
- * @example
- * ```ts
- * // Basic suggestions
- * const result = await suggest('I think we should');
- * result.suggestions.forEach(s => console.log(s.text));
- * // "schedule a meeting to discuss this further"
- * // "consider an alternative approach"
- * // "move forward with the plan"
- * ```
- *
- * @example
- * ```ts
- * // With context and tone
- * const result = await suggest('Dear Mr. Johnson,', {
- *   count: 5,
- *   context: 'writing a business email about project delays',
- *   tone: 'formal'
- * });
- * ```
- */
-export async function suggest(
-  partialText: string,
-  options?: LLMSuggestOptions
-): Promise<LLMSuggestResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { suggestions: [], raw: '' };
-  }
-
-  if (!partialText || partialText.trim().length === 0) {
-    throw new Error('partialText cannot be empty');
-  }
-
-  const count = options?.count ?? 3;
-  const tone = options?.tone ?? 'neutral';
-  const systemPrompt = buildSuggestPrompt(count, tone, options?.context);
-
-  const response = await sendMessage(
-    [{ role: 'user', content: partialText }],
-    { systemPrompt }
-  );
-
-  return {
-    suggestions: parseSuggestions(response.text),
-    raw: response.text,
-  };
-}
-
-/**
- * Generate text suggestions with streaming output.
- *
- * @param partialText - The text the user has typed so far
- * @param onToken - Callback for each token received
- * @param options - Optional settings for suggestions
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamSuggest(
- *   'The best way to',
- *   (event) => setRawSuggestions(event.accumulatedText)
- * );
- * const result = await promise;
- * // Parse suggestions from result.text
- * ```
- */
-export function streamSuggest(
-  partialText: string,
-  onToken: LLMStreamCallback,
-  options?: LLMSuggestOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!partialText || partialText.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('partialText cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const count = options?.count ?? 3;
-  const tone = options?.tone ?? 'neutral';
-  const systemPrompt = buildSuggestPrompt(count, tone, options?.context);
-
-  return streamMessage([{ role: 'user', content: partialText }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Generate smart reply suggestions for a conversation using on-device AI.
- *
- * Analyzes the conversation history and generates contextually appropriate
- * reply options, similar to Gmail/Messages smart replies.
- *
- * @param messages - The conversation history to generate replies for
- * @param options - Optional settings for reply generation
- * @returns Promise with an array of reply suggestions
- *
- * @example
- * ```ts
- * // Basic smart replies
- * const result = await smartReply([
- *   { role: 'user', content: 'Hey, are you free for lunch tomorrow?' }
- * ]);
- * result.suggestions.forEach(s => console.log(s.text));
- * // "Sure, what time works for you?"
- * // "Sorry, I already have plans."
- * // "Let me check my schedule and get back to you."
- * ```
- *
- * @example
- * ```ts
- * // With persona and tone
- * const result = await smartReply(
- *   [
- *     { role: 'user', content: 'We need the report by Friday.' },
- *     { role: 'assistant', content: 'I will do my best.' },
- *     { role: 'user', content: 'Can you confirm the deadline works?' }
- *   ],
- *   { tone: 'professional', persona: 'project manager', count: 4 }
- * );
- * ```
- */
-export async function smartReply(
-  messages: LLMMessage[],
-  options?: LLMSmartReplyOptions
-): Promise<LLMSuggestResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { suggestions: [], raw: '' };
-  }
-
-  if (!messages || messages.length === 0) {
-    throw new Error('messages array cannot be empty');
-  }
-
-  const count = options?.count ?? 3;
-  const tone = options?.tone ?? 'neutral';
-  const systemPrompt = buildSmartReplyPrompt(count, tone, options?.persona);
-
-  const conversation = messages
-    .map((m) => `${m.role === 'user' ? 'Them' : 'Me'}: ${m.content}`)
-    .join('\n');
-
-  const response = await sendMessage(
-    [{ role: 'user', content: `Conversation:\n${conversation}\n\nGenerate ${count} reply suggestions for my next message.` }],
-    { systemPrompt }
-  );
-
-  return {
-    suggestions: parseSuggestions(response.text),
-    raw: response.text,
-  };
-}
-
-/**
- * Generate smart reply suggestions with streaming output.
- *
- * @param messages - The conversation history to generate replies for
- * @param onToken - Callback for each token received
- * @param options - Optional settings for reply generation
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamSmartReply(
- *   [{ role: 'user', content: 'Want to grab coffee?' }],
- *   (event) => setRawReplies(event.accumulatedText)
- * );
- * const result = await promise;
- * ```
- */
-export function streamSmartReply(
-  messages: LLMMessage[],
-  onToken: LLMStreamCallback,
-  options?: LLMSmartReplyOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!messages || messages.length === 0) {
-    return {
-      promise: Promise.reject(new Error('messages array cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const count = options?.count ?? 3;
-  const tone = options?.tone ?? 'neutral';
-  const systemPrompt = buildSmartReplyPrompt(count, tone, options?.persona);
-
-  const conversation = messages
-    .map((m) => `${m.role === 'user' ? 'Them' : 'Me'}: ${m.content}`)
-    .join('\n');
-
-  return streamMessage(
-    [{ role: 'user', content: `Conversation:\n${conversation}\n\nGenerate ${count} reply suggestions for my next message.` }],
-    onToken,
-    { systemPrompt }
-  );
-}
-
-/**
- * Autocomplete the user's current text using on-device AI.
- *
- * Generates short, natural completions that seamlessly continue from
- * the user's partial input. Ideal for real-time typing suggestions.
- *
- * @param partialText - The text the user has typed so far
- * @param options - Optional settings for autocompletion
- * @returns Promise with an array of completion suggestions
- *
- * @example
- * ```ts
- * // Basic autocomplete
- * const result = await autocomplete('How do I');
- * result.suggestions.forEach(s => console.log(s.text));
- * // "reset my password"
- * // "contact support"
- * // "cancel my subscription"
- * ```
- *
- * @example
- * ```ts
- * // With context for better suggestions
- * const result = await autocomplete('The patient presents with', {
- *   context: 'medical notes',
- *   maxWords: 15,
- *   count: 5
- * });
- * ```
- */
-export async function autocomplete(
-  partialText: string,
-  options?: LLMAutocompleteOptions
-): Promise<LLMSuggestResponse> {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { suggestions: [], raw: '' };
-  }
-
-  if (!partialText || partialText.trim().length === 0) {
-    throw new Error('partialText cannot be empty');
-  }
-
-  const count = options?.count ?? 3;
-  const maxWords = options?.maxWords ?? 10;
-  const systemPrompt = buildAutocompletePrompt(count, maxWords, options?.context);
-
-  const response = await sendMessage(
-    [{ role: 'user', content: partialText }],
-    { systemPrompt }
-  );
-
-  return {
-    suggestions: parseSuggestions(response.text),
-    raw: response.text,
-  };
-}
-
-/**
- * Autocomplete text with streaming output.
- *
- * @param partialText - The text the user has typed so far
- * @param onToken - Callback for each token received
- * @param options - Optional settings for autocompletion
- * @returns Object with stop() function and promise
- *
- * @example
- * ```ts
- * const { promise } = streamAutocomplete(
- *   'I would like to',
- *   (event) => setRawCompletions(event.accumulatedText)
- * );
- * const result = await promise;
- * ```
- */
-export function streamAutocomplete(
-  partialText: string,
-  onToken: LLMStreamCallback,
-  options?: LLMAutocompleteOptions
-): { promise: Promise<LLMResponse>; stop: () => void } {
-  if (Platform.OS !== 'ios' && Platform.OS !== 'android') {
-    return { promise: Promise.resolve({ text: '' }), stop: () => {} };
-  }
-
-  if (!partialText || partialText.trim().length === 0) {
-    return {
-      promise: Promise.reject(new Error('partialText cannot be empty')),
-      stop: () => {},
-    };
-  }
-
-  const count = options?.count ?? 3;
-  const maxWords = options?.maxWords ?? 10;
-  const systemPrompt = buildAutocompletePrompt(count, maxWords, options?.context);
-
-  return streamMessage([{ role: 'user', content: partialText }], onToken, {
-    systemPrompt,
-  });
-}
-
-/**
- * Parse suggestion text from a suggest/smartReply/autocomplete response.
- *
- * Use this to parse the raw text from streaming responses into structured suggestions.
- *
- * @param raw - Raw text response from the model
- * @returns Array of parsed suggestions
- *
- * @example
- * ```ts
- * const { promise } = streamSuggest('Hello', (event) => {
- *   setText(event.accumulatedText);
- * });
- * const result = await promise;
- * const suggestions = parseSuggestResponse(result.text);
- * ```
- */
-export function parseSuggestResponse(raw: string): LLMSuggestResponse {
-  return {
-    suggestions: parseSuggestions(raw),
-    raw,
-  };
-}
-
 // ============================================================================
 // Model Management API
 // ============================================================================