@geenius/ai 0.1.0

package/packages/shared/src/convex/validators.ts

@@ -0,0 +1,136 @@
+// @geenius-ai/shared — src/convex/validators.ts
+
+/**
+ * Convex validators for AI actions, queries, and mutations.
+ */
+
+import { v } from 'convex/values'
+
+// ============================================================================
+// Action Validators
+// ============================================================================
+
+export const generateTextArgs = {
+    model: v.string(),
+    messages: v.any(), // AIMessage[]
+    temperature: v.optional(v.number()),
+    maxTokens: v.optional(v.number()),
+    topP: v.optional(v.number()),
+    caller: v.optional(v.string()),
+    responseFormat: v.optional(v.any()),
+}
+
+export const generateImageArgs = {
+    model: v.optional(v.string()),
+    prompt: v.string(),
+    n: v.optional(v.number()),
+    size: v.optional(v.string()),
+}
+
+export const generateAudioArgs = {
+    model: v.optional(v.string()),
+    prompt: v.string(),
+    voice: v.optional(v.string()),
+    speed: v.optional(v.number()),
+}
+
+export const transcribeAudioArgs = {
+    model: v.optional(v.string()),
+    audio: v.string(),
+    language: v.optional(v.string()),
+}
+
+export const generateVideoArgs = {
+    model: v.optional(v.string()),
+    prompt: v.string(),
+    duration: v.optional(v.number()),
+}
+
+export const testModelArgs = {
+    model: v.string(),
+    prompt: v.string(),
+    temperature: v.optional(v.number()),
+    maxTokens: v.optional(v.number()),
+}
+
+// ============================================================================
+// Log Validators
+// ============================================================================
+
+export const saveLogArgs = {
+    requestId: v.string(),
+    model: v.string(),
+    provider: v.string(),
+    caller: v.string(),
+    type: v.string(),
+    timestamp: v.number(),
+    durationMs: v.number(),
+    systemPrompt: v.string(),
+    userPrompt: v.string(),
+    hasImage: v.boolean(),
+    imageSizeBytes: v.optional(v.number()),
+    temperature: v.optional(v.number()),
+    maxTokens: v.optional(v.number()),
+    requestBodySize: v.number(),
+    status: v.union(v.literal('success'), v.literal('error')),
+    httpStatus: v.number(),
+    responseContent: v.string(),
+    responseSize: v.number(),
+    finishReason: v.optional(v.string()),
+    promptTokens: v.optional(v.number()),
+    completionTokens: v.optional(v.number()),
+    totalTokens: v.optional(v.number()),
+    errorMessage: v.optional(v.string()),
+    inputCostUsd: v.optional(v.number()),
+    outputCostUsd: v.optional(v.number()),
+    totalCostUsd: v.optional(v.number()),
+}
+
+export const listLogsArgs = {
+    model: v.optional(v.string()),
+    provider: v.optional(v.string()),
+    status: v.optional(v.union(v.literal('success'), v.literal('error'))),
+    caller: v.optional(v.string()),
+    limit: v.optional(v.number()),
+}
+
+// ============================================================================
+// Model Cost Validators
+// ============================================================================
+
+export const upsertModelArgs = {
+    model: v.string(),
+    provider: v.string(),
+    displayName: v.optional(v.string()),
+    inputCostPer1k: v.number(),
+    outputCostPer1k: v.number(),
+    capabilities: v.optional(v.array(v.string())),
+    contextWindow: v.optional(v.number()),
+    isActive: v.optional(v.boolean()),
+}
+
+// ============================================================================
+// Conversation Validators
+// ============================================================================
+
+export const createConversationArgs = {
+    title: v.string(),
+    model: v.string(),
+    systemPrompt: v.optional(v.string()),
+    tags: v.optional(v.array(v.string())),
+}
+
+export const updateConversationArgs = {
+    conversationId: v.id('aiConversations'),
+    title: v.optional(v.string()),
+    model: v.optional(v.string()),
+    systemPrompt: v.optional(v.string()),
+    tags: v.optional(v.array(v.string())),
+    favorite: v.optional(v.boolean()),
+    status: v.optional(v.union(v.literal('active'), v.literal('archived'))),
+}
+
+export const sendMessageArgs = {
+    conversationId: v.id('aiConversations'),
+    content: v.string(),
+}

package/packages/shared/src/index.ts

@@ -0,0 +1,107 @@
+// @geenius-ai/shared — src/index.ts
+
+// Types
+export type {
+    AIProviderType,
+    AIGenerationType,
+    AIModelCapability,
+    AIModel,
+    AIProviderConfig,
+    AIMessageRole,
+    AIMessageContentPart,
+    AIMessage,
+    AIToolCall,
+    AIToolDefinition,
+    AIConversation,
+    AIConversationStatus,
+    AIChatMessage,
+    AIAttachment,
+    AIGenerateTextOptions,
+    AIStructuredOutputOptions,
+    AIGenerateImageOptions,
+    AIEditImageOptions,
+    AIGenerateAudioOptions,
+    AITranscribeOptions,
+    AIGenerateVideoOptions,
+    AIEmbeddingOptions,
+    AIGenerateMusicOptions,
+    AIGenerateSoundEffectOptions,
+    AIVoiceCloneOptions,
+    AIGenerationResult,
+    AIEmbeddingResult,
+    AITranscriptionResult,
+    AITranscriptionSegment,
+    AITranscriptionWord,
+    AIRealtimeConfig,
+    AIRealtimeEvent,
+    AILogEntry,
+    AIRequestStatus,
+    AIUsageStats,
+    AIConfig,
+    ChatWindowProps,
+    ModelTestPageProps,
+    AILogTableProps,
+    ImageGeneratorProps,
+    AudioPlayerProps,
+    VoiceSelectorProps,
+    AIVoiceOption,
+    RealtimeAudioProps,
+} from './types'
+
+// Config
+export {
+    defineAIConfig,
+    mergeAIConfig,
+    findProviderConfig,
+    findModelConfig,
+    getProviderForModel,
+    getModelsByCapability,
+    DEFAULT_PROVIDERS,
+    DEFAULT_MODELS,
+    DEFAULT_VOICES,
+} from './config'
+
+// Content
+export type {
+    ContentType,
+    ContentAction,
+    ContentTone,
+    ContentTemplate,
+    ContentOutputField,
+    ContentItem,
+    ContentSchema,
+    ContentFieldMeta,
+    ContentGenerateOptions,
+    ContentResult,
+} from './content'
+export { BUILT_IN_TEMPLATES, buildContentPrompt, buildActionSystemPrompt } from './content'
+
+// Memory
+export type {
+    MemoryNamespace,
+    MemoryType,
+    MemoryImportance,
+    MemoryEntry,
+    MemoryQuery,
+    MemoryStore,
+    UserPreference,
+    OrgPattern,
+    MemoryConfig,
+} from './memory'
+export { DEFAULT_MEMORY_CONFIG, buildMemoryContext, extractPreferenceHints } from './memory'
+
+// Skills
+export type {
+    SkillCategory,
+    SkillParameterType,
+    SkillParameter,
+    SkillDefinition,
+    SkillContext,
+    SkillResult,
+    SkillRegistry,
+} from './skills'
+export { BUILT_IN_SKILLS, createSkillRegistry, buildSkillPrompt } from './skills'
+
+// Lightweight AI Client (fetch-based, no provider registry required)
+export type { AIProvider, AIClientConfig, AIClientMessage, ToolCall, ToolDefinition, StreamCallbacks, AIResponse } from './ai-client'
+export { resolveAIProvider, AIClient, createAI } from './ai-client'

package/packages/shared/src/memory.ts

@@ -0,0 +1,197 @@
+// @geenius-ai/shared — src/memory.ts
+
+/**
+ * AI memory system — persistent, contextual memory for AI interactions.
+ * Supports user-level preferences, org-level patterns, and conversation memory.
+ */
+
+import type { AIProviderType } from './types'
+
+// ============================================================================
+// Memory Types
+// ============================================================================
+
+export type MemoryNamespace = 'user' | 'org' | 'conversation' | 'global'
+export type MemoryType = 'preference' | 'pattern' | 'fact' | 'interaction' | 'feedback' | 'context'
+export type MemoryImportance = 'low' | 'medium' | 'high' | 'critical'
+
+export interface MemoryEntry {
+    id: string
+    namespace: MemoryNamespace
+    type: MemoryType
+    importance: MemoryImportance
+    /** The key identifying this memory (e.g., "preferred_tone", "date_format") */
+    key: string
+    /** The stored value */
+    value: string
+    /** Structured metadata */
+    metadata?: Record<string, unknown>
+    /** Embedding vector for semantic search */
+    embedding?: number[]
+    /** Namespace-scoped ID (userId, orgId, conversationId) */
+    scopeId: string
+    /** How many times this memory has been accessed */
+    accessCount: number
+    /** Last time this memory was accessed */
+    lastAccessedAt: number
+    /** When this memory expires (0 = never) */
+    expiresAt?: number
+    createdAt: number
+    updatedAt: number
+}
+
+export interface MemoryQuery {
+    namespace?: MemoryNamespace
+    type?: MemoryType
+    key?: string
+    scopeId?: string
+    /** Semantic search query */
+    query?: string
+    /** Min importance level */
+    minImportance?: MemoryImportance
+    /** Max results */
+    limit?: number
+    /** Include expired entries */
+    includeExpired?: boolean
+}
+
+export interface MemoryStore {
+    /** Store a new memory entry */
+    set(entry: Omit<MemoryEntry, 'id' | 'accessCount' | 'lastAccessedAt' | 'createdAt' | 'updatedAt'>): Promise<MemoryEntry>
+    /** Get a specific memory by key and scope */
+    get(namespace: MemoryNamespace, key: string, scopeId: string): Promise<MemoryEntry | null>
+    /** Search memories */
+    search(query: MemoryQuery): Promise<MemoryEntry[]>
+    /** Delete a memory */
+    delete(id: string): Promise<void>
+    /** Clear all memories in a namespace/scope */
+    clear(namespace: MemoryNamespace, scopeId: string): Promise<void>
+}
+
+// ============================================================================
+// User Preferences & Org Patterns
+// ============================================================================
+
+export interface UserPreference {
+    key: string
+    value: string
+    /** How confident we are in this preference (0-1) */
+    confidence: number
+    /** Number of times observed */
+    occurrences: number
+    /** Source of the preference */
+    source: 'explicit' | 'inferred' | 'default'
+}
+
+export interface OrgPattern {
+    key: string
+    value: string
+    /** How many users in the org follow this pattern */
+    adoptionRate: number
+    /** Description of the pattern */
+    description?: string
+    examples?: string[]
+}
+
+// ============================================================================
+// Memory Config
+// ============================================================================
+
+export interface MemoryConfig {
+    /** Enable memory system */
+    enabled: boolean
+    /** Max memories per namespace */
+    maxPerNamespace: number
+    /** Auto-expire memories after N days (0 = never) */
+    autoExpireDays: number
+    /** Enable semantic search (requires embedding model) */
+    semanticSearch: boolean
+    /** Embedding model to use */
+    embeddingModel?: string
+    /** Namespaces to enable */
+    enabledNamespaces: MemoryNamespace[]
+    /** Auto-extract preferences from conversations */
+    autoExtract: boolean
+    /** Max context tokens to inject from memory */
+    maxContextTokens: number
+}
+
+export const DEFAULT_MEMORY_CONFIG: MemoryConfig = {
+    enabled: true,
+    maxPerNamespace: 1000,
+    autoExpireDays: 0,
+    semanticSearch: false,
+    embeddingModel: 'text-embedding-3-small',
+    enabledNamespaces: ['user', 'org', 'conversation'],
+    autoExtract: true,
+    maxContextTokens: 2000,
+}
+
+// ============================================================================
+// Helpers
+// ============================================================================
+
+const IMPORTANCE_ORDER: Record<MemoryImportance, number> = {
+    low: 0,
+    medium: 1,
+    high: 2,
+    critical: 3,
+}
+
+/**
+ * Build a context string from relevant memories to inject into AI prompts.
+ */
+export function buildMemoryContext(
+    memories: MemoryEntry[],
+    maxTokens: number = 2000,
+): string {
+    if (memories.length === 0) return ''
+
+    // Sort by importance (desc), then recency (desc)
+    const sorted = [...memories].sort((a, b) => {
+        const impDiff = IMPORTANCE_ORDER[b.importance] - IMPORTANCE_ORDER[a.importance]
+        if (impDiff !== 0) return impDiff
+        return b.lastAccessedAt - a.lastAccessedAt
+    })
+
+    const lines: string[] = ['<context_memory>']
+    let approxTokens = 4 // opening tag
+
+    for (const mem of sorted) {
+        const line = `- [${mem.type}] ${mem.key}: ${mem.value}`
+        const lineTokens = Math.ceil(line.length / 4) // rough token estimate
+        if (approxTokens + lineTokens > maxTokens) break
+        lines.push(line)
+        approxTokens += lineTokens
+    }
+
+    lines.push('</context_memory>')
+    return lines.join('\n')
+}
+
+/**
+ * Extract potential preferences from a conversation message.
+ */
+export function extractPreferenceHints(text: string): Array<{ key: string; value: string }> {
+    const hints: Array<{ key: string; value: string }> = []
+
+    // Language preferences
+    const langMatch = text.match(/(?:in|to|use)\s+(english|french|german|spanish|dutch|italian|portuguese|japanese|chinese|korean|arabic)/i)
+    if (langMatch) {
+        hints.push({ key: 'preferred_language', value: langMatch[1].toLowerCase() })
+    }
+
+    // Tone preferences
+    const toneMatch = text.match(/(?:tone|style|voice).*?(?:be|is|use)\s+(professional|casual|friendly|formal|technical)/i)
+    if (toneMatch) {
+        hints.push({ key: 'preferred_tone', value: toneMatch[1].toLowerCase() })
+    }
+
+    // Currency preferences
+    const currencyMatch = text.match(/(?:in|use)\s+(EUR|USD|GBP|JPY|CHF|CAD|AUD)(?:\b)/i)
+    if (currencyMatch) {
+        hints.push({ key: 'preferred_currency', value: currencyMatch[1].toUpperCase() })
+    }
+
+    return hints
+}

package/packages/shared/src/providers/base.ts

@@ -0,0 +1,103 @@
+// @geenius-ai/shared — src/providers/base.ts
+
+/**
+ * Base provider interface — all AI providers implement this.
+ */
+
+import type {
+    AIProviderType,
+    AIGenerationResult,
+    AIEmbeddingResult,
+    AITranscriptionResult,
+    AIGenerateTextOptions,
+    AIStructuredOutputOptions,
+    AIGenerateImageOptions,
+    AIEditImageOptions,
+    AIGenerateAudioOptions,
+    AITranscribeOptions,
+    AIGenerateVideoOptions,
+    AIEmbeddingOptions,
+    AIGenerateMusicOptions,
+    AIGenerateSoundEffectOptions,
+    AIVoiceCloneOptions,
+    AIMessage,
+} from '../types'
+
+export interface AIProviderInterface {
+    type: AIProviderType
+    name: string
+
+    /** Generate text completion. */
+    generateText(options: AIGenerateTextOptions): Promise<AIGenerationResult>
+
+    /** Generate structured JSON output. */
+    generateStructuredOutput?(options: AIStructuredOutputOptions): Promise<AIGenerationResult>
+
+    /** Generate an image from a text prompt. */
+    generateImage?(options: AIGenerateImageOptions): Promise<string>
+
+    /** Edit an existing image. */
+    editImage?(options: AIEditImageOptions): Promise<string>
+
+    /** Generate speech/audio from text (TTS). Returns base64. */
+    generateAudio?(options: AIGenerateAudioOptions): Promise<string>
+
+    /** Transcribe audio to text (ASR). */
+    transcribeAudio?(options: AITranscribeOptions): Promise<string | AITranscriptionResult>
+
+    /** Generate video from text prompt. Returns URL. */
+    generateVideo?(options: AIGenerateVideoOptions): Promise<string>
+
+    /** Generate embeddings. */
+    generateEmbeddings?(options: AIEmbeddingOptions): Promise<AIEmbeddingResult>
+
+    /** Generate music from text. Returns base64. */
+    generateMusic?(options: AIGenerateMusicOptions): Promise<string>
+
+    /** Generate sound effects. Returns base64. */
+    generateSoundEffect?(options: AIGenerateSoundEffectOptions): Promise<string>
+
+    /** Clone a voice from samples. Returns voice ID. */
+    cloneVoice?(options: AIVoiceCloneOptions): Promise<string>
+
+    /** List available voices. */
+    listVoices?(): Promise<Array<{ id: string; name: string; preview?: string }>>
+}
+
+/**
+ * Extract text content from AIMessage array for logging.
+ */
+export function extractPromptText(messages: AIMessage[]): {
+    systemPrompt: string
+    userPromptText: string
+    hasImage: boolean
+    imageSizeBytes: number
+} {
+    let systemPrompt = ''
+    let userPromptText = ''
+    let hasImage = false
+    let imageSizeBytes = 0
+
+    for (const msg of messages) {
+        if (typeof msg.content === 'string') {
+            if (msg.role === 'system') systemPrompt = msg.content
+            else userPromptText += msg.content
+        } else if (Array.isArray(msg.content)) {
+            for (const part of msg.content) {
+                if (part.type === 'text' && part.text) {
+                    if (msg.role === 'system') systemPrompt = part.text
+                    else userPromptText += part.text
+                }
+                if (part.type === 'image_url' && part.image_url?.url) {
+                    hasImage = true
+                    imageSizeBytes = part.image_url.url.length
+                }
+                if (part.type === 'audio' && part.audio?.data) {
+                    imageSizeBytes += part.audio.data.length
+                }
+            }
+        }
+    }
+
+    return { systemPrompt, userPromptText, hasImage, imageSizeBytes }
+}

package/packages/shared/src/providers/elevenlabs.ts

@@ -0,0 +1,155 @@
+// @geenius-ai/shared — src/providers/elevenlabs.ts
+
+/**
+ * ElevenLabs provider — dedicated TTS, voice cloning, and sound effects.
+ */
+
+import type {
+    AIProviderType,
+    AIGenerationResult,
+    AIGenerateTextOptions,
+    AIGenerateAudioOptions,
+    AIVoiceCloneOptions,
+    AIGenerateSoundEffectOptions,
+} from '../types'
+import type { AIProviderInterface } from './base'
+
+export interface ElevenLabsConfig {
+    apiKey: string
+    baseUrl?: string
+    defaultModel?: string
+    defaultVoice?: string
+}
+
+export class ElevenLabsProvider implements AIProviderInterface {
+    type: AIProviderType = 'elevenlabs'
+    name = 'ElevenLabs'
+    private apiKey: string
+    private baseUrl: string
+    private defaultModel: string
+    private defaultVoice: string
+
+    constructor(config: ElevenLabsConfig) {
+        this.apiKey = config.apiKey
+        this.baseUrl = (config.baseUrl ?? 'https://api.elevenlabs.io').replace(/\/$/, '')
+        this.defaultModel = config.defaultModel ?? 'eleven_multilingual_v2'
+        this.defaultVoice = config.defaultVoice ?? 'pNInz6obpgDQGcFmaJgB'  // Adam
+    }
+
+    async generateText(_options: AIGenerateTextOptions): Promise<AIGenerationResult> {
+        throw new Error('ElevenLabs does not support text generation. Use generateAudio instead.')
+    }
+
+    async generateAudio(options: AIGenerateAudioOptions): Promise<string> {
+        const voiceId = options.voice ?? this.defaultVoice
+        const response = await fetch(
+            `${this.baseUrl}/v1/text-to-speech/${voiceId}`,
+            {
+                method: 'POST',
+                headers: {
+                    'Content-Type': 'application/json',
+                    'xi-api-key': this.apiKey,
+                    'Accept': 'audio/mpeg',
+                },
+                body: JSON.stringify({
+                    model_id: options.model ?? this.defaultModel,
+                    text: options.prompt,
+                    voice_settings: {
+                        stability: options.voiceSettings?.stability ?? 0.5,
+                        similarity_boost: options.voiceSettings?.similarityBoost ?? 0.75,
+                        style: options.voiceSettings?.style ?? 0.0,
+                        use_speaker_boost: options.voiceSettings?.useSpeakerBoost ?? true,
+                    },
+                }),
+            },
+        )
+
+        if (!response.ok) {
+            const errorText = await response.text()
+            throw new Error(`ElevenLabs TTS error (${response.status}): ${errorText}`)
+        }
+
+        const arrayBuffer = await response.arrayBuffer()
+        const bytes = new Uint8Array(arrayBuffer)
+        let binary = ''
+        for (let i = 0; i < bytes.length; i++) {
+            binary += String.fromCharCode(bytes[i]!)
+        }
+        return btoa(binary)
+    }
+
+    async cloneVoice(options: AIVoiceCloneOptions): Promise<string> {
+        const formData = new FormData()
+        formData.append('name', options.name)
+        if (options.description) formData.append('description', options.description)
+
+        for (let i = 0; i < options.samples.length; i++) {
+            const binaryStr = atob(options.samples[i]!)
+            const bytes = new Uint8Array(binaryStr.length)
+            for (let j = 0; j < binaryStr.length; j++) {
+                bytes[j] = binaryStr.charCodeAt(j)
+            }
+            const blob = new Blob([bytes], { type: 'audio/mp3' })
+            formData.append('files', blob, `sample_${i}.mp3`)
+        }
+
+        const response = await fetch(`${this.baseUrl}/v1/voices/add`, {
+            method: 'POST',
+            headers: { 'xi-api-key': this.apiKey },
+            body: formData,
+        })
+
+        if (!response.ok) {
+            const errorText = await response.text()
+            throw new Error(`ElevenLabs voice clone error (${response.status}): ${errorText}`)
+        }
+
+        const data = await response.json() as any
+        return data.voice_id
+    }
+
+    async generateSoundEffect(options: AIGenerateSoundEffectOptions): Promise<string> {
+        const response = await fetch(`${this.baseUrl}/v1/sound-generation`, {
+            method: 'POST',
+            headers: {
+                'Content-Type': 'application/json',
+                'xi-api-key': this.apiKey,
+                'Accept': 'audio/mpeg',
+            },
+            body: JSON.stringify({
+                text: options.prompt,
+                duration_seconds: options.duration,
+            }),
+        })
+
+        if (!response.ok) {
+            const errorText = await response.text()
+            throw new Error(`ElevenLabs SFX error (${response.status}): ${errorText}`)
+        }
+
+        const arrayBuffer = await response.arrayBuffer()
+        const bytes = new Uint8Array(arrayBuffer)
+        let binary = ''
+        for (let i = 0; i < bytes.length; i++) {
+            binary += String.fromCharCode(bytes[i]!)
+        }
+        return btoa(binary)
+    }
+
+    async listVoices(): Promise<Array<{ id: string; name: string; preview?: string }>> {
+        const response = await fetch(`${this.baseUrl}/v1/voices`, {
+            headers: { 'xi-api-key': this.apiKey },
+        })
+
+        if (!response.ok) {
+            throw new Error(`ElevenLabs list voices error (${response.status})`)
+        }
+
+        const data = await response.json() as any
+        return (data.voices ?? []).map((v: any) => ({
+            id: v.voice_id,
+            name: v.name,
+            preview: v.preview_url,
+        }))
+    }
+}