AI Commands

Przegląd

Unified AI to centralna warstwa AI w Viście. Wszystkie wywołania AI przechodzą przez backend Tauri (unified_ai_* commands).

Client: src/services/ai/UnifiedAIClient.ts Types: src/types/api/ai.ts

flowchart TB
    subgraph "Frontend"
        AISuite[AI Suite Panel]
        Chat[Chat Component]
        SOAP[SOAP Generator]
        Tasks[Task Suggestions]
    end

    subgraph "UnifiedAIClient"
        chat[chat / chatStream]
        transcribe[transcribe]
        synthesize[synthesize]
        generateSOAP[generateSOAP]
        analyzeDoc[analyzeDocument]
    end

    subgraph "Backend"
        unified_ai_chat
        unified_ai_chat_stream
        unified_ai_transcribe
        unified_ai_synthesize
        unified_ai_generate_soap
        unified_ai_analyze_document
    end

    AISuite --> chat
    Chat --> chat
    SOAP --> generateSOAP
    Tasks --> chat

    chat --> unified_ai_chat
    chat --> unified_ai_chat_stream

---

Chat Commands

unified_ai_chat

Główne API chatu w AI Suite.

UnifiedAIClient.chat(message: string, options?: ChatOptions): Promise<ChatResponse>

// Implementation
invokeWithSession('unified_ai_chat', buildChatPayload(...))
// → normalizeChatResponse(result)

unified_ai_chat_stream

Streaming odpowiedzi z tokenami.

UnifiedAIClient.chatStream(
  message: string,
  options: ChatOptions,
  onToken: (token: string) => void,
  onComplete: (response: ChatResponse) => void,
  onError: (error: Error) => void
)

// Events
'ai-stream-token:<requestId>'    → onToken
'ai-stream-complete:<requestId>' → onComplete
'ai-stream-error:<requestId>'    → onError

// Fire-and-forget
invokeWithSession('unified_ai_chat_stream', buildChatPayload(..., { request_id }))

unified_ai_chat_background

Chat w tle, bez pełnego UI.

UnifiedAIClient.chatBackground(message: string, options?: ChatOptions)
// → invokeWithSession('unified_ai_chat_background', payload)

unified_ai_cancel_stream

Przerwanie streamingu.

UnifiedAIClient.cancelStream(requestId: string)
// → safeInvoke('unified_ai_cancel_stream', { requestId })

Polling & Events

// Poll dłużej trwających odpowiedzi
UnifiedAIClient.pollResponse(responseId, { canonicalModel })
// → 'unified_ai_poll_response'

// Lista eventów odpowiedzi
UnifiedAIClient.listResponseEvents(responseId, { canonicalModel, starting_after })
// → 'unified_ai_list_response_events'

---

Transcription & TTS

unified_ai_transcribe

UnifiedAIClient.transcribe(audioBlob: Blob, options?: TranscribeOptions): Promise<TranscriptionResponse>

// Flow
Blob → Array<number> → invokeWithSession('unified_ai_transcribe', {
  audioData,
  language,
  format
})
// → normalizeTranscriptionResponse(result)

unified_ai_synthesize

Text-to-Speech.

UnifiedAIClient.synthesize(text: string, options?: TTSOptions): Promise<Blob>

// Flow
invokeWithSession('unified_ai_synthesize', {
  text,
  voice,
  language
})
// → number[] (audioData) → Blob('audio/mp3')

---

SOAP & Documents

unified_ai_generate_soap

Generowanie notatek SOAP.

UnifiedAIClient.generateSOAP(
  visitContext: VisitContext,
  templateType?: string,
  userId?: string
): Promise<SOAPResponse>

// Wymaga user_id
safeInvoke('unified_ai_generate_soap', {
  visit_context,
  template_type,
  user_id
})

unified_ai_analyze_document

Analiza obrazów/PDF/tekstu.

UnifiedAIClient.analyzeDocument(
  documentContent: string | ArrayBuffer,
  documentType: 'image' | 'pdf' | 'text',
  analysisPrompt: string
): Promise<AnalysisResponse>

// → invokeWithSession('unified_ai_analyze_document', { ... })

---

Task Suggestions

Service: src/services/aiTaskSuggestionsService.ts

Command	Description
unified_ai_generate_tasks	Generuje sugestie zadań
unified_ai_list_task_suggestions	Lista sugestii dla wizyty
unified_ai_accept_task_suggestions	Akceptacja sugestii
unified_ai_dismiss_task_suggestions	Odrzucenie sugestii

// Get suggestions via chat
UnifiedAIClient.getSuggestions(visitContext, suggestionType?)
// → chat(..., { canonicalModel: 'ai-suggestions' })

---

Health & Diagnostics

unified_ai_health_check

UnifiedAIClient.healthCheck(canonicalModel?: string): Promise<HealthStatus>
// → invokeWithSession('unified_ai_health_check', { canonical_model })

unified_ai_get_model_mapping

Mapowanie nazw kanonicznych na modele.

UnifiedAIClient.getModelMapping(canonicalModel: string): Promise<ModelMapping>
// → safeInvoke('unified_ai_get_model_mapping', { canonical_model })

// Canonical models
'chat'           → konkretny model chatu
'ai-suggestions' → model dla sugestii
'vision'         → model dla obrazów

unified_ai_test_providers

UnifiedAIClient.testProviders(apiKeyOverride?: string): Promise<ProviderStatus[]>
// → invokeWithSession('unified_ai_test_providers', { api_key_override })

Other Diagnostics

Command	Service
unified_ai_connectivity_probe	aiDiagnosticsService
unified_ai_mark_full_success	backgroundJobs
unified_ai_generate_suggestions	errorHandlingService
unified_ai_generate_differential	differentialService

---

Chat History & Drafts

Hook: src/features/ai-suite/hooks/useChatHistory.ts

ai_chat_draft_load

loadDraft(): Promise<ChatMessage[]>
// → safeInvoke('ai_chat_draft_load', { context_key: storageKey })

ai_chat_draft_save

persistDraft(messages: ChatMessage[]): Promise<void>

// Respects storage policy
if (!storagePolicy.shouldSaveChatHistory()) {
  messages = [];  // Send empty if history disabled
}

// → safeInvoke('ai_chat_draft_save', { context_key, messages })

Notatka

Hook auto-saves co sekundę (debounced) i przy unmount.

---

Chat Sessions

API: src/types/api/ai.ts → chatSessionsApi

ai_chat_session_save

Trwały zapis historii rozmowy.

chatSessionsApi.save({
  visitId?: string,
  messages: ChatMessage[]
})

// → safeInvoke('ai_chat_session_save', {
//     visit_id,
//     messages: [{ role, content, timestamp }]
//   })

ai_chat_sessions_list

chatSessionsApi.list({
  limit?: number,    // default 20
  offset?: number,   // default 0
  visitId?: string
})
// → safeInvoke('ai_chat_sessions_list', { limit, offset, visit_id })

---

AI Context Vault

Pamięć kontekstowa AI - przechowywanie snippetów do promptów.

API: src/types/api/ai.ts → contextVaultApi

ai_context_vault_upsert

contextVaultApi.upsert(entries: ContextEntry[])

interface ContextEntry {
  namespace: string;
  visit_id?: string;
  label: string;
  snippet: string;
  metadata?: Record<string, any>;
}

// → safeInvoke('ai_context_vault_upsert', { entries })

ai_context_vault_query

Wyszukiwanie podobnych fragmentów (embedding lub tekst).

contextVaultApi.query({
  namespace: string,
  query?: string,
  query_embedding?: number[],
  limit?: number
})
// → safeInvoke('ai_context_vault_query', params)

ai_context_vault_clear

contextVaultApi.clear(namespace?: string)
// → safeInvoke('ai_context_vault_clear', { namespace })

---

Detached Window

vista_apply_glass

Efekt “glass” dla odłączonego okna asystenta.

useAssistantDetachedWindow.ts

if (isTauriEnv) {
  safeInvoke('vista_apply_glass', { label: DETACHED_WINDOW_LABEL });
}

---

Commands Summary

Category	Commands
Chat	unified_ai_chat, unified_ai_chat_stream, unified_ai_chat_background, unified_ai_cancel_stream
Polling	unified_ai_poll_response, unified_ai_list_response_events
Audio	unified_ai_transcribe, unified_ai_synthesize
Documents	unified_ai_generate_soap, unified_ai_analyze_document
Tasks	unified_ai_generate_tasks, unified_ai_list_task_suggestions, unified_ai_accept_task_suggestions, unified_ai_dismiss_task_suggestions
Health	unified_ai_health_check, unified_ai_get_model_mapping, unified_ai_test_providers
Drafts	ai_chat_draft_load, ai_chat_draft_save
Sessions	ai_chat_session_save, ai_chat_sessions_list
Vault	ai_context_vault_upsert, ai_context_vault_query, ai_context_vault_clear

---

Powiązane dokumenty

• AI Suite Overview - Feature documentation
• SOAP Generation - SOAP notes pipeline
• Backend: Unified AI - Backend implementation