AI Suite Overview

Przegląd całego AI Suite: architektura Unified AI, kluczowe funkcje i modele kanoniczne.

Na tej stronie

Overview
Architecture
Core Features
Unified AI Client
Canonical Models
Context Building
Session Integration
Error Handling
Streaming
Context Vault
Feature Flags

Overview

AI Suite to zintegrowany system AI w Vista, oparty na architekturze Unified AI. Wszystkie wywołania AI przechodzą przez backend Tauri (unified_ai_* commands).

Architecture

flowchart TB
    subgraph Frontend["Frontend Components"]
        Chat[AI Chat Panel]
        SOAP[SOAP Generator]
        Tasks[Task Suggestions]
        PDF[PDF Analyzer]
        Assistants[AI Specialists]
    end

    subgraph Hooks["React Hooks"]
        useChat[useChatHistory]
        useSOAP[useSOAPGeneration]
        useTasks[useAiTaskSuggestions]
        useNotes[useNotesAgent]
    end

    subgraph UnifiedAI["Unified AI Client"]
        Client[UnifiedAIClient.ts]
    end

    subgraph Backend["Tauri Backend"]
        Commands[unified_ai_*<br/>commands]
        Providers[AI Providers<br/>OpenAI, Anthropic]
    end

    Frontend --> Hooks
    Hooks --> Client
    Client --> Commands
    Commands --> Providers

Core Features

Feature	Description	Command
SOAP Generation	Auto-generowanie notatek SOAP	`unified_ai_generate_soap`
AI Chat	Interaktywny chat z AI	`unified_ai_chat` / `_stream`
Task Suggestions	Sugestie zadań po wizycie	`unified_ai_generate_tasks`
PDF Analysis	Analiza dokumentów	`unified_ai_analyze_document`
AI Assistants	Specjaliści AI	Chat z kontekstem

Unified AI Client

Plik: src/services/ai/UnifiedAIClient.ts

// Main entry point for all AI operations
const UnifiedAIClient = {
  // Chat
  chat(message, options),
  chatStream(message, options, onToken, onComplete, onError),
  chatBackground(message, options),
  cancelStream(requestId),

  // SOAP
  generateSOAP(visitContext, templateType, userId),

  // Documents
  analyzeDocument(content, type, prompt),

  // Audio
  transcribe(audioBlob, options),
  synthesize(text, options),

  // Health
  healthCheck(canonicalModel),
  getModelMapping(canonicalModel),
  testProviders(apiKeyOverride),
};

Canonical Models

AI Suite używa abstrakcji “canonical models” do mapowania na konkretne modele:

Canonical	Use Case	Example Provider
`chat`	General chat	Libraxis AI, openAI
`ai-suggestions`	Task/SOAP suggestions	Libraxis AI
`vision`	Image analysis	Vision LLM
`transcription`	Audio STT	Whisper
`tts`	Text-to-speech	TTS Provider

// Get actual model for canonical name
const model = await UnifiedAIClient.getModelMapping('chat');
// → { provider: 'libraxis', model: 'chat' }

Context Building

AI potrzebuje kontekstu do generowania odpowiedzi:

Patient Context

interface AINotesPatientData {
  name: string;
  species: string;
  breed: string;
  age: number;
  weight: number;
  owner: string;
  medicalHistory?: string[];
  allergies?: string[];
}

Visit Context

interface AINotesVisitData {
  visitType: string;
  reasonForVisit: string;
  transcript: string;
  vetName: string;
  previousVisits?: VisitSummary[];
}

Session Integration

Wszystkie AI calls wymagają aktywnej sesji:

// invokeWithSession wraps all AI calls
invokeWithSession('unified_ai_chat', {
  message,
  options,
  session_id,  // Auto-injected
});

Error Handling

try {
  const response = await UnifiedAIClient.chat(message);
} catch (error) {
  if (error instanceof AIProviderError) {
    // Provider-specific error (rate limit, invalid key)
  } else if (error instanceof SessionNotReadyError) {
    // Session expired
  } else {
    // Generic error
  }
}

Streaming

Chat obsługuje streaming dla lepszego UX:

UnifiedAIClient.chatStream(
  message,
  options,
  (token) => {
    // Append token to response
    appendToResponse(token);
  },
  (response) => {
    // Complete response
    setFinalResponse(response);
  },
  (error) => {
    // Handle error
    showError(error);
  }
);

Context Vault

AI Suite ma “pamięć” - Context Vault:

// Save context snippets
await contextVaultApi.upsert([{
  namespace: 'visit',
  visit_id: visitId,
  label: 'patient_history',
  snippet: patientHistory,
}]);

// Query similar context
const similar = await contextVaultApi.query({
  namespace: 'visit',
  query: currentSymptoms,
  limit: 5,
});

Feature Flags

AI features mogą być włączane/wyłączane:

Feature	Setting	Default
SOAP Auto-generate	`ai_soap_enabled`	true
Task Suggestions	`ai_tasks_enabled`	true
Chat History	`ai_chat_history`	true
TTS	`tts_enabled`	false

Detailed Architecture

AI Suite to 149 plików i 22,600 LOC - największy moduł w Vista.

graph TB
    subgraph Frontend["AI Suite Frontend (149 files, 22.6k LOC)"]
        subgraph Engine["Engine Layer"]
            AIChatContext[AIChatContext]
            AIChatProvider[AIChatProvider]
            AssistantHostManager[AssistantHostManager]
        end

        subgraph Hosts["Host Components"]
            FloatingHost[AIFloatingHost<br/>Orb + Panel]
            SystemHost[AISystemHost<br/>Full-page]
            SpecialistHost[AISpecialistHost<br/>Visit context]
        end

        subgraph Hooks["Core Hooks"]
            useStreamingManager[useStreamingManager<br/>977 LOC]
            useUnifiedAI[useUnifiedAI<br/>439 LOC]
            useChatHistory[useChatHistory<br/>318 LOC]
            useChatSending[useChatSending<br/>380 LOC]
        end

        subgraph UI["UI Components"]
            InputArea[InputArea<br/>655 LOC]
            ChatCanvas[AIChatCanvas<br/>1,078 LOC]
            FloatingOrb[AIFloatingOrb<br/>570 LOC]
            MessageList[FloatingMessageList]
        end

        subgraph Agents["AI Agents"]
            NotesAgent[Notes Agent]
            SuggestionsAgent[Suggestions Agent]
            VoiceAgent[Voice Agent]
        end
    end

    subgraph Backend["Rust Backend (40 commands)"]
        UnifiedAI[unified_ai module]
        SOAP[soap_generation.rs<br/>24,961 LOC]
        ChatB[chat/service.rs]
        VectorMemory[vector_memory/]
        Diagnostics[diagnostics.rs<br/>28,669 LOC]
    end

    Engine --> Hosts
    Hooks --> Engine
    UI --> Hooks
    Agents --> Hooks
    Hooks --> UnifiedAI

AI Rust Commands (40 total)

Category	Commands
Chat	`unified_ai_chat`, `unified_ai_chat_stream`, `unified_ai_poll_response`
SOAP	`start_soap_generation`, `get_soap_generation_status`, `unified_ai_generate_soap`
Suggestions	`unified_ai_generate_suggestions`, `unified_ai_generate_tasks`, `unified_ai_generate_differential`
Memory	`get_vista_memory`, `save_vista_memory`, `ai_context_vault_*`
Sessions	`ai_chat_session_save`, `ai_chat_sessions_list`, `ai_chat_session_load`
Diagnostics	`diagnostics_ai_config`, `diagnostics_ai_curl`, `unified_ai_health_check`

Key Components LOC

Component	LOC	Odpowiedzialność
useStreamingManager	977	Token streaming, event handling
AIChatCanvas	1,078	Message rendering, scroll
InputArea	655	Message input, attachments
AIFloatingOrb	570	Floating AI button
useUnifiedAI	439	Main AI hook
useChatSending	380	Message sending logic
useChatHistory	318	Chat history management

Host Types

Host	Użycie	Kontekst
AIFloatingHost	Orb + sliding panel	Global, always available
AISystemHost	Full-page AI view	Standalone page
AISpecialistHost	Specialist mode	Visit context, patient data

AI Commands - Full API reference
SOAP Generation - SOAP details
AI Chat - Chat interface
Backend: Unified AI - Backend implementation