AI Chat

Chat AI z obsługą streamingu, draftów i trybów kontekstowych (visit/patient/specialist).

AI Chat Experimental

Na tej stronie

• Overview
• Architecture
• Components
• Hook: useChatHistory
• Chat Flow
• Chat Options
• Context Modes
• Message Types
• Draft Management
• Chat Sessions
• Streaming Controls
• Detached Window
• Error States
• Related

Overview

AI Chat to interaktywny interfejs do rozmów z AI w kontekście wizyty lub ogólnych konsultacji weterynaryjnych.

---

Architecture

flowchart TB
    subgraph UI["Chat UI"]
        Input[Message Input]
        Messages[Message List]
        Streaming[Streaming Display]
    end

    subgraph Hooks
        useChatHistory[useChatHistory]
        useChat[useChat]
    end

    subgraph Client["UnifiedAIClient"]
        chat[chat]
        chatStream[chatStream]
        cancelStream[cancelStream]
    end

    subgraph Backend
        unified_ai_chat
        unified_ai_chat_stream
    end

    UI --> Hooks
    Hooks --> Client
    Client --> Backend

---

Components

Chat Panel

┌─────────────────────────────────────┐
│ AI Assistant          [Detach] [X]  │
├─────────────────────────────────────┤
│                                     │
│ 🐾 Witaj! Jak mogę pomóc?           │
│                                     │
│ 👤 Pies ma wymioty od 2 dni         │
│                                     │
│ 🐾 Rozumiem. Kilka pytań:           │
│    - Czy pies je normalnie?         │
│    - Czy są inne objawy?            │
│    - Jaki jest wiek psa?            │
│                                     │
│ 👤 ...typing...                     │
│                                     │
├─────────────────────────────────────┤
│ [Type a message...]      [Send] 🎤  │
└─────────────────────────────────────┘

---

Hook: useChatHistory

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

const {
  messages,
  isLoading,
  addMessage,
  clearMessages,
  loadDraft,
  persistDraft,
} = useChatHistory(storageKey);

Auto-persistence

// Auto-save every second (debounced)
useEffect(() => {
  const timer = setTimeout(() => {
    persistDraft(messages);
  }, 1000);
  return () => clearTimeout(timer);
}, [messages]);

// Save on unmount
useEffect(() => {
  return () => persistDraft(messagesRef.current);
}, []);

---

Chat Flow

sequenceDiagram
    participant User
    participant UI as Chat Panel
    participant Hook as useChat
    participant Client as UnifiedAIClient
    participant BE as Backend

    User->>UI: Type message
    User->>UI: Press Send

    UI->>Hook: sendMessage(text)
    Hook->>Hook: Add user message to history

    alt Streaming enabled
        Hook->>Client: chatStream(message, options)
        Client->>BE: unified_ai_chat_stream

        loop Token streaming
            BE-->>Client: token
            Client-->>Hook: onToken(token)
            Hook->>UI: Update display
        end

        BE-->>Client: complete
        Client-->>Hook: onComplete(response)
    else Non-streaming
        Hook->>Client: chat(message, options)
        Client->>BE: unified_ai_chat
        BE-->>Client: Full response
        Client-->>Hook: response
    end

    Hook->>Hook: Add AI message to history
    Hook->>UI: Update messages

---

Chat Options

interface ChatOptions {
  canonicalModel?: string;  // 'chat' | 'ai-suggestions'
  visitId?: string;         // Link to visit context
  patientId?: string;       // Patient context
  systemPrompt?: string;    // Custom system prompt
  temperature?: number;     // 0-1, creativity level
  maxTokens?: number;       // Response length limit
}

---

Context Modes

Chat może działać w różnych kontekstach:

Mode	Context	Use Case
General	None	Ogólne pytania
Visit	visitId	Pytania o konkretną wizytę
Patient	patientId	Historia pacjenta
Specialist	Custom prompt	AI Specjalista

---

Message Types

interface ChatMessage {
  id: string;
  role: 'user' | 'assistant' | 'system';
  content: string;
  timestamp: number;
  metadata?: {
    visitId?: string;
    sources?: string[];  // References
    confidence?: number;
  };
}

---

Draft Management

Drafty są zapisywane lokalnie:

// Save draft
await safeInvoke('ai_chat_draft_save', {
  context_key: storageKey,
  messages: messages.map(m => ({
    role: m.role,
    content: m.content,
    timestamp: m.timestamp,
  })),
});

// Load draft
const draft = await safeInvoke('ai_chat_draft_load', {
  context_key: storageKey,
});

---

Chat Sessions

Trwałe sesje dla historii:

// Save session
await chatSessionsApi.save({
  visitId: currentVisitId,
  messages: messages,
});

// List sessions
const sessions = await chatSessionsApi.list({
  visitId: currentVisitId,
  limit: 20,
});

---

Streaming Controls

// Start streaming
const requestId = UnifiedAIClient.chatStream(
  message,
  options,
  onToken,
  onComplete,
  onError
);

// Cancel streaming
UnifiedAIClient.cancelStream(requestId);

---

Detached Window

Chat można odłączyć do osobnego okna:

useAssistantDetachedWindow.ts

if (isTauriEnv) {
  // Create new window
  await createWindow(DETACHED_WINDOW_LABEL, {
    width: 400,
    height: 600,
    title: 'AI Assistant',
  });

  // Apply glass effect
  await safeInvoke('vista_apply_glass', {
    label: DETACHED_WINDOW_LABEL,
  });
}

---

Error States

// Handle errors gracefully
try {
  await sendMessage(text);
} catch (error) {
  if (error.code === 'STREAM_CANCELLED') {
    // User cancelled - no error message
  } else if (error.code === 'RATE_LIMIT') {
    addMessage({
      role: 'system',
      content: 'Zbyt wiele wiadomości. Odczekaj chwilę.',
    });
  } else {
    addMessage({
      role: 'system',
      content: 'Wystąpił błąd. Spróbuj ponownie.',
    });
  }
}

---

Related

• AI Suite Overview - Full AI system
• AI Assistants - Specialist modes
• AI Commands - API reference