Jarvis

Voice Input (MentraOS glasses / mobile)
           ↓
    Speech-to-Text
           ↓
    jarvis_conversations / jarvis_messages
           ↓
    AI Processing (Claude/GPT)
           ↓
    MCP Tool Execution
           ↓
    jarvis_tool_logs
           ↓
    Response + Text-to-Speech

"How's the business doing this week?"

→ Queries across all tables
→ Returns: posts published, orders, revenue, deliverables

"What orders came in today?"

→ SELECT * FROM shopify_orders WHERE created_at::date = CURRENT_DATE
→ Returns: order list with customer names, totals

"Which travel posts are performing best?"

→ JOIN publisher_posts with publisher_post_analytics
→ Returns: top posts by views/engagement

"Create a ticket to fix the login bug"

→ Calls Linear MCP tool
→ Creates ENG issue with description

Jeff: "Jarvis, what's my top priority today?"

Jarvis: "You have 3 Linear tickets due today. The most impactful 
is ENG-3, the documentation site setup. Publisher scraping is 
running normally with 66 posts collected. No urgent orders."

// Create conversation
const { data: conversation } = await supabase
  .from('jarvis_conversations')
  .insert({ user_id, started_at: new Date() })
  .select()
  .single();

// Add message
await supabase
  .from('jarvis_messages')
  .insert({
    conversation_id: conversation.id,
    role: 'user',
    content: transcribedText
  });

// Process with AI + MCP tools
const response = await processWithAI(transcribedText, tools);

// Log tool usage
if (response.toolCalls) {
  await supabase
    .from('jarvis_tool_logs')
    .insert(response.toolCalls.map(tc => ({
      conversation_id: conversation.id,
      tool_name: tc.name,
      input: tc.input,
      output: tc.output
    })));
}