Skip to content

Latest commit

 

History

History
499 lines (401 loc) · 12.3 KB

VERTEX_AI_SETUP_GUIDE.md

File metadata and controls

499 lines (401 loc) · 12.3 KB

🚀 Vertex AI Integration Setup Guide

DevGruGold/suite - Complete Setup Instructions

📋 WHAT WAS CHANGED

1. Code Changes

  • ✅ Modified: supabase/functions/ai-chat/index.ts
    • Added Vertex AI as primary AI provider (with OAuth)
    • Maintains fallback chain: Vertex AI → Gemini API → DeepSeek
    • Added OAuth authentication flag in responses

2. New Database Migration

  • ✅ Created: supabase/migrations/20241228_vertex_ai_oauth_setup.sql
    • oauth_connections table for storing OAuth refresh tokens
    • ai_provider_usage_log table for tracking AI provider usage
    • google_cloud_service_log table for Google Cloud API calls
    • Performance monitoring views
    • Auto-disable triggers for failing connections

⚡ QUICK START (5 MINUTES)

Step 1: Apply Database Migration

cd /path/to/suite

# Apply the migration
supabase db push

# Verify tables were created
supabase db query "
SELECT tablename 
FROM pg_tables 
WHERE schemaname = 'public' 
  AND tablename IN ('oauth_connections', 'ai_provider_usage_log', 'google_cloud_service_log')
ORDER BY tablename;
"

Expected output:

              tablename
------------------------------------
 ai_provider_usage_log
 google_cloud_service_log
 oauth_connections

Step 2: Configure Google Cloud OAuth

Option A: Using Environment Variables (Fastest)

# Set required secrets in Supabase
supabase secrets set GOOGLE_CLIENT_ID=your_client_id_here
supabase secrets set GOOGLE_CLIENT_SECRET=your_client_secret_here
supabase secrets set GOOGLE_REFRESH_TOKEN=your_refresh_token_here
supabase secrets set GOOGLE_CLOUD_PROJECT_ID=your_project_id_here

# Verify secrets are set
supabase secrets list

Option B: Using Database (Recommended for Production)

-- Insert OAuth connection into database
INSERT INTO public.oauth_connections (
    provider,
    provider_email,
    refresh_token,
    scopes,
    is_active,
    metadata
) VALUES (
    'google_cloud',
    'xmrtsolutions@gmail.com',
    'YOUR_REFRESH_TOKEN_HERE',
    ARRAY[
        'https://www.googleapis.com/auth/gmail.send',
        'https://www.googleapis.com/auth/drive',
        'https://www.googleapis.com/auth/spreadsheets',
        'https://www.googleapis.com/auth/calendar',
        'https://www.googleapis.com/auth/cloud-platform'
    ],
    true,
    '{"source": "manual_setup", "setup_date": "2024-12-28"}'::jsonb
);

Step 3: Get OAuth Refresh Token (if you don't have it)

3a. Test OAuth Status

curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/google-cloud-auth \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{"action":"status"}'

If response shows "ready": false, continue with 3b-3d:

3b. Get Authorization URL

curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/google-cloud-auth \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{"action":"get_authorization_url"}'

Response contains:

{
  "authorization_url": "https://accounts.google.com/o/oauth2/v2/auth?client_id=...",
  "instructions": "Open this URL..."
}

3c. Complete OAuth Flow

  1. Copy the authorization_url from step 3b
  2. Open in browser
  3. Sign in with xmrtsolutions@gmail.com
  4. Grant all permissions
  5. You'll be redirected to callback with refresh token

3d. Store Refresh Token

# Store in Supabase secrets
supabase secrets set GOOGLE_REFRESH_TOKEN=<token_from_callback>

# OR insert into database (see Option B above)

Step 4: Deploy Updated Edge Function

# Deploy ai-chat function with Vertex AI integration
supabase functions deploy ai-chat

# Watch deployment logs
supabase functions logs ai-chat --follow

Step 5: Test the Integration

Test 1: Verify OAuth Status

curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/google-cloud-auth \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{"action":"status"}'

Expected Response:

{
  "success": true,
  "configured": {
    "client_id": true,
    "client_secret": true,
    "refresh_token": true
  },
  "ready": true,
  "available_services": ["gmail", "drive", "sheets", "calendar", "gemini"]
}

Test 2: Test Vertex AI Directly

curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/vertex-ai-chat \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "Hello, test Vertex AI"}
    ]
  }'

Expected Response:

{
  "success": true,
  "data": {
    "choices": [{"message": {"content": "Hello! I'm the Vertex AI..."}}],
    "provider": "vertex"
  },
  "executive": {
    "name": "ML Operations Specialist",
    "aiService": "vertex"
  }
}

Test 3: Test ai-chat (Should Use Vertex AI)

curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/ai-chat \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "What AI provider are you using?"}
    ]
  }'

Expected Response (KEY FIELD: "provider": "vertex-ai"):

{
  "success": true,
  "response": "I am currently using Vertex AI...",
  "executive": "ai-chat",
  "executiveTitle": "AI Assistant [Vertex AI]",
  "provider": "vertex-ai",
  "model": "gemini-1.5-pro",
  "oauth_authenticated": true
}

✅ SUCCESS INDICATOR: Response shows "provider": "vertex-ai"

📊 MONITORING & VERIFICATION

Query 1: Check OAuth Connection Health

SELECT * FROM public.oauth_connection_health;

Expected Result:

provider      | total_connections | active_connections | most_recent_use
--------------|-------------------|--------------------|-----------------
google_cloud  | 1                 | 1                  | 2024-12-28 ...

Query 2: Verify Vertex AI is Being Used

SELECT 
    provider,
    COUNT(*) as request_count,
    COUNT(*) FILTER (WHERE success = TRUE) as successful,
    AVG(execution_time_ms) as avg_time_ms,
    COUNT(*) FILTER (WHERE oauth_authenticated = TRUE) as oauth_requests
FROM public.ai_provider_usage_log
WHERE timestamp > NOW() - INTERVAL '1 hour'
GROUP BY provider
ORDER BY request_count DESC;

Expected Result:

provider    | request_count | successful | avg_time_ms | oauth_requests
------------|---------------|------------|-------------|---------------
vertex-ai   | 15            | 15         | 850         | 15
gemini-api  | 2             | 2          | 920         | 0

✅ Success: vertex-ai should have highest request count with oauth_requests > 0

Query 3: Check Recent AI Requests

SELECT 
    provider,
    model,
    oauth_authenticated,
    execution_time_ms,
    success,
    fallback_level,
    timestamp
FROM public.ai_provider_usage_log
ORDER BY timestamp DESC
LIMIT 10;

Query 4: Monitor Google Cloud Service Usage

SELECT 
    service,
    operation,
    COUNT(*) as operation_count,
    AVG(execution_time_ms) as avg_time_ms,
    COUNT(*) FILTER (WHERE success = TRUE) as successful
FROM public.google_cloud_service_log
WHERE timestamp > NOW() - INTERVAL '24 hours'
GROUP BY service, operation
ORDER BY operation_count DESC;

🔍 TROUBLESHOOTING

Issue 1: OAuth Status Shows "ready": false

Diagnosis:

supabase secrets list

Look for:

  • GOOGLE_CLIENT_ID
  • GOOGLE_CLIENT_SECRET
  • GOOGLE_REFRESH_TOKEN
  • GOOGLE_CLOUD_PROJECT_ID

Solution: Complete Step 3 (Get OAuth Refresh Token) above

Issue 2: Vertex AI Returns 403 Forbidden

Cause: Vertex AI API not enabled in Google Cloud Project

Solution:

  1. Go to: https://console.cloud.google.com/apis/library/aiplatform.googleapis.com
  2. Click "Enable"
  3. Wait 2-3 minutes for propagation
  4. Test again

Issue 3: Response Still Shows "provider": "gemini-api"

Possible Causes:

  1. Function not deployed
  2. Vertex AI failing silently
  3. OAuth not configured

Debug Steps:

# Check deployment
supabase functions list

# Watch logs in real-time
supabase functions logs ai-chat --follow

# Send test message and watch for:
# - "🔵 Attempting Vertex AI"
# - Any error messages

Look for these log messages:

  • ✅ 🔵 Attempting Vertex AI (Google Cloud OAuth)...
  • ✅ ✅ Vertex AI succeeded
  • ⚠️ ⚠️ Vertex AI unavailable or failed, falling back to Gemini API (indicates OAuth issue)

Issue 4: Database Migration Failed

Check Error:

supabase db query "SELECT * FROM _supabase_migrations ORDER BY inserted_at DESC LIMIT 5;"

Common Errors:

  • "already exists" - Tables created, migration safe to skip
  • "permission denied" - Run as database owner/service role

Force Reapply (if needed):

# Drop and recreate (CAUTION: loses data in these tables)
supabase db query "
DROP TABLE IF EXISTS public.google_cloud_service_log CASCADE;
DROP TABLE IF EXISTS public.ai_provider_usage_log CASCADE;
DROP TABLE IF EXISTS public.oauth_connections CASCADE;
"

# Reapply migration
supabase db push

📈 EXPECTED BEHAVIOR AFTER SETUP

Before Fix:

User → ai-chat → Gemini API → Response
  • No OAuth
  • No Google Cloud integration
  • Direct API calls only

After Fix:

User → ai-chat → Vertex AI (OAuth) → Response
                    ↓ (if fails)
                Gemini API → Response
                    ↓ (if fails)
                DeepSeek → Response
  • ✅ OAuth authenticated
  • ✅ Google Cloud integration (Gmail, Drive, Sheets, Calendar)
  • ✅ Enterprise-grade Vertex AI
  • ✅ Maintained fallback chain

🎯 SUCCESS CHECKLIST

  • Database migration applied successfully
  • Tables created: oauth_connections, ai_provider_usage_log, google_cloud_service_log
  • Google OAuth secrets configured (either env or database)
  • OAuth status shows "ready": true
  • vertex-ai-chat test returns success
  • ai-chat test returns "provider": "vertex-ai"
  • Frontend chat uses Vertex AI (check browser console)
  • Monitoring queries return data
  • Logs show "Vertex AI succeeded" messages

📚 ADDITIONAL RESOURCES

Database Schema Documentation

Monitoring Views

  • oauth_connection_health - OAuth connection status
  • ai_provider_performance - 24-hour AI provider metrics
  • google_cloud_service_summary - 7-day Google Cloud usage

Edge Functions

  • google-cloud-auth - OAuth flow and token management
  • vertex-ai-chat - Vertex AI API integration
  • ai-chat - Main chat router (now includes Vertex AI)

🔐 SECURITY NOTES

  1. Refresh Token Security

    • Stored in oauth_connections.refresh_token (plaintext)
    • PRODUCTION RECOMMENDATION: Encrypt with Supabase Vault or similar
    • Auto-disabled after 5 consecutive failures

  2. Row Level Security (RLS)

    • Enabled on all new tables
    • Service role has full access
    • Adjust policies based on your auth model

  3. Token Rotation

    • Access tokens automatically refreshed
    • Last refresh tracked in last_refreshed_at
    • Error tracking in error_count column

📞 SUPPORT

If you encounter issues:

  1. Check logs:

    supabase functions logs ai-chat --follow
supabase functions logs google-cloud-auth --follow

  2. Verify database:

    SELECT * FROM public.oauth_connection_health;
SELECT * FROM public.ai_provider_performance;

  3. Test OAuth directly:

    curl -X POST https://YOUR_PROJECT.supabase.co/functions/v1/google-cloud-auth \
  -H "Authorization: Bearer YOUR_ANON_KEY" \
  -d '{"action":"get_access_token"}'

Setup Time: ~15-20 minutes
Testing Time: ~10 minutes
Total Time: ~30 minutes for complete integration

🎉 Congratulations! Your Vertex AI integration with Google Cloud OAuth is now live!