FedMCP - Federal Parliamentary Information

# Launch Readiness Status **Last Updated**: 2025-12-01 05:25 UTC ## P0 Critical Blocker - RESOLVED ✅ ### Issue: Missing Platform Anthropic API Key **Status**: ✅ **FIXED** **Problem**: FREE tier users could not use chat without providing their own API key, contradicting the pricing promise of "10 free queries". **Solution Implemented**: 1. ✅ Verified `anthropic-api-key` secret exists in GCP Secret Manager (created 2025-12-01T05:17:49) 2. ✅ Granted Cloud Run service account access: ```bash gcloud secrets add-iam-policy-binding anthropic-api-key \ --member="serviceAccount:canadagpt-frontend-sa@canada-gpt-ca.iam.gserviceaccount.com" \ --role="roles/secretmanager.secretAccessor" ``` 3. ✅ Updated deployment script (`scripts/deploy-frontend-cloudrun.sh`) to include ANTHROPIC_API_KEY 4. ✅ Deployed new revision: `canadagpt-frontend-00052-jvf` 5. ✅ Verified environment variable configuration in Cloud Run **Verification**: ```bash # Secret exists gcloud secrets describe anthropic-api-key # Created: 2025-12-01T05:17:49.646061Z # IAM policy grants access gcloud secrets get-iam-policy anthropic-api-key # Role: roles/secretmanager.secretAccessor # Cloud Run configuration gcloud run services describe canadagpt-frontend --region=us-central1 # Environment: ANTHROPIC_API_KEY → secretKeyRef: anthropic-api-key:latest ``` **Chat API Logic** (packages/frontend/src/app/api/chat/route.ts:248-258): - Checks for user BYOK keys (Anthropic or OpenAI) - Falls back to `process.env.ANTHROPIC_API_KEY` if no BYOK key found - Platform key is now available for FREE tier users --- ## Remaining Issues ### P1 - High Priority (Implement Soon After Launch) #### Issue: No Response Caching **Impact**: High cost risk - every query generates fresh AI response **Problem**: - Popular questions asked hundreds of times = 100x API costs - Example: "Who are current MPs?" could be cached for 24 hours - Estimated 30-40% cost savings with basic caching **Quick Win Solutions**: 1. Cache tool call results (GraphQL queries) - 2 days 2. Pre-compute daily summaries - 1 week 3. Implement Redis for response caching - 2 weeks ### P2 - Medium Priority (Fix Before Scale) #### Issue: Tier Definition Inconsistencies **Impact**: Potential quota bypass, billing disputes, support confusion **Problems Identified**: 1. **Case sensitivity**: - `auth.ts:153` sets `subscription_tier: 'FREE'` (uppercase) - `quota/route.ts:134` checks `subscription.tier === 'free'` (lowercase) 2. **Query limit mismatch**: - Pricing page promises "200 queries per month" (BASIC), "1,000 queries per month" (PRO) - `quota/route.ts:143-150` implements daily quota checks, not monthly 3. **Multiple definition locations**: - Tiers defined in 5+ files (auth.ts, tierLimits.ts, pricing page, migrations) - No single source of truth **Solution**: Consolidate tier definitions into single constants file --- ## Testing Required ### Manual Testing Checklist - [ ] **FREE tier chat flow**: - [ ] Sign up new FREE user (no BYOK key) - [ ] Send a chat query - [ ] Verify response uses platform API key (check logs: "Using platform key") - [ ] Verify quota decrements correctly - [ ] Test 10-query limit enforcement - [ ] Verify error message when quota exhausted - [ ] **BYOK flow**: - [ ] Add user's Anthropic API key in settings - [ ] Send chat query - [ ] Verify response uses user's key (check logs: "Using user Anthropic key") - [ ] Verify quota doesn't decrement for BYOK users - [ ] **Tier upgrades**: - [ ] Upgrade FREE → BASIC (BYOK option) - [ ] Upgrade FREE → BASIC (+Usage option) - [ ] Verify quota limits change correctly - [ ] Verify billing starts/stops appropriately ### Automated Testing Needed - [ ] Unit tests for quota enforcement - [ ] Integration tests for chat API key selection logic - [ ] End-to-end tests for tier transitions --- ## Architecture Notes ### Chat Flow ``` User Message ↓ Chat API (/api/chat/route.ts) ├─ Check quota (can_user_query RPC) ├─ Check for user's BYOK key (Anthropic/OpenAI) ├─ Fallback to platform key (NOW AVAILABLE ✅) ↓ Claude Sonnet 4.5 (streaming) ├─ 17 tools available ├─ Multi-round tool calling ↓ Tool Executor (toolExecutor.ts) ├─ Maps to GraphQL queries ├─ Apollo Client (cache-first) ↓ GraphQL API (@neo4j/graphql) ↓ Neo4j Database (2M+ nodes) ``` ### Current Caching Layers 1. ✅ Apollo Client (GraphQL data) - 60-80% hit rate 2. ✅ Next.js ISR (static pages) - 1 hour TTL 3. ❌ AI response caching - **NONE** (P1 to implement) 4. ❌ Daily summaries - **NONE** (P1 to implement) --- ## Deployment Details **Deployment Script**: `scripts/deploy-frontend-cloudrun.sh` **Key Changes Made**: - Line 62: Added documentation for `anthropic-api-key` secret - Line 137: Added `ANTHROPIC_API_KEY=anthropic-api-key:latest` to `--set-secrets` **Service Configuration**: - Service: `canadagpt-frontend` - Region: `us-central1` - Latest Revision: `canadagpt-frontend-00052-jvf` - Service Account: `canadagpt-frontend-sa@canada-gpt-ca.iam.gserviceaccount.com` - URL: `https://canadagpt-frontend-i7veukpr2q-uc.a.run.app` - Custom Domain: `https://canadagpt.ca` (via domain mapping) **Secrets Configured**: - supabase-url, supabase-anon-key, supabase-service-role-key - canadagpt-auth-secret (NextAuth) - google-client-id, google-client-secret, github-client-id, github-client-secret, facebook-client-id, facebook-client-secret, linkedin-client-id, linkedin-client-secret (OAuth) - encryption-key (for BYOK key encryption) - stripe-secret-key, stripe-publishable-key, stripe-webhook-secret - **anthropic-api-key** ← **NEWLY ADDED** --- ## Next Steps ### Immediate (Before Public Launch) 1. **Manual Testing**: Test FREE tier chat flow end-to-end with real user 2. **Monitor Costs**: Track API usage for first 24-48 hours 3. **Set Alerts**: Configure billing alerts for unexpected API costs ### Short-Term (Week 1-2) 1. **Implement Response Caching**: Start with tool result caching (2 days) 2. **Fix Tier Inconsistencies**: Consolidate to single source of truth (1 day) 3. **Add Pre-Query Warnings**: Alert users at 80% quota (4 hours) ### Medium-Term (Month 1) 1. **Pre-compute Daily Summaries**: Cache common queries (1 week) 2. **Implement Redis**: Full response caching layer (2 weeks) 3. **Add Monitoring**: Track quota usage patterns, identify abuse (1 week) --- ## Questions for User 1. **Manual Testing**: Would you like to test the FREE tier chat flow now, or wait until after launch? 2. **Cost Monitoring**: Should we set a daily API cost alert threshold? (Recommended: $50/day) 3. **Tier Inconsistencies**: Should we prioritize fixing the case sensitivity issue before launch? (Recommended: Yes) 4. **Response Caching**: What's your priority - implement before launch or shortly after?