A technical deep-dive into the infrastructure, security, integrations, and SLAs behind MoeCloud's autonomous AI agent platform.
Every component is fully managed, auto-scaling, and globally distributed on Google Cloud Platform.
Anthropic Claude (Haiku/Sonnet) via shared callLLM abstraction. Prompt engineering with agent-specific personas, guardrails, and confidence thresholds.
Firebase Cloud Functions v2 (Gen 2) on Google Cloud Run. Node.js 22 runtime, auto-scaling, 900s max timeout for long-running pipelines.
Firestore for all operational data — tickets, leads, invoices, agent health, action logs. Real-time listeners for dashboard and agent coordination.
Google Cloud Storage for campaign assets, video renders, audio files, and document archives. Signed URLs for secure access.
Twilio for inbound/outbound calls with StatusCallback tracking. ElevenLabs for agent TTS voices. DTMF fallback IVR.
HeyGen Avatar IV for talking-head renders. Submagic for post-production (captions, B-roll, effects). Upload-Post for multi-platform publishing.
Every layer of the platform is designed with security, compliance, and auditability as first-class requirements.
All API keys and credentials stored in GCP Secret Manager. No plaintext secrets in code or environment variables. Runtime-only access via defineSecret().
Firebase Authentication with role-based permissions. Portal access gated by canAccessPage() middleware. Admin/viewer/agent role hierarchy.
All data encrypted at rest (AES-256) and in transit (TLS 1.3). Firestore and GCS use Google-managed encryption keys by default.
Every agent action written to action_log with agent_id, timestamp, description, and outcome. Full audit trail queryable by client, agent, or time range.
Critical decisions require Moses approval via approval_queue. Confidence thresholds prevent autonomous action on low-certainty classifications.
Pre-flight resource checks before expensive pipeline runs. Submagic credit exhaustion, ElevenLabs character limits, and HeyGen quota monitoring.
Inbound/outbound calls, SMS, StatusCallback webhooks
TTS generation for 5 agent voices (turbo v2.5)
Avatar IV talking-photo video rendering
AI captions, B-roll, zoom effects, music
TikTok, Instagram, LinkedIn distribution
Remote monitoring, device inventory, ticket sync
Agent email pipes, SMTP relay, push-based ingestion
Haiku (fast) + Sonnet (complex) via shared client
Every metric tracked in real-time and available to clients via the dashboard. No vanity numbers — real operational targets.
Schedule a 30-minute architecture walkthrough with our engineering team. We'll show you the live system, answer your integration questions, and discuss your security requirements.