Quickstart: Deploy 7 AI Agents in 30 Minutes
From defining your VIP Commerce thresholds to monitoring all 7 dedicated AI agents (4 event-driven + 3 operational-input). Every invocation is logged to the Concierge audit trail so you can trace every AI action back to its trigger and result.
5-Step Agent Deployment
Enable the Module
Navigate to Admin > System > Modules. Enable Concierge. Ensure you also have the modules needed by each agent active: CRM + Inbox (VIP Welcome, Sentiment Guard, Lead Enrichment), Commerce (VIP Welcome), Auth/Identity (Lead Enrichment), Hospitality (Loyalty Liaison), Tasks + Calendar (Chief of Staff, Assistant, Travel). Missing modules trigger graceful degradation — the agent logs the gap and exits cleanly for that invocation instead of throwing.
Configure the 4 Event-Driven Agent Thresholds
Go to Admin > Automation > Concierge.
Schedule the 3 Operational-Input Agents
Different shape — these are cron / API-triggered, not event-driven.
Trigger Test Events
- VIP Welcome: Commerce module → mark a $2,500 test order as "Paid"
Verify the Orchestration in the Audit Log
Open Admin → Concierge → Audit Log. Filter by agent or trigger. Every test invocation is logged as in-progress with the trigger details, then updated with the full result and an AI-generated flag on success — or marked as failed with a machine-readable reason (`ai_not_configured` / `ai_call_failed` / `ai_response_malformed`) when something goes wrong. Check the **AI Generated** quick-filter in your CRM contacts to see the test customer tagged `VIP` and `UltraVIP`.
typescript
// A Concierge audit log entry — logged before the AI is ever called{ agent: "VIP Welcome", trigger: "Order paid ($2,500)", status: "completed", summary: { orderId: "ord_789", total: 250000, isUltraVip: true }, aiGenerated: true, // Flag for the admin "AI Generated" quick-filter confidence: 92, // 0-100, above threshold → auto-acted impact: "Recognized high-value customer ($2,500) and initiated white-glove onboarding.", createdAt: "2026-05-19T10:30:00Z", completedAt: "2026-05-19T10:30:00.008Z" // Sub-10ms event-bus reaction}
// Idempotency: before acting, the agent checks for an existing action record// tied to this order — a replayed "order paid" event for the same order// won't re-fire the AI or double-tag the customer.
// Honest no-op: when AI providers aren't configured, the entry records:{status: "failed",reason: "ai_not_configured", // Surfaced directly in the admin filter// ... rather than fabricating a success}Absolutely not. Concierge reacts asynchronously, in the background. When a customer checks out, Commerce confirms the order and immediately responds to the user. The Concierge background worker picks up the event and processes it independently in single-digit milliseconds.
Graceful degradation: the agent detects the missing CRM connection, logs the gap, and exits cleanly for that invocation rather than crashing or half-applying side effects. The admin filter surfaces the failed run for operator review.
No. Cross-module idempotency is enforced before the AI is ever called, tied to the originating record (the order for VIP Welcome, the user for Lead Enrichment). A replayed "order paid" event for the same order returns immediately without re-running the AI — saving cost AND preventing duplicate CRM tags / emails / tasks.
Event-driven agents react to things happening elsewhere in the platform (an order is paid, a message arrives, a user registers, a guest checks in) in single-digit ms. Operational-input agents take structured input, run it through the AI, and produce a typed analysis. They're schedule-driven (Chief of Staff @ 7 AM) or triggered on demand (Assistant 30 min pre-meeting; Travel on itinerary submit). Both shapes write to the same audit trail.
The 7 built-in agents cover the most common customer-intelligence and executive-productivity workflows out of the box. For automations beyond those, use Blueprint — our visual, no-code workflow builder — or talk to our solutions team about a bespoke agent built as part of an Enterprise engagement.
Blueprint is a visual, general-purpose node builder for mapping out standard business processes (multi-step approvals, branching workflows). Concierge is 7 pre-built, opinionated AI agents specifically designed for deep customer intelligence (VIP routing, sentiment escalation, lead enrichment), personal-assistant workflows (morning briefings, meeting prep, travel logistics), and hospitality (returning-guest welcome). The two compose: Concierge agents can trigger Blueprint workflows through the sentiment alerts they raise.
Every agent is built on an honest-failure design principle: it returns a machine-readable reason (`ai_not_configured` / `ai_call_failed` / `ai_response_malformed`) instead of fabricating success. The audit log records the reason. Admin filters surface failures for operator review. Compare to Zapier's opaque "Zap failed" — Concierge tells you exactly what went wrong and why.
The 3 operational-input agents (Chief of Staff, Assistant, Travel) flag any output below the configured confidence threshold (default 60/100) for human review rather than auto-acting. Those items are marked for review in the audit log. Operators review, then promote to completed (firing any downstream side effects) or dismiss (with an audit-trail record of the human override).
Ready to deploy 7 AI agents?
Full audit trail. Honest failure reporting. Cross-module duplicate protection. See /platform/concierge for the full marketing overview or /platform/concierge/tech-specs for the architecture deep-dive.