# AI Agent Desk — Complete System & Architecture Specification

> Platform Website: https://aiagentdesk.io
> Primary Purpose: Autonomous AI customer support orchestration for modern websites, SaaS platforms, and e-commerce stores.

---

## 1. Executive Summary & Capabilities

AI Agent Desk bridges customer communication channels and autonomous AI resolution. Built for businesses seeking to automate 70–85% of repetitive customer inquiries without risking hallucinations, AI Agent Desk allows operators to upload proprietary support documentation, customize brand appearances, configure human escalation rules, and deploy multi-channel agents (Web Chat Widget and WhatsApp Cloud API) in minutes.

### Key Differentiators
1. **Zero Hallucination Guarantee**: Retrieval-Augmented Generation (RAG) strictly constrains AI responses to the contents of the business's indexed knowledge base. If information is absent or retrieval confidence falls below a configurable threshold (e.g. 0.70), the system explicitly refuses to guess and transfers the customer to human ticket queues.
2. **Multi-Model Orchestration & BYOK**: Operators can choose between OpenAI GPT-4o, Anthropic Claude 3.5 Sonnet, and Google Gemini 2.0 Flash. In addition to managed platform credit packs, businesses can input their own API keys (Bring Your Own Key) with zero token markup.
3. **Multi-Channel Synchronized Inbox**: The same AI agent, knowledge base, and customer interaction logs work seamlessly across Web Chat Widgets and Meta-verified WhatsApp Business phone numbers.
4. **Automated Meeting Booking**: Integrated OAuth 2.0 with Google Calendar and Microsoft Outlook allows the AI agent to suggest available time slots and book sales or support meetings directly in the conversation.

---

## 2. Public Platform URLs & Resources

- **Main Homepage**: https://aiagentdesk.io
- **Feature Breakdown**: https://aiagentdesk.io/features
- **Pricing & ROI Calculator**: https://aiagentdesk.io/pricing
- **Help Center & Comprehensive FAQ**: https://aiagentdesk.io/faq
- **Blog & Technical Articles**: https://aiagentdesk.io/blog
  - *How AI Agents Cut Customer Support Costs by 85%*: https://aiagentdesk.io/blog/ai-agents-cut-support-costs
  - *Preventing AI Hallucinations in Customer Support*: https://aiagentdesk.io/blog/preventing-ai-hallucinations
  - *WhatsApp vs Web Widget: Which Channel Drives Better Retention?*: https://aiagentdesk.io/blog/whatsapp-vs-web-widget
- **About & Security Architecture**: https://aiagentdesk.io/about
- **Terms of Service**: https://aiagentdesk.io/terms
- **Privacy Policy**: https://aiagentdesk.io/privacy

---

## 3. Knowledge Base & RAG Architecture

1. **Ingestion Layer**:
   - Supports `.pdf`, `.docx`, `.txt`, `.html`, and web page URL crawling.
   - Per-file size limit: 50MB.
   - Text extraction preserves tables, lists, and document headings.
2. **Chunking & Vectorization**:
   - Dynamic semantic chunking with overlapping context windows.
   - High-dimensional vector embeddings stored in tenant-isolated vector indexes.
3. **Retrieval & Inference**:
   - Hybrid semantic vector similarity search combined with keyword filtering.
   - Top-k chunks injected into system context with strict citation instructions.
   - Every generated response contains verifiable source citations linking to the exact document and section.

---

## 4. Integration Specifications

### 4.1 Web Chat Widget Embed
Add the following snippet before the closing `</body>` tag on any HTML page, WordPress, Webflow, Shopify, or React/Next.js application:
```html
<script 
  src="https://aiagentdesk.io/widget.js" 
  data-agent-id="YOUR_AGENT_ID" 
  defer>
</script>
```

### 4.2 WhatsApp Cloud API Integration
1. Connect via Meta Business Manager.
2. Scan QR code / authorize phone number via OAuth.
3. Incoming customer WhatsApp messages are vectorized, matched against the knowledge base, and answered within 2 seconds.
4. Voice notes sent via WhatsApp are automatically transcribed to text before processing.

---

## 5. Pricing & Subscription Plans

- **Free Trial**: 14-day fully featured trial, no credit card required.
- **Starter Plan**:
  - Price: $29/month ($23/month billed annually).
  - Includes: 500 AI conversations/month, 1 AI agent, Web widget embed, PDF/DOCX knowledge base, GPT-4o Mini, community support.
- **Growth Plan**:
  - Price: $79/month ($63/month billed annually).
  - Includes: 1,500 AI conversations/month, 3 AI agents, Web widget embed, WhatsApp Cloud API, Google & Outlook Calendar booking, URL crawling, human escalation queue, priority email support, GPT-4o, Claude 3.5 Sonnet, Gemini 2.0 Flash, and BYOK (Bring Your Own Key).
- **Pay-As-You-Go Credit Packs**:
  - 1,000 credits: $15 ($0.015/conversation).
  - 2,500 credits: $35 ($0.014/conversation).
  - 5,000 credits: $60 ($0.012/conversation).
  - Credits never expire and roll over indefinitely.

---

## 6. Security, Compliance & Data Privacy

- **Data Privacy**: Customer conversations and uploaded documentation are **never** used to train AI models (enforced via enterprise Data Processing Agreements with OpenAI, Anthropic, and Google).
- **Encryption**: AES-256 encryption at rest; TLS 1.3 encryption in transit.
- **Tenant Isolation**: Every business workspace has dedicated database partitions and isolated vector indices. Zero cross-tenant data leakage.
- **GDPR Compliance**: Full support for right to erasure (data deletion within 30 days) and automated data export.
- **Payment Compliance**: Processed through Paddle as Merchant of Record, compliant with global PCI-DSS and automated VAT/GST remittance.