# Dversi - Complete Platform Documentation

> AI-powered business automation platform that creates and manages specialized AI agents for workflow orchestration and system integration.

## Overview

Dversi is a full-stack AI platform designed for SMBs (20-300 employees) to automate business operations. The core differentiator: AI creates and configures the automation, rather than users manually building it.

**One-line pitch:** "Tell us your business. We create your AI team."

**Core philosophy:** "Solve First, Automate Later" - Users see immediate value before committing to automation. Every conversation builds business context, not just transactional exchanges.

---

## Problems Dversi Solves

1. **Blank Canvas Paralysis** - Users given thousands of integration options get overwhelmed
2. **Integration Tax** - Manual data copying between 5+ SaaS tools
3. **Expertise Gap** - Non-technical users can't configure traditional automation tools
4. **Maintenance Burden** - Static workflows break and require constant fixing

---

## Platform Architecture

Dversi operates on a three-layer architecture:

### Layer 1: Dversi Assistant

The AI architect that orchestrates everything:
- Discovers what systems users have
- Connects systems via OAuth or credentials
- Creates specialized agents with appropriate tools
- Hands off tasks to the right agent
- Monitors and reports on automation health

### Layer 2: Specialized Agents

Domain experts that handle specific business functions:

**Agent Types:**
- **Sales Agent** - Lead management, pipeline reporting, meeting prep
- **Finance Agent** - Invoice processing, expense management, reconciliation
- **Support Agent** - Ticket handling, knowledge lookup, escalation
- **Operations Agent** - Order sync, inventory management, data flows
- **Custom Agents** - User-defined agents for specific business needs

**Agent Capabilities:**
- Execute tasks within their domain
- Remember context about the user's business
- Create workflows to automate repetitive tasks
- Report results, issues, and recommendations
- Learn from interactions to improve over time

### Layer 3: Workflows & Triggers

Automation execution layer:

**Workflow Features:**
- Visual drag-and-drop workflow builder
- Three step types: Tool (API calls), Resource (data fetch), Prompt (AI reasoning)
- Bulk processing via iteration source
- Zero tokens for tool-only steps (95%+ cost reduction vs chat)

**Trigger Types:**
- Manual - User-initiated execution
- Scheduled - Time-based (cron) execution
- Event-based - Triggered by external events (webhooks, system changes)

**Cost Economics:**
Example: Lead enrichment for 100 leads
- Via Chat: ~250,000 tokens
- Via Workflow: 0 tokens (tool steps only)

---

## MCP Integration Framework

Dversi uses the Model Context Protocol (MCP) for integrations:

### Core Servers (Self-Hosted)

**Business Systems:**
- Odoo - Full ERP integration (CRM, Sales, Inventory, Accounting)
- Tripletex - Norwegian accounting system
- Strapi - Headless CMS integration
- FTP - File transfer operations

**Productivity:**
- Microsoft 365 - Mail, Calendar, OneDrive, Teams
- Google Workspace - Gmail, Calendar, Drive, Docs

**Professional Networks:**
- LinkedIn - Profile data, company pages, posting

**Data Sources:**
- Brreg - Norwegian Business Registry (unique competitive advantage)
- Icecat - Product information database
- PostgreSQL - Direct database operations

**Intelligence:**
- Planning - Task decomposition and scheduling
- Memory - Persistent context storage
- Knowledge - Document and information retrieval
- Time - Date/time operations and scheduling

### Community Servers (Remote)

31+ integrations including:
- **Project Management:** Notion, Asana, Linear, Jira
- **Design:** Figma, Canva
- **Finance:** Stripe, QuickBooks
- **Sales:** HubSpot, Salesforce
- **Development:** GitHub, GitLab
- **Communication:** Slack, Discord

### Authentication Methods

**Manual Authentication:**
- API keys
- Username/password
- Service accounts
- Works with all transport types

**OAuth 2.1:**
- Standard-based authentication
- Automatic token refresh
- HTTP transport only
- Secure credential storage

---

## Entity-Centric Knowledge Graph

Dversi maintains an entity-centric data model that accumulates business intelligence:

### Core Tables

- **entities** - Unified business objects (companies, contacts, products)
- **entity_relations** - Relationships between entities
- **entity_documents** - Associated documents and files
- **entity_memories** - Contextual information from interactions
- **intelligence_settings** - Configuration for intelligence features

### Cross-System Identity Mapping

Dversi maps identities across systems:
- Odoo Customer ID ↔ Tripletex Contact ID ↔ Brreg Org Number
- Enables unified view of business entities across all connected systems

### Why It's a Competitive Moat

- **Unique:** Only Dversi sees data flowing between all connected business systems
- **Defensible:** Competitors would need months to recreate entity mappings
- **Compounding:** Every workflow execution adds to the knowledge graph

---

## Business Verticals

### Sales Automation

**Capabilities:**
- Lead enrichment from multiple data sources
- Pipeline reporting and analytics
- Meeting preparation with context
- Follow-up automation and sequences
- CRM data synchronization

**Example Workflow:** New Lead Enrichment
1. Trigger: New lead in CRM
2. Fetch company data from Brreg
3. Enrich with LinkedIn company info
4. Update CRM record
5. Notify sales rep with summary

### Finance Automation

**Capabilities:**
- Invoice processing and routing
- Expense report automation
- Account reconciliation
- Financial reporting
- ERP data synchronization

**Example Workflow:** Invoice Processing
1. Trigger: New invoice received
2. Extract data via OCR
3. Match to purchase order
4. Route for approval
5. Post to accounting system

### Operations Automation

**Capabilities:**
- Order synchronization across systems
- Inventory level monitoring and alerts
- Data reconciliation workflows
- Vendor management automation
- Shipping and logistics coordination

**Example Workflow:** Low Stock Alert
1. Trigger: Scheduled daily check
2. Query inventory levels
3. Compare to reorder thresholds
4. Create purchase orders
5. Notify procurement team

### Support Automation

**Capabilities:**
- Ticket triage and routing
- Knowledge base lookup
- Escalation management
- Customer onboarding flows
- SLA monitoring and alerts

**Example Workflow:** Ticket Triage
1. Trigger: New support ticket
2. Analyze content with AI
3. Categorize and prioritize
4. Route to appropriate agent
5. Suggest relevant KB articles

---

## Multi-Tenant Architecture

### Organization Isolation

- Separate databases per organization
- Isolated Redis namespaces
- Tenant-specific MCP server instances
- Role-based access control

### Security Model

- OAuth 2.1 for external integrations
- Encrypted credential storage
- Audit logging for all operations
- GDPR-compliant data handling

---

## User Experience

### Time to Value: < 10 Minutes

1. **Sign up** (1 minute) - Create account
2. **Onboard** (2-3 minutes) - Chat with Dversi Assistant about your business
3. **First value** (1-2 minutes) - Complete first task with connected system
4. **Automation offered** (30 seconds) - Accept workflow suggestion

### Interface Modes

**Chat Mode:**
- Conversational interface with agents
- Context-aware responses
- Tool execution in conversation

**Workflow Mode:**
- Visual workflow builder
- Step-by-step execution view
- Performance analytics

**Dashboard Mode:**
- Agent activity overview
- Workflow execution history
- Business intelligence displays

---

## Competitive Positioning

### vs. Zapier/Make
- Don't build workflows manually - let AI build them
- Agents understand business context
- Accumulated intelligence over time

### vs. Lindy.ai/Relevance AI
- Multi-agent team with specialized domains
- Workflow engine for bulk operations
- Entity-centric knowledge graph

### vs. Enterprise (Workato/Tray)
- Enterprise power at SMB prices
- Minutes to setup, not weeks
- AI-first architecture

---

## Technical Stack

- **Frontend:** Next.js 15, React, TypeScript
- **Backend:** Node.js, Express
- **Database:** PostgreSQL (multi-tenant), Redis
- **Real-time:** Socket.io
- **AI:** Claude (Anthropic), Google Gemini
- **Protocol:** MCP (Model Context Protocol)

---

## Links

- **Website:** https://dversi.com
- **Platform:** https://app.dversi.com
- **Summary:** /llms.txt