Architecture Overview

Understanding the Crella.ai system architecture.

High-Level Architecture

flowchart TB
    subgraph clients [Client Layer]
        Web[Web Apps]
        API[API Clients]
        Mobile[Mobile Apps]
    end
    
    subgraph gateway [Gateway Layer]
        LB[Load Balancer]
        Auth[Authentication]
        Rate[Rate Limiter]
    end
    
    subgraph orchestration [Orchestration Layer]
        Lima[Lima Orchestrator]
        Queue[Task Queue]
        State[State Manager]
    end
    
    subgraph agents [Agent Layer]
        Alpha[Alpha]
        Bravo[Bravo]
        Charlie[Charlie]
        Delta[Delta]
        Echo[Echo]
        Foxtrot[Foxtrot]
        Golf[Golf]
        Hotel[Hotel]
        India[India]
        Juliet[Juliet]
        Kilo[Kilo]
    end
    
    subgraph human [Human Layer]
        Arena[Crella Arena]
        Staff[Human Operators]
    end
    
    subgraph storage [Storage Layer]
        DB[(Database)]
        Cache[(Redis)]
        Files[(File Storage)]
    end
    
    clients --> gateway
    gateway --> orchestration
    orchestration --> agents
    orchestration --> human
    agents --> storage
    human --> storage

Component Overview

1. Client Layer

• Web Applications: React/Next.js frontends
• API Clients: REST/GraphQL integration
• Mobile Apps: React Native (planned)

2. Gateway Layer

• Load Balancer: Nginx reverse proxy
• Authentication: JWT + API keys
• Rate Limiter: Per-client throttling

3. Orchestration Layer

Lima (The Orchestrator) is the brain of the system:

flowchart LR
    Input[Task Input] --> Lima
    Lima --> Route{Route Decision}
    Route -->|Automated| Agent[AI Agent]
    Route -->|Complex| Human[Human Review]
    Route -->|Hybrid| Both[Both]
    Agent --> Output[Task Output]
    Human --> Output
    Both --> Output

Key responsibilities:

• Task classification
• Agent selection
• Workflow orchestration
• Error recovery
• State management

4. Agent Layer

The 12 BPO agents operate in specialized pipelines:

Intake Pipeline

flowchart LR
    Raw[Raw Data] --> Alpha[Alpha - Validate]
    Alpha --> Bravo[Bravo - Triage]
    Bravo --> Next[Next Stage]

Processing Pipeline

flowchart LR
    Doc[Document] --> Charlie[Charlie - Extract]
    Charlie --> Delta[Delta - Decide]
    Delta --> Result[Decision]

Enrichment Pipeline

flowchart LR
    Lead[Lead Data] --> Echo[Echo - Research]
    Echo --> Foxtrot[Foxtrot - Content]
    Foxtrot --> Hotel[Hotel - Send]

5. Human Layer

The Crella Arena provides:

• Task escalation handling
• Quality assurance
• Complex decision support
• Relationship management

6. Storage Layer

Component	Technology	Purpose
Database	PostgreSQL/SQLite	Structured data
Cache	Redis	Session, rate limiting
Files	S3/Local	Documents, media
Search	Elasticsearch	Full-text search

Data Flow Example

Lead Processing Workflow

sequenceDiagram
    participant Client
    participant Lima
    participant Alpha
    participant Echo
    participant Foxtrot
    participant Hotel
    participant Client
    
    Client->>Lima: Submit Lead
    Lima->>Alpha: Validate Data
    Alpha-->>Lima: Validated Lead
    Lima->>Echo: Enrich Lead
    Echo-->>Lima: Enriched Data
    Lima->>Foxtrot: Generate Email
    Foxtrot-->>Lima: Email Content
    Lima->>Hotel: Send Email
    Hotel-->>Lima: Delivery Status
    Lima-->>Client: Task Complete

Scaling Strategy

Horizontal Scaling

• Each agent can run multiple instances
• Lima coordinates work distribution
• Stateless design enables easy scaling

Vertical Specialization

• Agents can be fine-tuned per client
• Custom training for specific domains
• Specialized pipelines for industries

Security Architecture

flowchart TB
    subgraph external [External]
        User[User]
        API[API Client]
    end
    
    subgraph perimeter [Perimeter]
        WAF[Web Application Firewall]
        CDN[Cloudflare CDN]
    end
    
    subgraph auth [Authentication]
        JWT[JWT Validation]
        RBAC[Role-Based Access]
    end
    
    subgraph internal [Internal Network]
        Services[Crella Services]
        Agents[AI Agents]
    end
    
    external --> perimeter
    perimeter --> auth
    auth --> internal

Technology Stack

Layer	Technologies
Frontend	Next.js, React, Tailwind CSS
Backend	Node.js, TypeScript, Python
AI/ML	Claude, GPT-4, Rasa, Ollama
Database	PostgreSQL, SQLite, Redis
Infrastructure	PM2, Nginx, Docker
Monitoring	Prometheus, Grafana

Next Steps

• 12-Agent System — Detailed agent documentation
• API Overview — Integration guide
• Deployment — Self-hosting guide