Agents at a glance

AI agents are the heart of Capellaris. Each one specialises in a particular area of knowledge and can work alone or alongside other agents on tasks that need more than one speciality.

Agent ecosystem

mindmap
    root((Capellaris))
        Marketing
            Campaigns
            SEO
            Content
            Metrics
        Finance
            Budgets
            Analysis
            Reporting
            Forecasts
        Engineering
            Code
            Debugging
            Documentation
            Architecture
        HR
            Recruiting
            Training
            Policies
            Performance
        Sales
            Strategy
            Pipeline
            Proposals
            Forecasts
        Operations
            Processes
            Projects
            Logistics
            Quality
Natural language processing
Continuous learning
Standard communication
Domain specialisation
Human confirmation
Teamwork

Agent types

Marketing agent

@agente_marketing

Specialist in marketing strategy, content creation and campaign metrics.

Campaign and metrics analysis
Multi-channel content creation
SEO optimisation
Market research

Finance agent

@agente_financeiro

Specialist in financial analysis, budget planning and forecasting.

Financial statement analysis
Budget preparation
Feasibility and ROI analysis
Financial reporting

Engineering agent

@agente_dev

Specialist in code, debugging, review and technical documentation.

Code review and optimisation
Debugging and bug identification
Architecture recommendations
Technical documentation

HR agent

@agente_rh

Specialist in human resources, recruiting and people development.

Job descriptions and openings
Candidate screening
Training plans
HR policies

Sales agent

@agente_vendas

Specialist in commercial strategy, pipeline management and proposals.

Sales performance analysis
Commercial strategy
Commercial proposals
Sales forecasting

Operations agent

@agente_ops

Specialist in process optimisation, project management and operational efficiency.

Process mapping
Project planning
Efficiency analysis
Continuous improvement

Squad formation

One of the defining features of Capellaris is that it forms squads — temporary teams of agents that collaborate on complex tasks requiring several specialities.

How a squad comes together

flowchart LR
    A[📋 Complex
task] --> B[🔍 Requirements
analysis]
    B --> C[🎯 Specialities
identified]
    C --> D[🤖 Agents
selected]
    D --> E[👤 Lead
assigned]
    E --> F[📊 Workflow
created]
    F --> G[⚡ Collaborative
execution]
    G --> H[✅ Consolidated
result]
    
    style A fill:#e3f2fd
    style H fill:#e8f5e9

Example squads

Product launch squad

Plans and runs the launch of a new product.

Marketing Finance Sales Operations

Market analysis squad

Produces a full analysis of a specific market.

Marketing Finance Sales

Engineering squad

Plans and tracks software projects.

Engineering Finance Operations

Optimisation squad

Analyses and optimises internal processes.

Operations Finance RH Engineering

Agent-to-agent communication

Agents talk to each other over Google's A2A (Agent-to-Agent) protocol, which keeps collaboration between specialists efficient and coordinated.

Communication flow

sequenceDiagram
    participant MKT as 📢 Marketing
    participant FIN as 💰 Finance
    participant ORCH as ⚙️ Orchestrator
    
    MKT->>ORCH: Requests a cost analysis
    ORCH->>FIN: Forwards the request
    FIN->>FIN: Runs the analysis
    FIN->>ORCH: Returns the result
    ORCH->>MKT: Delivers the response
    MKT->>ORCH: Consolidates the final report

Message structure

  • Sender: who is sending it
  • Recipient: the target agent or squad
  • Content: text, data or references
  • Type: request, response or notification
  • Context: anything else it needs

Routing process

  1. An agent creates the message
  2. The communication system receives it
  3. The router identifies the recipient
  4. The message is delivered to the queue
  5. The recipient processes it and replies
  6. The interaction is recorded