Integration overview

Chat integration is central to Capellaris: people reach the agents through the interface they already have open all day, rather than through one more tool to learn.

Integration architecture

graph TB
    subgraph Platforms["📱 Chat platforms"]
        SLACK[fa:fa-slack Slack]
        TEAMS[fa:fa-microsoft Microsoft Teams]
    end
    
    subgraph Connectors["🔌 Connectors"]
        SC[Slack connector]
        TC[Teams connector]
    end
    
    subgraph Core["⚙️ Capellaris core"]
        ROUTER[Message
router]
        CONFIRM[Human
confirmation]
        AGENTS[Specialised
agents]
    end
    
    subgraph Formatters["📝 Formatters"]
        BLOCKS[Interactive
blocks]
        CARDS[Adaptive
cards]
    end
    
    SLACK --> SC
    TEAMS --> TC
    SC --> ROUTER
    TC --> ROUTER
    ROUTER --> CONFIRM
    CONFIRM --> AGENTS
    AGENTS --> BLOCKS
    AGENTS --> CARDS
    BLOCKS --> SC
    CARDS --> TC
    
    style Platforms fill:#e3f2fd
    style Core fill:#fff3e0
    style Connectors fill:#e8f5e9
    style Formatters fill:#f3e5f5
Interfaces familiares
Collaboration in channels
Conversation context
Built-in notifications
History preserved
Enterprise security

Plataformas Suportadas

Slack

Full integration

Full integration across public channels, private channels and direct messages, with rich message support.

@agent_name mentions
Direct messages
Threaded conversations
Interactive blocks
Feedback via reactions

Microsoft Teams

Full integration

Integration with Teams channels and chats, with adaptive cards and action buttons.

@agent_name mentions
Private chats
Adaptive cards
Action buttons
Custom tabs

Next integrations (roadmap)

Discord Telegram WhatsApp Business

Interaction flow

What a typical exchange between a person and an agent looks like through the connected chat platforms.

Example interaction

sequenceDiagram
    participant U as 👤 User
    participant S as 💬 Slack/Teams
    participant C as 🔌 Connector
    participant A as 🤖 Agent
    participant H as ✅ Approver
    
    U->>S: @agente_marketing analyse the campaign
    S->>C: Forwards the message
    C->>A: Processes the request
    A->>C: Asks for approval
    C->>S: Sends an approval card
    S->>H: Notifies the approver
    H->>S: Clicks "Approve"
    S->>C: Confirmation received
    C->>A: Runs the task
    A->>C: Returns the analysis
    C->>S: Formats the response
    S->>U: Shows the result

1. The request

JD
Jane Doe 10:30

@agente_marketing analyse the results of our email campaign from last quarter

2. The approval request

Approval request

The marketing agent would like to analyse campaign data.

The human confirmation process

Human confirmation is what stops an agent taking an action nobody authorised. Control stays with the operators.

Confirmation flow

flowchart LR
    A[📋 Task
requested] --> B[🔍 Complexity
assessment]
    B --> C{Approval
level?}
    C -->|Low| D[👤 User
approves]
    C -->|Medium| E[👨‍💼 Manager
approves]
    C -->|Alto| F[👔 Diretoria
Aprova]
    D --> G[⚡ Execution]
    E --> G
    F --> G
    G --> H[📊 Report]
    
    style A fill:#e3f2fd
    style G fill:#e8f5e9
    style H fill:#fff3e0

Approval levels

Low

Simple tasks with no access to sensitive data.

Examples: basic analysis, general summaries

Medium

Tasks touching internal data that is not sensitive.

Examples: detailed reports, recommendations

High

Tasks touching sensitive or confidential data.

Examples: financial analysis, customer data

Security and control

Every approval is logged and auditable. A request left unanswered for 30 minutes is escalated or cancelled automatically, so nothing sits pending indefinitely.