ENVOY

Multi-Channel Integration

Envoy is the Integration Layer in the MADEIRA architecture, connecting AI agents to external communication channels and systems. It serves as the bridge between the intelligence layer and the outside world, handling message routing, protocol translation, and channel-specific adaptations.

Built on the Channel Adapter pattern, Envoy enables MADEIRA agents to interact seamlessly across Telegram, Slack, WhatsApp, and the A2UI Canvas web interface. Each adapter translates platform-specific events into the unified MADEIRA messaging format while respecting platform constraints and capabilities.

Envoy is not a chatbot. It is the gateway that allows intelligent agents to communicate through multiple channels while maintaining consistent behavior and context across all interactions.

Multi-Channel Integration

Request/Response Pipeline

Every message flows through a standardized pipeline that ensures consistent processing regardless of the originating channel.

Control Plane Integration

Envoy adapters connect to the MADEIRA Gateway (Control Plane) which handles authentication, agent routing, and channel multiplexing. The gateway is bound to 127.0.0.1:18789 (localhost) for security.

Each channel adapter is stateless. Session state and conversation context are managed by the MADEIRA Memory Server (Intelligence Layer), allowing seamless cross-channel continuity.

Channel Bindings

Channels are bound to agents via the openclaw.json configuration. Each agent can optionally declare channel bindings for external routing.

// openclaw.json
{
  "agents": {
    "list": [
      {
        "id": "madeira",
        "identity": { "name": "Madeira", "emoji": "🌊" },
        "bindings": {
          "telegram": { "botToken": "${TELEGRAM_BOT_TOKEN}" },
          "slack": { "appToken": "${SLACK_APP_TOKEN}" }
        }
      }
    ]
  }
}