Skip to content

How Agents Work

Architecture Overview

Each Nexus AI agent runs in its own isolated Docker container. This means your agent has dedicated resources, its own memory, and cannot interfere with other agents on the platform.

The Agent Stack

  • Gateway — Handles all incoming and outgoing messages, authentication, and channel routing
  • Brain — The AI model (Claude, GPT, etc.) that processes your messages and generates responses
  • Memory — Persistent storage that lets your agent remember context across conversations
  • Skills — Modular capabilities like web search, calendar management, or content creation
  • Channels — Communication endpoints (Telegram, WhatsApp, web chat, Discord, Slack)

Message Flow

When you send a message to your agent, it flows through the following path:

  • Your message arrives via a channel (e.g., Telegram, web chat)
  • The gateway authenticates and routes it to your agent's container
  • The agent loads relevant memory and conversation context
  • The brain (AI model) processes your message and generates a response
  • The response streams back through the gateway to your channel

This entire process typically takes 1-3 seconds for the first token, with the response streaming in real-time.