Source - OpenClaw System Architecture Overview

Source - OpenClaw System Architecture Overview

Origin: https://ppaolo.substack.com/p/openclaw-system-architecture-overview
Author: ppaolo (Substack)
Fetched: 2026-04-15

Summary

Overview of OpenClaw's full system architecture: hub-and-spoke Gateway design, channel adapters for multiple messaging platforms, control interfaces, the Agent Runtime, Canvas/A2UI system, six-phase execution flow, layered security model, and deployment topologies.

Key Takeaways

Hub-and-Spoke Architecture

The central Gateway (Node.js 22+, binds 127.0.0.1:18789 by default) is the control plane. Everything — channel adapters, control interfaces, and the agent runtime — routes through it.

Channel Adapters

Platform-specific connectors handling auth, message parsing, access control, and formatting for: WhatsApp, Telegram, Discord, Slack, iMessage, and other messaging platforms.

Control Interfaces

Web UI, CLI, macOS app, and mobile nodes for human interaction with the gateway.

Agent Runtime

Executes the AI loop: assembles context (AGENTS.md, SOUL.md, TOOLS.md system-prompt files + semantically-retrieved memory via hybrid search), invokes the model, executes tools, persists state. Docker-based sandboxing for untrusted tool sessions.

Canvas / A2UI

Agent-driven visual workspace. Uses a declarative A2UI (Agent-to-UI) framework for agents to render structured UI components.

Six-Phase Execution Flow

  1. Ingestion
  2. Access control & routing
  3. Context assembly
  4. Model invocation
  5. Tool execution
  6. Response delivery

Security Model (Layered)

Deployment Options

Mode Description
Local dev All components on developer machine
macOS production LaunchAgent background service + menu bar app
Linux/VPS Remote Gateway via SSH tunnel or Tailscale
Fly.io containers Cloud-native with persistent volumes

Pages Created/Updated from This Source