Architecture & Deep-Dive

The OpenClaw System

ClawdBot → MoltBot → OpenClaw.The world's most flexible open-source AI agent gateway. Built for mission-critical autonomy, cross-platform messaging, and deep hardware integration.

View Codebase Explore Architecture

774K+

Lines of Code

Native Apps

25+

Channel Extensions

50+

Built-in Skills

100+

Doc Pages

What is OpenClaw?

A High-Volume AI Agent Gateway

OpenClaw is more than just a chatbot wrapper. It is a central control plane designed to host long-lived, autonomous agents that can interact with the physical world through nodes and digital platforms through messengers.

State Persistence

Markdown-canonical memory architecture ensures long-term context recall without losing historical nuance.

Zero-Trust Connectivity

Devices pair via cryptographic handshakes with explicit approval workflows for sensitive tool execution.

Gateway

Mobile Node

CLI Operator

LLM Providers

Messengers

Architecture

System Components

OpenClaw is divided into three functional layers that separate orchestration, execution, and reasoning.

OpenClaw Gateway

The hub of the entire system. Maintains all websocket connections, validates protocol frames, and routes events.

Exposes typed WebSocket API
Owns all messaging provider sessions
Canvas host for A2UI HTML serving
State versioning & session deduplication

Node Clients

Native applications that expose hardware capabilities (camera, screen, GPS) as remote-invokable tools.

Native macOS, iOS, Android builds
System-level automation (Bash/Python)
Encrypted SQLite state storage
Seamless device-based pairing

LLM Brain

The logic engine. Pluggable adapter system supports every major model provider for complex reasoning.

Native Tool Calling (Anthropic/OpenAI)
Fallback & Failover model chains
System prompt injection & templating
Local execution via Ollama support

Deep Dive

Architecture Explainer

A deeper walkthrough of the repo structure and the runtime surfaces that make OpenClaw feel unified across devices.

Client Apps (Android, iOS, macOS)

Native node clients expose cameras, screens, sensors, and voice as secure, remote-invokable tools.

Android foreground node runtime
SwiftUI iOS + macOS control surfaces
Local capture pipelines for media + location
Session-key pairing with device identity

Gateway + Session Router

The always-on control plane that brokers agents, routes events, and synchronizes state across channels.

Typed WebSocket protocol frames
Channel adapters for chat platforms
Pairing approvals and session dedupe
Canvas hosting for A2UI experiences

OpenClawKit Shared Core

Shared protocol types, UI primitives, and tool schemas so every platform speaks the same language.

Protocol + capability definitions
Chat UI components and rendering
Canvas scaffolding + tool registry
Cross-platform session payloads

CLI + Local Operator

Command-line tooling for onboarding, diagnostics, configuration, and node lifecycle control.

Installer + runtime discovery
Gateway and node health checks
Config schema + environment helpers
Logs, traces, and session tooling

Browser Automation Bridge

A Chrome extension bridge enabling DOM-aware agent actions and live A2UI workflows.

DOM tree serialization + commands
Headful + headless workflows
Session-aware cookie handling
A2UI event streaming

Deployment + Infra

Multi-target deployment recipes for running the gateway across containers and hosted platforms.

Dockerfiles + docker-compose flows
Fly.io and Render definitions
Environment templates + secrets
Sandbox + browser runner images

Integration Surface

Connect Anywhere.

OpenClaw connects to your existing messaging surface. It doesn't force move your data—it bridges your AI assistant to where you already live.

WhatsApp (Baileys)

Telegram (grammY)

Discord

Slack

Signal-CLI

iMessage (Mac Bridge)

iCloud Desktop

WebChat

📱WhatsApp

📨Telegram

💬Slack

🎮Discord

📡Signal

🔵iMessage

🔷Matrix

👥MS Teams

💚Google Chat

📝Mattermost

☁️Nextcloud

⭐Tlon

🔮Nostr

📺Twitch

💚LINE

🔵BlueBubbles

💙Zalo

📞Voice Call

🌐WebChat

🤖LLM Task

Execution

Tool Architecture

How OpenClaw extends its capabilities beyond simple text generation.

PeekabooBridge

A high-performance Chrome Extension bridge that allows the agent to interact with live DOM elements, execute scripts, and automate web workflows as if it were a human user.

DOM Tree Serialization & Navigation

A2UI (Agent-to-User) Interactivity

Headless & Headful automation modes

Cookie & Session management

Agent Workspace

A sandboxed file system environment where the agent can store its own logs, memory files, and generated assets without polluting your host machine.

Markdown Canonical Search Index

JSONL Transcript Logging

Local Asset Storage (Images/Videos)

Isolated Skill Registry (.mjs files)

Connection Lifecycle (WS API)

ConnectOperator / Node

First handshake frame includes device ID & role.

Res: Hello-OkGateway

Gateway acknowledges and pushes initial state snapshot.

Req: AgentOperator

Request for reasoning session with specific workspace.

Event: AgentGateway

Streaming response with tool usage & text tokens.

Req: ExecGateway → Node

Hardware command execution (Camera/Screen/Shell).

Trust

Security: Defense in Depth

OpenClaw implements a 4-layer security model to protect your data and execution context.

Network Privacy

Encrypted tunnels via WireGuard, Tailscale, or SSH.

Gateway Auth

Token-based handshake prevents unauthorized access.

Device Pairing

Signature-based pairing for remote nodes.

Manual Approval

Human-in-the-loop for destructive tool usage.

Automated Deployment (Ansible)

The recommended way to deploy OpenClaw to production is via openclaw-ansible. It automates security-first setup including firewalls, automatic updates, and audit logging.

Read Deployment Specs →View Ansible Scripts →

Autonomous Lifecycle

Cron Jobs & Heartbeats

OpenClaw agents don't just react. They have an autonomous heartbeat that triggers cron jobs (scheduled tasks) and system health checks.

"cron": {
"schedule": "0 9 * * 1-5",
"tool": "email.summarize_inbox"
}

Deterministic Selection

Intelligent Routing

Multiple agents can run the same gateway. Routing ensures messages reach the correct "brain" based on channel, account, or group markers.

Account-based Bindings
Discord Guild/Team Isolation
Parallel Broadcast Strategies
Topic-aware Thread Routing

50+ Skills

Skills Library

Pre-built agent capabilities that extend OpenClaw's functionality across productivity, media, development, and smart home domains.

Apple Notes

Productivity

Apple Reminders

Productivity

Notion

Productivity

Obsidian

Productivity

Trello

Productivity

Things Mac

Productivity

GitHub

Development

Coding Agent

Development

tmux

Development

OpenAI Whisper

Voice

Sherpa TTS

Voice

Voice Call

Voice

Peekaboo

Browser

Canvas

A2UI

CamSnap

Media

Video Frames

Media

OpenAI Image

Media

OpenHue

Smart Home

Sonos CLI

Smart Home

Spotify

Smart Home

Weather

Location

Local Places

Location

1Password

Security

Himalaya Mail

...and many more. Create custom skills with Markdown + Python/Shell scripts.

Developer Experience

Quick Start Commands

Get up and running with OpenClaw in minutes using these essential CLI commands.

npx openclaw setup

Initialize OpenClaw and create config

openclaw gateway start

Start the local gateway server

openclaw channels add whatsapp

Connect WhatsApp via QR code

openclaw agent chat

Start interactive chat session

openclaw cron add

Schedule autonomous tasks

openclaw skills install github

Add skills from the registry

openclaw nodes pair

Pair mobile/desktop devices

openclaw doctor

Diagnose configuration issues

Full documentation available at openclaw.ai/docs with 100+ pages covering every feature.

View Full CLI Reference →

Let Us Build It For You

Want Your Own AI Agent?

Whether you want to deploy OpenClaw for your team or build a custom autonomous agent tailored to your workflow, our experts can help.

$99

Setup Consultation

1-hour expert consultation
OpenClaw setup & configuration
Channel integration walkthrough
Custom skill recommendations

Book Consultation