OpenClaw Masterclass: From Installation to Production-Grade AI Agent Runtime
Master the full capabilities of the self-hosted AI Agent runtime OpenClaw from scratch. 40-episode tutorial covering installation, CLI commands, permissions, Skill & Plugin development, Telegram/Discord integration, multi-Agent orchestration, Memory systems, AgentMail, MCP ecosystem, and VPS deployment.
Start Learning
Curriculum
Course Introduction and Phase 40 Panoramic Roadmap (Syllabus)
This section introduces the end-to-end learning path of the 40-session masterclass, from installation and deployment to production-grade multi-Agent systems, guiding you to fully master all capabilities of the OpenClaw self-hosted AI Agent runtime.
Phase 1 / Ep 01: What is OpenClaw? — Understanding the Revolutionary Significance of Self-Hosted AI Agents
A paradigm shift from traditional ChatBots to self-hosted Agent Runtimes. Understanding OpenClaw's core concepts: Gateway, Agent, Channel, Skill, Plugin, MCP, and why it can change your relationship with AI.
Phase 1 / Ep 02: macOS Native Installation —— 5-Minute Quick Start
Install OpenClaw on macOS with a single command, walk through the Onboarding wizard, and complete the LLM configuration and initial health check.
Phase 1 / Ep 03: Docker Containerized Installation —— Best Practices for Secure Isolation
Run OpenClaw using Docker, understand the security value of containerized isolation, and master Volume mounting, environment variable management, and container health check configuration.
Phase 1 / Ep 04: UTM Virtual Machine Deployment —— Complete Practical Implementation in a Linux Environment
Running Ubuntu + OpenClaw in a UTM virtual machine on an Apple Silicon Mac, achieving kernel-level isolation, and configuring systemd daemon and SSH remote management.
Phase 1 / Ep 05: Directory Structure Dissection —— What Does an Agent's "Body" Look Like?
A comprehensive interpretation of the function of each file and directory within the ~/.openclaw/ directory. SOUL.md is the soul, MEMORY.md is the long-term memory, and sessions/ stores the conversation history. Understanding the structure is crucial for mastering the Agent.
Phase 1 / Ep 06: Your First Message — The Complete Conversation Flow from CLI to Telegram
Initiate the first conversation with the Agent via two channels: CLI and Telegram. Trace the complete message flow from user input to LLM invocation, and finally to the returned response.
Phase 2 / Ep 07: Gateway Management —— start, stop, restart, and daemon mode
Master OpenClaw Gateway's lifecycle management. Understand the distinction between the control plane and the data plane, and configure daemon mode to enable automatic startup at boot.
Phase 2 / Ep 08: Configuration Management —— config get/set and openclaw.json Deep Dive
Proficient in OpenClaw's configuration hierarchy, mastering the `openclaw config` command and full field interpretation of `openclaw.json`.
Phase 2 / Ep 09: Skills Command —— Discover and Install Skills from ClawHub
Master Skill's full lifecycle management commands: list, install, check, remove. Understand browsing and trust evaluation of the ClawHub marketplace.
Phase 2 / Ep 10: Channels Commands —— The Art of Multi-Channel Management
Mastering access and management commands for multiple messaging platforms, including adding, status checking, and troubleshooting for Telegram, Discord, and WhatsApp.
Phase 2 / Ep 11: Agent Management —— agents add, list, switch
Understand the management commands of multi-agent architecture. Create specialized agents, workspace isolation mechanisms, and independent configuration for Bootstrap files.
Phase 2 / Ep 12: Logging System —— logs, sessions, Practical Troubleshooting
Proficient in OpenClaw's multi-layer logging system: Gateway logs, Agent session logs, LLM invocation logs. Master troubleshooting methodology.
Phase 3 / Ep 13: Permission Model Overview —— tools.profile and Security Philosophy
Understand OpenClaw's tiered permission design. The four levels from basic to full, and the capability boundaries and applicable scenarios of each level. Always follow the principle of least privilege.
Phase 3 / Ep 14: Coding Permissions in Action — Make the Agent Your Coding Assistant
Configure coding permissions and enable the Agent to complete real programming tasks: clone a repository, create a branch, write code, run tests, and make a commit.
Phase 3 / Ep 15: Full Permissions and Security Policies —— Governance of High-Risk Environments
Understand the risks and mitigation measures of full permissions. Run the full-permission Agent in isolated Docker/VM environments, and configure command blacklists/whitelists and manual approval mechanisms.
Phase 3 / Ep 16: Multi-Agent Privilege Isolation —— Assigning Least Privilege by Role
Design a differentiated permission architecture for multiple Agents. Each Agent has independent TOOLS.md and agent.json configurations to achieve segregation of duties.
Phase 4 / Ep 17: Understanding Skill Architecture — SKILL.md is the Soul
In-depth understanding of Skill's definition specifications and operational mechanisms. SKILL.md's YAML frontmatter format, Markdown instruction authoring, and scripts directory extension.
Phase 4 / Ep 18: Built-in Core Skill Tour —— Top 10 Must-Install Skills
Explore the 10 most practical built-in and community Skills for OpenClaw: web-search, file-manager, shell-exec, calendar, etc. Each Skill comes with a dual-platform usage comparison between CLI and Telegram.
Phase 4 / Ep 19: Develop Your First Skill From Scratch —— "Package Tracking" Practical Application
Develop a complete custom Skill from scratch: create the directory structure, write SKILL.md, write auxiliary scripts, perform local testing, and register with the Agent.
Phase 4 / Ep 20: Advanced Skill Development —— Complex Skill with Scripts and Resource Files
Develop enterprise-grade Skills with multiple scripts and template files. Resources directory usage, cross-Agent sharing strategies, and version management.
Phase 4 / Ep 21: ClawHub Marketplace — Publish Your Skill to the Global Community
Publish custom Skills to the ClawHub marketplace. Account registration, publishing process, version management, documentation writing, and community review mechanism.
Phase 4 / Ep 22: Skill Security Audit — Essential Preparations Before Installing ClawHub
Establish Skill security review awareness. Identify red flags in SKILL.md, review the scripts directory, and cross-validate permission declarations.
Phase 5 / Ep 23: Plugin vs Skill —— When to Use a Plugin?
Clearly distinguish the use cases of Plugin and Skill. Plugin's independent process model, zero Token consumption feature, and middleware pipeline position.
Phase 5 / Ep 24: Plugin Pipeline Architecture —— Onion Model for Message Processing
Understand the Plugin's position and working mechanism within the message pipeline. The complete pipeline flow of messages from user to Agent and back to user.
Phase 5 / Ep 25: Develop Your First Plugin —— \"Message Counter\"
Develop a message counter Plugin from scratch. Learn about the Plugin directory structure, manifest file, writing Hooks, and Gateway registration.
Phase 5 / Ep 26: Hands-on Plugin —— \"Notion Synchronizer\"
Develop a practical Plugin: Automatically sync the summary of each conversation to a Notion database. Notion API integration, error handling, and batch synchronization.
Phase 5 / Ep 27: Plugin Release and Lifecycle Management
Publish Plugin to ClawHub, manage version updates, hot updates and cold restarts, and deprecation and migration strategies.
Phase 6 / Ep 28: Telegram Deep Integration —— From BotFather to Production-Grade Bot
Fully configure a Telegram channel and achieve production quality. Full BotFather workflow, Pairing, group permissions, message format optimization, and custom commands.
Phase 6 / Ep 29: Discord Deep Integration —— Bot Creation to Server Deployment
Deploy OpenClaw Bot on a Discord server. Developer Portal configuration, OAuth2 permissions, channel binding, and Slash Commands.
Phase 6 / Ep 30: Multi-frequency Path Routing Strategy —— Cross-platform Management of a Single Agent
Design of a multi-platform access routing architecture. Single Agent multi-channel, Multi-Agent multi-channel, routing rules, and failover.
Phase 6 / Ep 31: Message Channel Security Governance —— Abuse Prevention · Leak Prevention · Audit Trail
Ensure message channel security. User whitelists, rate limiting, sensitive information filtering, and a comprehensive audit log system.
Phase 7 / Ep 32: Build Your Agent Legion — Multi-Agent Architecture with Specialized Division of Labor
Design and deploy a multi-agent collaboration system. Division of responsibilities and channel binding for the four major agents: Personal, Code, Ops, Research.
Phase 7 / Ep 33: Agent Customization —— SOUL.md and Personality Engineering
Customize the Agent's behavioral persona through Bootstrap files. The art of writing SOUL.md, USER.md user profile, IDENTITY.md identity marker.
Phase 7 / Ep 34: Agent Custom Workflow —— Enabling Agents to Define Their Own Working Style
Enable Agents to autonomously create and manage workflows. Cron scheduled tasks, inter-Agent collaboration, workflow templates, and Agent self-evolution.
Phase 7 / Ep 35: Session In-depth Analysis —— Lifecycle of Dialogue
Understanding the full process of Session creation, maintenance, compression (Compaction), and archiving. Context window management and Session storage format interpretation.
Phase 7 / Ep 36: Memory System —— MEMORY.md and Hierarchical Memory Architecture
Master OpenClaw's long-term memory mechanism. Three-layer memory model: Bootstrap (always loaded) → Memory Index (retrieved on demand) → Session (within the session).
Phase 7 / Ep 37: MemSkill Practical Application — Teaching Agents to "Remember" and "Forget"
Configure and use Memory-related Skills: memory-manager (journal management), memory-search (semantic search), memory-compact (memory compression).
Phase 8 / Ep 38: AgentMail —— Enabling Agents to Send and Receive Emails
Configure AgentMail MCP Server to enable the Agent to have a dedicated mailbox. Practical scenarios: automatic email replies, scheduled sending of daily reports, reading emails to extract tasks.
Phase 8 / Ep 39: MCP Ecosystem Tour —— 5 Essential MCP Servers
Understand the MCP (Model Context Protocol) ecosystem. A detailed explanation of the installation and usage of five major MCP Servers: AgentMail, Browser, Database, GitHub, and Calendar.
Phase 8 / Ep 40: Production Deployment —— 24/7 Operation of VPS + Cloudflare Tunnel
Deploy OpenClaw to a VPS to achieve 24/7 unattended operation. Docker Compose production configuration, Cloudflare Tunnel secure exposure, systemd daemonization, monitoring and alerting, and backup strategies.