Introduction
OpenClaw: Under the Hood
Section titled “OpenClaw: Under the Hood”This course is a deep technical dive into how OpenClaw works internally — written for developers and power users who want to understand the system beyond what the official documentation covers.
What this is not:
- A getting-started guide (see docs.openclaw.ai for that)
- A guide to model selection or the TUI
- A tutorial on connecting channels
What this is:
- An architectural deep dive into every major subsystem
- Explanations of how things fit together internally
- Source-backed analysis of real implementation decisions
- A reference you can return to when building on top of OpenClaw
The 10 Chapters
Section titled “The 10 Chapters”| # | Chapter | What you’ll learn |
|---|---|---|
| 1 | Gateway Architecture | The WebSocket broker, routing, agent dispatch |
| 2 | Workspace Structure | File hierarchy, loading order, precedence |
| 3 | The Agent Loop | Turn lifecycle, tool execution, thinking levels |
| 4 | Session Management | Key construction, isolation, compaction, storage |
| 5 | Channel & Account Model | How channels work, multi-account, plugin interface |
| 6 | Multi-Agent Routing | Bindings, isolation, cross-agent communication |
| 7 | Memory System | File-based memory, context injection, search |
| 8 | Skills & Tools | Built-in tools, AgentSkills spec, skill loading |
| 9 | Automation | Cron, hooks, standing orders, TaskFlow |
| 10 | Security & Permissions | Exec policy, sandboxing, auth profiles |
Built by LangCraft. Source analysis based on OpenClaw v2026.4.x.