OpenClaw
Use OpenClaw as a real-world case study for self-hosted agent gateways, chat channels, tools, plugins, nodes, memory, and operations.
What is OpenClaw?
OpenClaw is a self-hosted gateway for AI agents. You run one Gateway process on your own machine or server, connect it to chat apps, the browser Control UI, optional desktop/mobile nodes, and one or more AI agent runtimes.
The useful mental model is simple: OpenClaw is not just a chat app and not just a coding agent. It is the control plane that lets an agent receive messages, choose tools, route work, remember context, and act across surfaces you own.
- Surfaces
- Telegram, WhatsApp, Slack, Discord, iMessage, WebChat, Control UI, mobile nodes
- Gateway
- Auth, pairing, routing, sessions, events, health, usage
- Agent Runtime
- Models, prompts, memory, tools, subagents, automation
- Capabilities
- Plugins, skills, browser, media, MCP, ACP, sandbox backends
Start Here
5 min
Install & First Run
Install OpenClaw, run onboarding, verify the Gateway, and open the Control UI.
Context
Agent Workspace
Understand AGENTS.md, SOUL.md, MEMORY.md, daily memory files, workspace roots, and backup boundaries.
Gateway
Chat Channels
Connect Telegram, WhatsApp, Slack, Discord, iMessage, Signal, WebChat, and more.
Platform Capabilities
Tools
File, exec, browser, web, message, session, media, automation, and Tool Search surfaces.
Skills
Reusable SKILL.md instruction packs for workflows, review rubrics, and operating rules.
Plugins
Packaged runtime capabilities: tools, channels, providers, hooks, speech, media, and skills.
Control UI & Nodes
Browser dashboard, WebChat, Windows Hub, mobile nodes, Canvas, camera, screen, and voice.
Agent System Topics
Multi-Agent Routing
Isolated agents with their own workspaces, auth profiles, session stores, and channel bindings.
Memory & Sessions
MEMORY.md, daily notes, session history, compaction, QMD, active memory, dreaming, and memory wiki.
Automation
Cron jobs, background tasks, Task Flow, hooks, standing orders, and heartbeat behavior.
Models & Providers
Provider-qualified model refs, OAuth/API credentials, fallbacks, usage tracking, and SecretRefs.
Operations & Internals
Security & Trust
Gateway auth, DM pairing, shared workspace risk, tool policy, sandboxing, and elevated access.
Operations
Update, doctor, status, backup, logs, usage, troubleshooting, and release lane habits.
Architecture
Gateway protocol, runtime loop, tools, plugins, sessions, nodes, and state boundaries.
ACP & MCP
Use OpenClaw with editors, external coding harnesses, and MCP clients without confusing the roles.
Why learn OpenClaw on AgentWay?
OpenClaw exposes the hard parts of agent platforms in one system: ingress, auth, routing, tools, memory, background work, and operator safety. Treat it as a living implementation map, not a static product manual.
Continue on AgentWay
Turn OpenClaw patterns into guided practice, chat and review