OpenClaw QuickStart
Self-hosted AI agent gateway from zero to a real working stack — install, channels, skills, MCP.
01OpenClaw QuickStart (1): What This Thing Actually Is
What OpenClaw is, where it sits vs ChatGPT/Claude, and which problems it solves.
02OpenClaw QuickStart (2): Install and First Chat in 10 Minutes
Install OpenClaw, plug in a model provider, and have a working agent in ten minutes.
03OpenClaw QuickStart (3): The Six Layers That Make the Agent Loop Work
Gateway, Pi Agent, Tools, Skills, Memory, Channels — what each layer does.
04OpenClaw QuickStart (4): Configuration, Model Providers, and the Coding Plan Trick
openclaw.json walkthrough, model providers, and the Bailian Coding Plan trick.
05OpenClaw QuickStart (5): Wiring Telegram, DingTalk, and the WeChat Reality
Telegram in five minutes, DingTalk in fifteen, WeChat via WorkBuddy.
06OpenClaw QuickStart (6): Skills, MCP, and Shipping Something Real
Write a Skill, attach a Playwright MCP server, and ship a morning-briefing agent.
07OpenClaw QuickStart (7): The Memory System, Without the Magic
MEMORY.md as index, memoryFlush, bge-m3 search, and the v2026.3.7 ContextEngine.
08OpenClaw QuickStart (8): Heartbeat, Cron, and Getting Pinged at 7am
Heartbeat patrols and Cron timers — configs for daily briefings and weekly reports.
09OpenClaw QuickStart (9): The China IM Picker, with Honest Tradeoffs
DingTalk, WeCom, WeChat WorkBuddy — a picker matrix with actual config blocks.
10OpenClaw QuickStart (10): Production Deploy and the Failure Modes Nobody Warns You About
ECS, pm2, nginx, acme.sh — plus eight production failure modes and their fixes.