| name | openclaw-docs-expert |
| description | Expert on the official OpenClaw documentation (docs.openclaw.ai). OpenClaw is a self-hosted multi-channel AI agent gateway — NOT a skills framework. Always consult before answering any OpenClaw question. Trigger on: openclaw, AGENTS.md, SOUL.md, IDENTITY.md, USER.md, TOOLS.md, HEARTBEAT.md, BOOT.md, MEMORY.md, openclaw workspace, openclaw config, openclaw gateway, openclaw onboard, openclaw channels, openclaw skills, openclaw tools, openclaw sessions, openclaw heartbeat, openclaw cron, openclaw doctor, openclaw sandbox, openclaw telegram, openclaw whatsapp, openclaw discord, openclaw plugins, openclaw memory, openclaw dashboard, openclaw multi-agent, openclaw security, openclaw remote, openclaw tailscale, openclaw hooks, gateway port 18789, pi-mono, "what is openclaw", "how to set up openclaw". Trigger even for simple or casual OpenClaw questions — details are easy to confuse with other tools.
|
| version | 2.0.0 |
| metadata | {"category":"documentation","domain":"openclaw","sources":["https://docs.openclaw.ai","https://github.com/openclaw/openclaw"]} |
OpenClaw Docs Expert
OpenClaw is a self-hosted multi-channel AI agent gateway written in TypeScript.
A single Gateway process connects messengers (WhatsApp, Telegram, Discord, Slack,
iMessage, Signal, and others) with an Agent Runtime on a local or self-hosted server.
⚠️ Rule: answer ONLY based on docs.openclaw.ai and
github.com/openclaw/openclaw. If a fact is not officially confirmed — explicitly
mark it as ⚠️ Not confirmed in official documentation.
Documentation Map
| Section | URL | What it covers |
|---|
| Gateway Architecture | docs.openclaw.ai/concepts/architecture | Hub-and-spoke, WS protocol, clients, nodes |
| Agent Runtime | docs.openclaw.ai/concepts/agent | Bootstrap files, tools, skills, sessions |
| Agent Workspace | docs.openclaw.ai/concepts/agent-workspace | Workspace file map, git backup |
| Configuration | docs.openclaw.ai/gateway/configuration | openclaw.json, hot reload, env |
| Config Reference | docs.openclaw.ai/gateway/configuration-reference | All configuration fields |
| Tools & Plugins | docs.openclaw.ai/tools | Built-in tools, skills, plugins |
| Skills | docs.openclaw.ai/tools/skills | What a skill is, SKILL.md structure |
| Creating Skills | docs.openclaw.ai/tools/creating-skills | Writing custom skills |
| Channels | docs.openclaw.ai/channels | WhatsApp, Telegram, Discord, etc. |
| Security | docs.openclaw.ai/gateway/security | Sandboxing, auth, threat model |
| Heartbeat | docs.openclaw.ai/gateway/heartbeat | Periodic check-ins |
| Cron | docs.openclaw.ai/automation/cron-jobs | Scheduled tasks |
| Getting Started | docs.openclaw.ai/start/getting-started | Installation, onboarding |
| Doctor | docs.openclaw.ai/gateway/doctor | Diagnostics and auto-fix |
| Multi-Agent | docs.openclaw.ai/concepts/multi-agent | Routing, multiple agents |
| Session | docs.openclaw.ai/concepts/session | Scoping, identity links |
| Memory | docs.openclaw.ai/concepts/memory | Persistent memory workflow |
Detailed extracts for each section → references/sections.md
Answer Workflow
Step 1 — Classify the question:
- Install/Setup →
start/getting-started, openclaw onboard
- Workspace/Files →
concepts/agent-workspace + references/workspace-files.md
- Config →
gateway/configuration + references/sections.md#configuration
- Tools/Skills/Plugins →
tools/ + references/sections.md#tools--skills--plugins
- Channels →
channels/<provider>
- Security/Sandbox →
gateway/security, gateway/sandboxing
- Automation →
automation/cron-jobs, automation/hooks, gateway/heartbeat
- Architecture →
concepts/architecture + references/sections.md#architecture
- Troubleshooting →
gateway/doctor, gateway/troubleshooting
Step 2 — For workspace files (AGENTS.md, SOUL.md, etc.):
→ read references/workspace-files.md
Step 3 — Compose the answer:
- Short answer (1–3 sentences)
- Details + code/config from official docs
- Link to the docs section
⚠️ if the behavior is not officially confirmed
Quick Answers
"What is OpenClaw?"
Self-hosted AI agent gateway in TypeScript (111k+ ⭐). A single Gateway process
unifies all messengers into a single control plane, routes messages to the
Agent Runtime, and supports tools, skills, plugins, cron, and webhooks.
"How to install?"
npm install -g openclaw@latest
openclaw onboard --install-daemon
openclaw dashboard
Requirements: Node 24 (recommended) or Node 22.14+.
"Where does the config live?"
~/.openclaw/openclaw.json (JSON5). The Gateway watches the file and
applies most changes hot, without a restart.
"Where does the agent workspace live?"
~/.openclaw/workspace (by default). Changed via
agents.defaults.workspace in the config.
"What is the Gateway?"
A WebSocket server at 127.0.0.1:18789 — the single control plane between
messengers and the Agent Runtime. One Gateway per host.
"Skill vs Plugin vs Tool?"
- Tool — a typed function the agent calls (
exec, browser, message...)
- Skill — a
SKILL.md in the system prompt, teaches the agent WHEN and HOW to use tools
- Plugin — an npm package, can register channels/tools/skills/providers
"What is openclaw doctor?"
openclaw doctor — a diagnostic command that detects configuration issues.
openclaw doctor --fix or --yes — automatically applies safe fixes.
Related Skills
skill-creator — for writing new OpenClaw skills
