OpenClaw — Personal Agent Gateway Cheatsheet

One-line: OpenClaw is a self-hosted personal AI gateway that connects messaging channels such as WhatsApp, Telegram, Slack, Discord, and Matrix to agents while keeping workspace, credentials, and integrations under your control.

Installation

Install from the official OpenClaw site or npm package, then run onboarding.

# Official installer for macOS and Linux
curl -fsSL https://openclaw.ai/install.sh | bash

# Package-manager alternative
npm i -g openclaw

# First run
openclaw onboard

For a no-terminal setup, use the OpenClaw Easy desktop project linked in Reference and follow its current installer instructions.

First-run checklist:

1. Create a private workspace for agent files and context.
2. Connect one AI provider before adding messaging channels.
3. Add one channel first, such as WhatsApp or Telegram.
4. Test with harmless prompts before exposing personal data or automations.

Step-by-Step Setup & Optimization

Step 1 — Separate config, credentials, and workspace

Goal: Config and credentials outside the workspace, which acts as the agent’s homeTime: ~10 minLevel: beginner

OpenClaw stores app configuration and credentials under ~/.openclaw/, outside the agent workspace (~/.openclaw/workspace by default). Treat the workspace as the agent’s home directory and memory surface.

openclaw onboard   # creates ~/.openclaw/workspace with starter files

Keep sensitive files out of the workspace unless the agent truly needs them. Remember that the workspace is a default working directory, not a complete sandbox by itself.

Step 2 — Add one channel at a time

Goal: One low-risk channel live on a dedicated accountTime: ~15 minLevel: beginner

Start with a low-risk channel and a dedicated account or number.

Step 3 — Configure the agent boundary

Goal: A written operating policy the assistant must followTime: ~10 minLevel: intermediate

Write a short operating policy for the assistant:

# OpenClaw assistant policy
- Ask before sending messages to other people.
- Do not reveal secrets, tokens, or private files.
- Summarize planned actions before using integrations.
- Keep personal reminders separate from work tasks.

Always allowlist who may talk to the assistant in ~/.openclaw/openclaw.json — never run a personal gateway open to the world:

{
  channels: { whatsapp: { allowFrom: ["+15555550123"] } }
}

If your install supports sandbox settings, enable them for tools that can read or write outside the workspace.

Step 4 — Connect integrations deliberately

Goal: Each integration’s blast radius recorded before it’s enabledTime: ~15 minLevel: intermediate

OpenClaw’s value comes from channels and plugins, but each integration expands the blast radius. For each new integration, record:

• What data it can read.
• What actions it can take.
• How to revoke credentials.
• Which prompts require confirmation.

Step 5 — Operate like a production service

Goal: The gateway running as a service with backup, monitoring, and key rotationTime: ~30 minLevel: advanced

Run the gateway as a local service or server process only after backup and recovery are clear.

openclaw --help
openclaw onboard

Monitor logs, rotate provider keys periodically, and keep the channel list small enough to audit.

Best Practices

Do

• ✅ Use the official installer or npm package.
• ✅ Start with a dedicated test identity or channel.
• ✅ Keep the workspace private and free of unnecessary secrets.
• ✅ Document confirmation rules for outbound messages and integrations.
• ✅ Use sandboxing when tools can access the host filesystem.
• ✅ Maintain a credential revocation checklist.

Don’t

• ❌ Connect every personal and team channel during the first session.
• ❌ Treat the workspace as a hard security boundary without sandboxing.
• ❌ Let the agent send messages, payments, or production changes without confirmation.
• ❌ Store provider tokens in shared notes or prompts.
• ❌ Ignore logs; gateway behavior should be observable.

Quick Command Reference

Expected Outcomes

After setup, you should have:

• A working OpenClaw install and completed onboarding flow.
• A private workspace with clear boundaries and no unnecessary secrets.
• One connected AI provider and one tested messaging channel.
• Written confirmation rules for outbound messages and sensitive integrations.
• A safety checklist for adding future channels, plugins, and automations.

Reference