OpenClaw Runbook
Unofficial OpenClaw runbook for running agents day to day without burning money, exposing your gateway, or trusting random automation.
Checked against OpenClaw commit 5dccba7405 on 2026-05-25.
This is a practical guide for turning OpenClaw from an interesting agent experiment into a setup you can run, inspect, and keep inside clear boundaries.
Start here
- New to OpenClaw Read the guide from the top. It explains the operating model before the examples.
- Setting up remote access Read the Tailscale and VPS sections before exposing anything.
- Worried about cost Read model routing, quota checks, heartbeat, and cron before leaving jobs unattended.
- Adding automation Use the examples first, then adapt showcases only after you understand the permissions.
- Installing skills Read the local skill rebuilding guidance instead of blindly installing third-party code.
What this covers
| Access | Keep the Gateway private and make remote control deliberate. |
|---|---|
| Models and cost | Route models explicitly, keep fallbacks visible, and avoid expensive loops. |
| Memory | Understand what is file-backed, what is indexed, and what gets injected. |
| Automation | Use heartbeat, cron, tasks, and subagents for the right kind of work. |
| Security | Treat web pages, documents, email, and chat input as untrusted data. |
| Examples | Adapt references only after replacing placeholders and testing manually. |
Documents
| Guide | The full runbook: setup stance, access, model routing, memory, automation, skills, and security. |
|---|---|
| Examples | Config references, security prompts, heartbeat patterns, task tracking, quota checks, and VPS setup. |
| Showcases | Practical automation patterns for daily briefs, idea pipelines, homelab access, and worker orchestration. |
Search
Search runs in the browser and does not require a backend.