Skip to content

Clawford

Clawford Guide (v3)

Last updated: 2026-04-15 · Live guide

Guide v3 is the live guide. Every chapter in the table below is drafted against the post-liberation Clawford-native runtime. The OpenClaw-era guide v2 is frozen and preserved as historical record. For the strategic reasoning behind the rewrite, start with Ch 02 — What Isn't Clawford?.

Full table of contents

# Chapter
Overview
01 What is Clawford?
02 What Isn't Clawford?
Setup
03 Before you start
04 VPS setup
05 Dev setup
06 Infra setup
Agents
07 Intro to agents
08 Your first agent
09 Mr Fixit 🦊🔧
10 Lowly Worm — newsfeed 🐛📰
11 Lowly Worm — social 🐛📰
12 Mistress Mouse 🐭📅
13 Sergeant Murphy 🐷🔍
14 Huckle Cat 🐱🤝
15 Hilda Hippo 🦛🛒
Architecture
16 The shared brain
17 Auth architectures
18 The inbox: making agents conversational
19 Security and hardening
Reference
20 Scripts and configs reference
21 Glossary

What's changing from v2

  • OpenClaw is gone. Clawford now runs on a Clawford-native stack — codex for LLM calls (riding ChatGPT Plus), system crontab for scheduling, a real shared library under agents/shared/ for the world-access layer, and a shared brain on git + Dropbox. Ch 02 explains why.
  • Nothing is buried. Every chapter that mentioned OpenClaw in v2 either gets rewritten or migrated with the scar tissue stripped. Where v2 spent paragraphs explaining how to work around OpenClaw's 600-second cron timeout or its exec-approvals allowlist, v3 doesn't have to — those problems are gone.
  • The shared library is real. v2 treated agents/shared/ as a handful of deployment helpers. v3 describes a proper three-tier library for world access (clean APIs / stock Playwright / hardened Camoufox), plus a brain.py module for the shared-brain read/write pattern, plus an llm.py broker for the codex backend.

See also