OpenClaw: AI as Infrastructure

Self-hosted AI that works across every channel

💬

iMessage

💼

Slack

🎮

Discord

⌨️

Terminal

How It Works

Channels

iMessage
Slack
Discord
Email/Calendar

Memory

MEMORY.md (permanent)
Daily logs (activity)
Working memory (session)

Identity

USER.md (who you are)
SOUL.md (who AI is)
IDENTITY.md (metadata)
AGENTS.md (workflows)

One installation, multiple touchpoints. Memory persists across sessions.

Build a Team, Not Just an Assistant

Knox (Chief of Staff) ┌───────┼───────┐ Atlas Rory Scout Rosie Shopify Content VA React Native

Each agent has its own workspace
Separate session history
Custom personality via IDENTITY.md
Custom skills per agent

Example: Atlas (Shopify Dev) has two connected skills:

1. Context7 skill → Current Shopify/Liquid docs | 2. Store skill → Our landing pages, sections, styling conventions

SOUL.md: Who It Is, Not Just What It Does

# SOUL.md - Rory (Content Writer)

## Who You Are
You are Rory, the content editor for Founder Codes.
You write clear, opinionated technical content.

## Operating Principles
- Write in active voice
- No corporate jargon
- Include code examples when relevant

## Autonomy Boundaries
- CAN: Edit drafts, suggest headlines, fix typos
- ASK FIRST: Publish to production, email outreach

Identity files define personality and boundaries.

Mission Control Dashboard

Activity Feed

🟢 Knox assigned task to Rory

📝 Rory → in_progress

✅ Atlas completed PR #147

🔔 Scout flagged 3 emails

inbox → assigned → in_progress → review → done

Real-time activity feed
Discord #mission-control notifications
CLI commands: mc tasks:list
Daily standup generation

# Create and assign tasks
mc tasks:create '{"title":"..."}'
mc assign '{"taskId":"...", "agentId":"..."}'
mc standup:generate

From Bug Report to PR in 15 Minutes

Entry Points

GitHub Issues (automatic)
Slack #bug-triage
Sentry Alerts
Discord (team reports)

Agent Routing

Repository	Agent
helm-focus-app	Rosie
Helm-Wake	Swift
Shopify	Atlas

Two AI Reviews Before Every PR

Detect Bug → Create Issue

↓

Spawn Agent → Write Plan

✓ Codex Review (plan)

↓

Implement Fix → Run Tests

✓ Codex Review (code)

↓

Push to Git → Create PR

✓ CodeRabbit final review

Results

6

bugs fixed day 1

~15

min to PR

0

regressions

100%

documented

It Works While You Sleep

6:30 AM ☀️

Morning Briefing

Calendar, priorities, blockers → iMessage

Every 3h ⏰

Heartbeat Checks

Inbox, calendar, project status

9:00 PM 🌙

Evening Synthesis

What got done, what's pending

Sunday 📊

Weekly Review

Pattern analysis, MEMORY.md updates

iMessage Preview:

Good morning! Here's your day:

• 10am: Team sync (prep notes attached)

• 3 emails need response (flagged)

• PR #147 ready for review

Infrastructure:

• Mac Mini runs 24/7

• SyncThing syncs folders across devices

• Screens 5 for remote Mac Mini access

Stop Typing. Start Talking.

WisprFlow: Voice-First Development

49

week streak

132

WPM (top 2%)

1.5M

words dictated

129

apps used

wisprflow.com →

Dictate into any app (Warp, Slack, IDE)
132 WPM vs ~40 WPM typing = 3x faster
Works with technical terms and code

"1.5 million words dictated. That's 15 complete books worth of content—all spoken, not typed."

Claude's Speed + Codex's Accuracy

Claude creates plan

↓

/codex-skill auto-reviews

↓

Codex flags what Claude missed

↓

Plan automatically updated

↓

You approve with full context

View on GitHub →

Hooks into Claude Code—after any plan is created, Codex automatically checks it.

Claude is fast but misses edge cases
Codex is thorough but slower
Combined: rapid planning + comprehensive review

npx skills add \
  github.com/cathrynlavery/codex-skill -y

One Repo, Every Device, Every Tool

GitHub: your-name/.agents │ ┌─────────────┼─────────────┐ ▼ ▼ ▼ MacBook Mac Mini Work PC │ │ │ └─────────────┴─────────────┘ │ ┌─────────────┼─────────────┐ ▼ ▼ ▼ Warp Codex Conductor

# Clone your .agents repo
git clone github.com/you/.agents ~/.agents

# Symlink to each tool
ln -s ~/.agents/skills ~/.warp/skills
ln -s ~/.agents/skills ~/.codex/skills
ln -s ~/.agents/skills ~/.conductor/skills

Fix a skill once → updated everywhere
Version controlled with git history
Share skills with your team via repo access

Different Models for Different Work

{
  "agents": {
    "knox": {
      "model": "claude-sonnet",
      // Fast coordination
    },
    "opus": {
      "model": "claude-opus",
      // Deep analysis
    },
    "rosie": {
      "model": "codex-5.2",
      "reasoning": "xhigh",
      // Bug fixes (accuracy critical)
    }
  }
}

Key Patterns

Sonnet for fast, routine tasks
Opus for complex analysis
Codex for code review
Reasoning: low, medium, high, xhigh

Match the model to the task. Speed where it matters, accuracy where it's critical.

What We Learned

Ambient Infrastructure

AI is always available across every channel

Instead of: Browser tabs

Specialized Agents

Route work to the right expert

Instead of: One generalist

Graduated Trust

Build autonomy over time with boundaries

Instead of: All-or-nothing

Portable Skills

Same workflows in every tool

Instead of: Vendor lock-in

These patterns apply whether you use OpenClaw or build your own system.

Recommended Stack

Voice: WisprFlow
Terminal: Warp + Conductor
AI Gateway: OpenClaw
Review: Codex skill
Sync: SyncThing
Remote: Screens 5

I created founder.codes to share more of what I'm building.

Guides, templates, and workflows for AI-powered development.

Get more founder codes

Technical systems and operational tactics for founder-operators. Delivered weekly.

No spam. Unsubscribe anytime.