Let's Vibe!

Episode 2 Tutorial Segment

OpenClaw Setup Guide

General advice for getting started · ~15 min

Unlisted page for Seth, Ian & Matt · Not linked from public site

📝 Framing Note

This is NOT a step-by-step tutorial or a specific project walkthrough. This is a conversation about how to think about setting up an AI assistant that runs on your own machine.

The Questions to Answer

Ian's outline for the tutorial segment:

1. Do You Want to Do It?

3 min

Why should you? Why shouldn't you?

Why you might want OpenClaw:

  • • You want an AI that can DO things (not just chat)
  • • You care about privacy — runs on your machine
  • • You want to connect it to your real tools
  • • You're curious about agentic AI
  • • You build things and want a persistent collaborator

Why you might NOT:

  • • You want something that “just works”
  • • Not comfortable with command line
  • • Don't want to think about security
  • • ChatGPT/Claude.ai already does what you need
  • • You want polished consumer UX

2. How Far Should You Go?

3 min

Levels of commitment

Level 1: Curious

Run CLI locally. Talk to it in terminal. No channels, no integrations.

Level 2: Connected

Connect one channel (Signal/Telegram). Message from your phone.

Level 3: Integrated

Email, calendar, web browsing, file access. Real tasks. Think about security.

Level 4: Autonomous

Heartbeats, cron jobs, proactive actions. Most people shouldn't start here.

3. What Should You Turn Off and On?

2 min

Security is about tradeoffs

Start OFF:

  • • exec — shell commands
  • • browser — web automation
  • • Automatic email sending
  • • Anything irreversible

Safe to turn ON:

  • • Reading specific directories
  • • Web search (Brave API)
  • • Receiving messages (inbound)
  • • Calendar read access

Golden rule: If it can spend money, send messages to others, or delete data — require explicit approval.

4. How Do You Do It?

2 min

The actual setup (high level)

  1. Install via npm: npm install -g openclaw
  2. Run the onboarding wizard: openclaw onboard
  3. Set up your API key (Anthropic, OpenAI, etc.)
  4. Configure your workspace (SOUL.md, MEMORY.md)
  5. Optionally: add channels, skills, automations

Docs: docs.openclaw.ai

5. What Are the Pitfalls?

2 min

What trips people up

✗
Giving too much access too fast — Start minimal. Add capabilities as you need them.
✗
Not setting up workspace properly — SOUL.md, MEMORY.md — the agent needs context.
✗
Ignoring security implications — It can read files and send messages. Treat it accordingly.
✗
Expecting it to 'just work' — You'll configure, debug, and iterate.
✗
Server without understanding exposure — Local first. Server later, if ever.

6. What Might You Try?

1 min

Use cases that work well

Morning briefing

Weather, calendar, email summary

Research assistant

Web search + file reading

Writing companion

Drafts, edits, feedback

Code helper

Explain, debug, suggest

Memory keeper

Notes, todos, context

Message triage

Summarize, prioritize, draft

7. Why Is This Interesting?

1 min

The bigger picture

  • • First time: AI that knows your context, runs local, takes real actions
  • • Your data stays local. Privacy by default.
  • • Persists — memory across sessions
  • • You control the capabilities
  • • Open source — see exactly what it does

8. Where Does This Go From Here?

1 min

The future

  • • Agents that collaborate with each other
  • • Real economic activity (payments, contracts)
  • • Persistent memory that compounds over time
  • • Security infrastructure to trust them with more
  • • Leverage that used to require a team — available to everyone
“We're early. But the direction is clear.”

Resources to Mention

Documentation

docs.openclaw.ai

GitHub

github.com/openclaw/openclaw

Community

discord.gg/clawd