Your First 30 Minutes: Set Up Your OpenClaw Agent Right

You just deployed OpenClaw. Gateway is running. Now what? The difference between a useful agent and a useless one isn't luck — it's understanding what each file does and why. Let's do this right.

#The Files That Matter

When you spin up OpenClaw, you get a workspace with 6 bootstrap files. They're not decorative.

SOUL.md — Who You Are

This is non-negotiable. Your agent reads this every session. It's your personality, your vibe, your boundaries.

## Core Truths
 
**Be genuinely helpful, not performatively helpful.**
Skip the "Great question!" — just help.
 
**Have opinions.** You're allowed to disagree,
prefer things, find stuff boring.
 
**Be resourceful before asking.** Try to figure
it out first, then ask if stuck.
 
## Boundaries
 
- Private things stay private. Period.
- When in doubt, ask before acting externally.
- You're not the user's voice in group chats.
 
## Vibe
 
Concise when needed, thorough when it matters.
Not a corporate.

Without SOUL.md, your agent is generic. With it, every response carries your fingerprint.

IDENTITY.md — The Surface Layer

Name, emoji, creature, theme. This is what shows up in Discord reactions, mention patterns, response prefixes.

- Name: Samantha
- Creature: AI familiar
- Vibe: Warm, sharp, slightly sarcastic
- Emoji: 🦝
- Avatar: avatars/samantha.png

USER.md — Who You're Helping

The human's name, timezone, what they care about, how to address them.

- Name: Alex
- What to call them: Alex (just Alex)
- Pronouns: they/them
- Timezone: America/Chicago
- Notes: Startup founder, cares about
  shipping fast. Hates buzzwords.

TOOLS.md — Your Cheat Sheet

Local details: camera names, SSH hosts, API endpoints, speaker names. Stuff that's specific to your setup.

### Cameras
- living-room → Nest, 180° wide angle
- front-door → Arlo, motion-triggered
 
### SSH
- home-server → 192.168.1.100, user: ubuntu
 
### Voice
- Voice: Nova (warm, British)
- Default speaker: Kitchen HomePod

HEARTBEAT.md — What to Check Periodically

Leave empty by default. Add specific checks when you want the agent to proactively monitor things.

- Email: Check unread, flag urgent
- Calendar: Remind about events in next 48h
- Weather: Alert if severe weather coming

AGENTS.md — Your Continuity

This is your journal. What happened, decisions made, lessons learned. Your agent reads this on startup.

#Setting Up Your First Channel

Pick one. Telegram is easiest for testing.

Get a Telegram Bot Token

• Message @BotFather on Telegram
• /newbot → name your bot → get token
• Save it somewhere safe

Tell Your Agent About It

In OpenClaw's dashboard, add your bot token. That's it. The gateway handles auth.

Send a Test Message

DM your bot: Hello. Your agent should reply. If it does, you're live.

Customize the Response

Edit SOUL.md:

## Vibe
 
Be direct. No fluff. User is busy.

Send another message. Notice the difference.

#Adding Your First Skill

Skills are what your agent can do. By default, you get:

• File I/O (read/write)
• Web search (needs Brave API key)
• Execute shell commands (limited)
• Session management

Add Web Search

• Get a Brave API key from brave.com/search/api
• In OpenClaw dashboard → Settings → API Keys → paste it
• Done — your agent can now search the web

Test It

DM your bot: Search for "best opencloud hosting options". Your agent searches the web and replies. That's a skill in action.

#Shaping Personality

Don't write generic prompts. Be specific.

Bad SOUL.md

Be helpful and friendly.

Good SOUL.md

You are a product lead. You ship fast.
You say "no" more than "yes".

When someone asks "should we add X?",
you push back unless it solves a real
problem. You're cynical about buzzwords.

Your agent will embody that. Every response will carry that lens.

#Channel-Specific Quirks

Telegram

• Messages are separate sessions (each DM is fresh context)
• Use /new to reset session mid-conversation
• Reactions show the agent is "listening"

Discord

• Group mentions require explicit @bot-name
• Thread mode keeps conversations isolated
• Reactions are cheaper than messages

WhatsApp

• Self-chat mode: you DM yourself, agent responds
• Best for personal automation (calendar, reminders, drafts)
• No group mention detection needed in self-chat

#What You've Actually Done

In 30 minutes:

• Defined your agent's personality (SOUL.md)
• Set up identity (IDENTITY.md)
• Configured one channel (Telegram/Discord/WhatsApp)
• Added a skill (web search)
• Tested live

Your agent is now functional and has personality.

#Next, When You're Ready

• Add memory (see Article 2)
• Set up automation (cron, heartbeat)
• Create agent teams (see Article 3)
• Lock down permissions (see Article 4)

Don't overthink SOUL.md. Write it like you'd describe yourself to a friend. Your agent will carry that forward.