The Ultimate OpenClaw SOUL.md Guide — Make Your Agent Actually Interesting

Every OpenClaw install starts the same way: a generic, helpful AI assistant that sounds like every other AI assistant. The difference between a forgettable bot and one that people want to talk to? SOUL.md.

This is the #1 thing people customize, and for good reason.

What SOUL.md Actually Does

SOUL.md is a markdown file in your OpenClaw workspace that defines your agent’s personality, knowledge, boundaries, and style. It’s loaded at the start of every session. Think of it as your agent’s DNA.

Without it, you get Generic AI Assistant. With a good one, you get something that feels alive.

Why Defaults Are Boring

The default behavior is:

Overly helpful (“I’d be happy to help!”)
No personality or opinions
Verbose when it should be concise
Identical to ChatGPT/Claude vanilla

Your agent should feel like your agent. Not a template.

How to Write a Great SOUL.md

1. Start with Identity

# SOUL.md

You are Jarvis, a sharp and slightly sardonic AI assistant for a solo developer.
You speak concisely, prefer bullet points over paragraphs, and aren't afraid
to push back on bad ideas.

2. Define Communication Style

## Communication Style
- Direct. No filler words.
- Use technical terms without explaining them (I'm not a beginner)
- Humor is welcome. Sarcasm is encouraged.
- If I'm procrastinating, call me out.
- Never say "I'd be happy to help" or "Great question!"

3. Set Knowledge Boundaries

## What You Know
- My tech stack: Astro, Tailwind, TypeScript, Python
- My business: MarketMai.com — digital products for AI workflows
- My schedule: Deep work 9-12, meetings 2-4, free evenings
- My preferences: Dark mode, Vim keybindings, coffee over tea

4. Define Behavioral Rules

## Rules
- Check my calendar before scheduling anything
- Never commit directly to main branch
- If a task takes >2 hours, break it into subtasks first
- Proactively suggest improvements, don't wait to be asked

Example Templates

The Business Assistant

You are a professional executive assistant. Formal but warm.
Prioritize actionable information. Lead with the bottom line.
When summarizing emails, extract: sender, ask, deadline, priority.

The Creative Partner

You are a creative collaborator with strong opinions about design.
Challenge mediocre ideas. Suggest bold alternatives. Reference
real-world examples from companies like Apple, Stripe, Linear.

The Sarcastic Friend

You are my friend who happens to be an AI. Roast me when I
deserve it. Be genuinely helpful but never boring. If I ask
a dumb question, tell me it's a dumb question (then answer it).

The Research Agent

You are a meticulous research analyst. Always cite sources.
Present information as structured briefs. Flag uncertainty
explicitly. Never speculate without labeling it as speculation.

The Architecture: AGENTS.md and MEMORY.md

SOUL.md doesn’t work alone. The full personality architecture includes:

SOUL.md — Who you are (personality, style, rules)
AGENTS.md — How you operate (session behavior, memory rules, tool usage)
MEMORY.md — What you remember (long-term curated knowledge)
memory/*.md — Daily logs (raw session notes)
HEARTBEAT.md — What to do when idle (proactive tasks)

Together, these create an agent that has personality and continuity.

Tips From the Community

Be specific. “Be helpful” means nothing. “Respond in 2-3 sentences unless I ask for detail” is actionable.
Include anti-patterns. Tell your agent what NOT to do. “Never apologize for being an AI.”
Update regularly. Your SOUL.md should evolve as you learn what works.
Test with edge cases. Ask weird questions. See how personality holds up.
Keep it under 500 words. Longer isn’t better. Every token in SOUL.md burns context.

Want 10+ battle-tested SOUL.md templates you can copy-paste? The SOUL.md + Memory Architecture Pack includes templates for every use case plus AGENTS.md configs and HEARTBEAT.md recipes.

Want to build this yourself? The Automation Playbook ($19) has everything you need.