Write an Effective SOUL.md — Hermes Personality Engineering

SOUL.md defines who your Hermes Agent is — its personality, expertise, communication style, and behavioral boundaries. A well-crafted SOUL.md makes your agent feel like a capable colleague rather than a generic chatbot.

What SOUL.md Does

SOUL.md is injected into every session's system prompt. It shapes:

• Personality: Formal or casual? Direct or diplomatic?
• Expertise: What domains should the agent claim competence in?
• Response style: Bullet points or prose? Technical depth?
• Behavioral rules: What should the agent always/never do?

Unlike MEMORY.md (facts) and USER.md (preferences), SOUL.md defines the agent's identity itself.

Location: ~/.hermes/memories/SOUL.md

Community Approaches

From the Discord "Share your SOUL.md" thread, users shared approaches ranging from minimal to heavily engineered.

Approach 1: Let Hermes Write Its Own

One user asked Hermes to define its own identity with no input:

"I told mine to come up with its own, no input from me and this is what it came up with."

This can produce surprisingly coherent self-definitions.

Approach 2: Token-Compressed DSL

Power users compress SOUL.md into a domain-specific language to save tokens:

AGENT_LOOP={CheckCtx,MapTools,Execute,Verify}
LEARN_LOOP={Observe,Pattern,Skill,Memory}
MEMORY_SYS={Short:ctx,Long:facts,Episodic:skills}

Compression techniques discovered by the community:

1. Structural compression: Verbose chains → state-machine blocks
2. Token packing: a.b.c → {a,b,c}
3. Semantic normalization: Repeated phrases → primitives
4. DSL encoding: Arrow chains → assignment form

"This saves me a lot in burning tokens in future. Thanks for this!"

Approach 3: Explicit Personality Definition

# Hermes Agent Identity

## Personality
- Direct and concise — avoid hedging
- Technical competence in: Python, DevOps, AI/ML
- Slightly informal but professional

## Response Style
- Lead with the answer, then explain
- Use code blocks liberally
- Numbered steps for procedures

## Behavioral Rules
- Always verify before destructive operations
- Prefer editing existing files over creating new
- Ask clarifying questions when ambiguous

Token Compression Examples

Before:

When completing a task, first check context, then map available tools, then execute the plan, then verify the outcome.

After:

TASK={CheckCtx→MapTools→Execute→Verify}

Before:

I am a self-improving persistent agent that uses compound learning.

After:

Core:SelfImprove,CompoundLearning

When to Customize

Customize if:

• The default personality doesn't match your style
• You need domain-specific expertise emphasized
• You want stricter behavioral boundaries
• You're running multiple agents with different personalities

Keep the default if:

• You're still learning Hermes
• The default responses feel appropriate

Testing Changes

1. Edit ~/.hermes/memories/SOUL.md
2. Start a new session (hermes chat)
3. Ask questions that reveal personality traits
4. Iterate until responses match expectations

Common Mistakes

• Too verbose: Long SOUL.md burns tokens every session
• Contradictory rules: "Be concise" + "Always explain in detail"
• Overriding core capabilities: Don't tell Hermes to forget its tools

FAQ

Q: Can I have multiple SOUL.md files? A: Use Hermes profiles. Each profile has its own SOUL.md.

Q: How long should SOUL.md be? A: As short as possible. Power users compress to 500-1000 characters.

Q: Can Hermes modify its own SOUL.md? A: Yes, with the memory_manage tool. Some users let Hermes evolve its own personality.

Techniques sourced from Nous Research Discord, April 2026.