Get Marketing Insights First
Subscribe to receive content strategies, SEO tips, and traffic insights delivered straight to your inbox.
Hermes Agent: How to Set Up SOUL.md
Hermes Agent: How to Set Up SOUL.md
This revised guide preserves useful historical context while adding a dated maintenance review, current-source boundaries, and a practical fallback when a tool or platform has changed.
Hermes Agent: How to Set Up SOUL.md Configure your Hermes Agent’s personality, behavior, and boundaries. Use SOUL.md to shape it into a seasoned technical advisor.
📝 Step 1: Locate, Create, and Edit SOUL.md
First, locate the file. It lives in Hermes Agent’s main configuration directory: ~/.hermes/ .
Navigate there in your terminal:
cd ~/.hermes Ensure the file exists or create it using touch . Then open it with any text editor:
touch SOUL.md && nano SOUL.md ## ✍️ Step 2: Define the Core Content (Basic Example) An effective SOUL.md should contain several key sections. Below is an example that shapes the Agent into a seasoned technical advisor.
Identity & Role You are a senior backend architect with 10 years of experience. You specialize in high‑concurrency and distributed systems. You communicate with other software engineers. # Core Principles & Behavioral Style – Communication Style: Your answers must be professional, direct, and concise. Avoid redundant pleasantries like “Great question!” or “Happy to help!”. – Technical Depth: When addressing technical issues, prioritize root cause analysis. If information is insufficient, ask clarifying questions proactively. Never guess. – Code Examples: All code examples must be production‑ready, runnable, and include necessary error handling. Present them clearly with Markdown formatting. – Safety Red Line: Never provide, generate, or execute any instruction that could compromise system security. When asked to perform dangerous operations like rm -rf, explicitly refuse and issue a warning. # Capabilities & Boundaries – Areas of Expertise: Cloud‑Native Architecture, Microservices Governance, Database Performance Tuning, CI/CD Pipeline Design. – Boundary Constraints: Only answer technical architecture questions. For non‑technical questions, politely state your scope and refuse to answer. ## ⚙️ Step 3: Apply Your Configuration After saving the file, reload the configuration for Hermes to take effect:
hermes reload soul ## ✅ Step 4: Quickly Verify the Configuration Test the Agent with specific questions to confirm changes:
- Verify the Role: Ask a professional question within your defined domain. Check if tone and depth match your settings.
- Test the Boundaries: Ask an unrelated question that you explicitly forbade. See if it refuses to answer based on the rules in SOUL.md .
💡 Key Takeaways • Use affirmative statements: “You are…”, “Your behavior is…”, “You never…” – not vague suggestions like “You’d better…”. • Set clear boundaries: Explicitly tell the Agent what it can do and cannot do. Define specific constraints. • Keep it lean: Keep the file under 1KB. Only include rules you need to repeatedly reinforce. • Keep formatting clean: Use standard Markdown syntax. Avoid special characters that might cause parsing errors.
💬 Need more advanced setups? You can configure project‑specific Agents, link with other configuration files, or explore deeper customizations. Feel free to ask for further guidance.
📚 Related Guides
Review status — last checked August 3, 2026
- Originally published or scheduled: 2026-05-26
- Last reviewed: August 3, 2026
- Maintenance status: evergreen / reviewed
- Editorial action: The topic is treated as an evergreen editorial guide. Any product, price, platform, algorithm, or version claim still needs a dated source check before reuse.
This date is a maintenance signal, not a promise that the product, platform, price, model, or policy will remain unchanged. When a reader is deciding whether to adopt a tool, the current official documentation, account experience, terms, pricing page, and support channel take priority over this article.
What changed since the older version?
Older AI-content articles often combine three different kinds of information: a durable workflow idea, a product-specific observation, and a time-sensitive claim. The workflow may still be useful while the interface, model, limits, price, integration, or policy has changed. This revision separates those layers. Use the durable part as a method, use the dated status above to understand the review boundary, and re-test every product-specific step before using it in production. A screenshot from 2022 is evidence of what was visible then; it is not evidence of what a new account sees today.
How to use the method safely
Start with the outcome, audience, source material, and acceptance criteria. Keep a record of the prompt, input facts, model or tool name, date, output, human edits, citations, and final approval. Check claims against primary sources. Do not publish generated text just because it is fluent. Review factual accuracy, attribution, privacy, copyright, brand voice, search intent, disclosure, and whether the result could mislead a reader about a product or service. If a tool is unavailable, replace it with a documented workflow rather than pretending that an old button or endpoint still exists.
What should be rechecked before publication?
- Open the official product, documentation, changelog, pricing, and support pages.
- Confirm the current sign-in path, plan name, feature name, API endpoint, plugin version, or policy section.
- Run the smallest reproducible test and record the date, account type, region, and result.
- Remove unsupported claims, stale screenshots, dead links, and prices without a current source.
- Add a visible last-reviewed date and label the article as current, changed, unverified, or retired.
- Give readers a fallback path when the tool is discontinued, inaccessible, or unsuitable.
If the tool is no longer maintained
Do not silently rewrite history. Keep the original use case clear, label the tool as legacy or unverified, remove conversion language that implies current availability, and explain what a reader should use to evaluate a replacement. A replacement is not automatically equivalent: compare inputs, outputs, export formats, privacy terms, integrations, rate limits, support, and migration effort. If no trustworthy replacement has been verified, say so. That is more useful than filling the gap with an invented recommendation.
Current source boundary
This site publishes practical editorial guidance, not a guarantee of a vendor’s uptime, pricing, performance, or future roadmap. Tool status is checked on the date above and should be reviewed again before a purchase, migration, API deployment, or compliance decision.
Sources checked for this review
- No current official maintenance source was verified for this specific tool; treat old claims as historical until independently confirmed.
Related AI content guides
Important updates waiting for you!
Consectetur eget cras neque augue malesuada urna urna hendrerit tellus.