Log in

OpenClaw vs Hermes Agent: Which One Fits Your Business, and How Personas Work in Each

Updated: November 12, 2026 · By Brian Arrington (founder, BotPlaybooks)*

The short answer

OpenClaw is best for people who want one self-hosted gateway across their chat apps, running several agents that each have their own workspace and SOUL.md.

Hermes Agent is best for people who want an agent that builds and improves its own skills, with a SOUL.md per profile plus quick /personality switches.

Both are free and open source, both need you to bring a model, and both can run BotPlaybooks personas.

I am not going to name a winner. These are two good tools built around different ideas, and the right pick depends on how you want to work. This comparison is written from the persona side: what changes for your SOUL.md and your agent roles when you choose one or the other. Every fact comes from each project's own docs, linked in the Sources list at the end.

OpenClaw vs Hermes at a glance

OpenClaw Hermes Agent
Maker OpenClaw Foundation, "an independent 501(c)(3)" [1] Nous Research [2]
License and cost MIT licensed, "No paid tier" [1] MIT licensed; "free and open-source" [3][4]
Model Bring your own: needs "an API key from your chosen provider," or a local model [1][5] Bring your own: pay "only for the LLM API usage from your chosen provider," or run local models [4]
Core idea "Self-hosted gateway" connecting chat apps to AI agents [1] "Self-improving AI agent" with a "built-in learning loop" that creates and improves skills [2]
Where SOUL.md lives Per-agent workspace, default ~/.openclaw/workspace [6] ~/.hermes/SOUL.md or $HERMES_HOME, never the working directory; each profile has its own [7][8]
How SOUL.md is used Injected into the system prompt's Project Context on the first turn of a new session [9] Slot #1 of the system prompt, replacing the default identity [10]
Size limit 20,000 characters per file and 60,000 total by default, configurable [6] Set by context_file_max_chars. Unset: scales with the model's context window, 20,000 to 500,000 characters. Set: your fixed cap [11][12]
SOUL.md injection scan Not found in the OpenClaw docs we checked Scanned; a hit warns but the file still loads [10]
Temporary persona switch No equivalent found in the docs /personality overlays, with built-in presets and custom ones in config.yaml [7]
Persona-related files AGENTS.md, SOUL.md, IDENTITY.md, USER.md, BOOTSTRAP.md, MEMORY.md [9] SOUL.md, AGENTS.md and context files; MEMORY.md and USER.md for memory [7][13]
Running several roles One workspace per configured agent [9] Profiles: independent agents on one machine [8]
Skills and plugins ClawHub plugin marketplace [1] Skills Hub, agentskills.io compatible [2]

Who makes them and what they cost

OpenClaw

OpenClaw is run by the OpenClaw Foundation, which its docs describe as "an independent 501(c)(3)." It is MIT licensed, and the docs say plainly: "No paid tier." The software is free, but it does not ship with a model. The docs say you need "an API key from your chosen provider," or OpenClaw can install and manage a local model or connect to a model server you already run. One caution from the same docs: local models do not come with the safety filters hosted providers apply.

Hermes Agent

Hermes Agent is made by Nous Research. It is MIT licensed, and its FAQ says it is "free and open-source." On cost, the FAQ is just as clear: "You pay only for the LLM API usage from your chosen provider. Local models are completely free to run." It supports local models through Ollama, vLLM, llama.cpp, SGLang or any OpenAI-compatible server.

Nous Research also offers Nous Portal, a subscription gateway its docs call the recommended way to run Hermes, with many models and tools on one bill. It is optional.

The cost that actually matters

With either tool, your running cost is the model you connect. Pick the provider first, look up its pricing, and budget for that. The agent software itself is free.

How each one handles a persona

OpenClaw: one SOUL.md per agent workspace

In OpenClaw, "Each configured agent (see Multi-agent routing for running several) has its own workspace, bootstrap files, and session store." That means each role agent gets its own SOUL.md. Your Chief of Staff and your Inbox Triage Assistant can live side by side, each with its own identity, voice and boundaries.

SOUL.md is one of six bootstrap files. OpenClaw injects them into the system prompt's Project Context on the first turn of a new session. The default budget is 20,000 characters per file and 60,000 total, and both numbers can be changed.

Hermes: SOUL.md first, plus /personality for quick modes

In Hermes Agent, SOUL.md is "the primary identity" and slot #1 of the system prompt. It replaces the built-in default identity. Hermes reads it only from ~/.hermes/SOUL.md (or $HERMES_HOME/SOUL.md), so a SOUL.md in your project folder is ignored. Hermes seeds a starter file if you do not have one and never overwrites yours.

Hermes scans SOUL.md for prompt-injection patterns. A hit shows a warning, but the file still loads. The size cap is set by context_file_max_chars. Left unset, it scales with your model's context window, from 20,000 to 500,000 characters; set a number to pin your own cap. Text over the cap is truncated, so shorter is safer.

For short-term changes, Hermes has /personality, a temporary overlay for the current session. It ships with built-in presets, and you can add your own under agent.personalities in ~/.hermes/config.yaml.

What goes in AGENTS.md

In Hermes, put the persona (including standing approval rules, and any workflow that should follow the agent everywhere) in SOUL.md, and put only project-specific instructions (commands, paths and repo conventions) in AGENTS.md. In OpenClaw, official docs keep voice, stance and style in SOUL.md and operating rules in AGENTS.md; you can paste the whole Persona / System Prompt into SOUL.md, or use that official split. Hermes can also use skills.

Our recommendation: In Hermes, put the persona (including standing approval rules, and any workflow that should follow the agent everywhere) in SOUL.md, and put only project-specific instructions (commands, paths and repo conventions) in AGENTS.md. In OpenClaw, official docs keep voice, stance and style in SOUL.md and operating rules in AGENTS.md; you can paste the whole Persona / System Prompt into SOUL.md, or use that official split.

Running a team of business roles

Several roles in OpenClaw

Configure one agent per role. Each gets its own workspace with its own SOUL.md and AGENTS.md, plus its own session store. Multi-agent routing in the OpenClaw docs covers running several at once.

Several roles in Hermes

Use Profiles. The docs describe them as a way to "Run multiple independent Hermes agents on the same machine, each with its own config, API keys, memory, sessions, skills, and gateway state." Each profile has its own SOUL.md, and creating a profile named coder gives you a coder command. The FAQ says "There is no hard limit" on profiles. Two cautions from the docs: never point two agent processes at the same profile, and each profile needs its own WhatsApp number if you use WhatsApp.

Inside one profile, /personality covers quick, temporary switches. For a lasting role, give it its own profile.

Chat apps

Both projects publish official channel lists. Both include Telegram, Discord, Slack, WhatsApp, Signal and Microsoft Teams, among many others. Check the OpenClaw channels page and the Hermes messaging page for the full, current lists.

Where Grok Bot fits

Grok Bot is a different kind of tool. It is not self-hosted. You run many named Bots on a persistent cloud computer, and there is no SOUL.md. In your main Grok Bot, use the plus button to give it the persona's Markdown file and tell it to build the persona as a new agent. Task instructions go in your messages. Grok Bot runs in desktop apps (macOS, Windows, Linux) and mobile apps (iOS, Android).

If you would rather not manage a server, Grok Bot is worth a look. To use a BotPlaybooks persona there, use the plus button in your main Grok Bot to give it the persona's Markdown file. Tell it to take that file and build the persona as a new agent. A Chief of Staff or CEO bot helps. If you have one, tell it that it's in charge of the new persona and that the new persona reports to it. Then send the Example First Task.

Installing a BotPlaybooks persona in either one

Each BotPlaybooks persona is one Markdown file with a system prompt (role, inputs, workflow, output format, guardrails), tools and connectors, routines and an example first task. It is not named SOUL.md out of the box. The steps below differ by tool: for Hermes, the whole persona stays in SOUL.md; for OpenClaw, you can paste the whole Persona / System Prompt into SOUL.md, or use the official split.

In OpenClaw:

  1. Open the workspace for the agent that will do the job (default ~/.openclaw/workspace).
  2. Back up the existing SOUL.md and AGENTS.md.
  3. You can paste the whole Persona / System Prompt into SOUL.md, or split it: voice and boundaries in SOUL.md, workflow and output format in AGENTS.md. If you use that official split, put the full guardrails in AGENTS.md. In most persona files, that means moving the GUARDRAILS block out of the Persona / System Prompt section.
  4. Paste the workflow, routines and tool notes into AGENTS.md.
  5. Start a new session, so the bootstrap files load, and run the example first task.

In Hermes Agent:

  1. Pick or create the profile for the role.
  2. Back up its SOUL.md.
  3. Paste the whole persona, including standing approval rules and any workflow that should follow the agent everywhere, into SOUL.md. Keep the GUARDRAILS block in the Persona / System Prompt section.
  4. Put only project-specific instructions (commands, paths and repo conventions) in AGENTS.md or a skill.
  5. Run the example first task, then check /context for any scan warning.

Good roles to start with: Chief: Chief of Staff, Docket: Inbox Triage Assistant and Scout: SDR Prospect Researcher. For 10 ready-made SOUL.md excerpts, see SOUL.md Templates for Business Agents. Our What Is SOUL.md? explainer covers it in plain English.

FAQ

What is the difference between OpenClaw and Hermes Agent?

OpenClaw is a self-hosted gateway that connects your chat apps to AI agents, and each configured agent gets its own workspace and SOUL.md. Hermes Agent is a self-improving agent with a built-in learning loop that creates and improves its own skills. Both are free, MIT licensed and open source, and both need a model you connect.

Is OpenClaw free?

Yes. OpenClaw is MIT licensed, run by the OpenClaw Foundation, and its docs say "No paid tier." The software costs nothing, but it does not include a model. You need an API key from your chosen provider, whose usage you pay for, or you can run a local model that OpenClaw installs or connects to.

Who makes Hermes Agent?

Hermes Agent is made by Nous Research. It is MIT licensed, and its FAQ calls it free and open source. You pay only for LLM API usage from the provider you choose, and local models are free to run. Nous Research also offers Nous Portal, an optional subscription gateway for models and tools.

Where does SOUL.md go in OpenClaw vs Hermes?

In OpenClaw, SOUL.md goes in each agent's workspace, which is ~/.openclaw/workspace by default. In Hermes Agent, it goes in ~/.hermes/SOUL.md or $HERMES_HOME/SOUL.md, and Hermes ignores a SOUL.md in your working directory. Each Hermes profile has its own SOUL.md, so separate roles can each keep their own identity file.

Can I run several personas at once?

Yes, in both. OpenClaw gives each configured agent its own workspace, bootstrap files and session store, so each role has its own SOUL.md. Hermes Agent uses Profiles: independent agents on one machine, each with its own config, memory, skills and SOUL.md. Hermes also has /personality for temporary switches inside a single session.

Which is better for a small business?

It depends on how you work. OpenClaw fits if you want one gateway across several chat apps and a separate agent for each role. Hermes Agent fits if you want an agent that grows its own skills, with profiles for separate roles. Both are free; your real cost is the model you connect.

Do BotPlaybooks personas work in both?

Yes. Each persona is one plain Markdown file. In Hermes, put the persona (including standing approval rules, and any workflow that should follow the agent everywhere) in SOUL.md, and put only project-specific instructions (commands, paths and repo conventions) in AGENTS.md. In OpenClaw, official docs keep voice, stance and style in SOUL.md and operating rules in AGENTS.md; you can paste the whole Persona / System Prompt into SOUL.md, or use that official split. The file location is the agent's workspace in OpenClaw, or the Hermes profile.

Run either one with a ready-made role

Every persona drafts. You approve anything sent, posted or paid.

Related guides: SOUL.md Templates for Business Agents

Sources

Official docs only, checked Oct 16, 2026.

  1. OpenClaw docs home: https://docs.openclaw.ai/
  2. Hermes Agent docs home: https://hermes-agent.nousresearch.com/docs/
  3. Hermes Agent license: https://github.com/NousResearch/hermes-agent (LICENSE)
  4. Hermes Agent FAQ: https://hermes-agent.nousresearch.com/docs/reference/faq
  5. OpenClaw local models: https://docs.openclaw.ai/gateway/local-models
  6. OpenClaw, where things live on disk: https://docs.openclaw.ai/help/faq/where-things-live-on-disk
  7. Hermes personality: https://hermes-agent.nousresearch.com/docs/user-guide/features/personality
  8. Hermes profiles: https://hermes-agent.nousresearch.com/docs/user-guide/profiles
  9. OpenClaw agent concepts: https://docs.openclaw.ai/concepts/agent
  10. Use SOUL.md with Hermes: https://hermes-agent.nousresearch.com/docs/guides/use-soul-with-hermes
  11. Hermes context files (size limits): https://hermes-agent.nousresearch.com/docs/user-guide/features/context-files
  12. Hermes configuration (context file truncation): https://hermes-agent.nousresearch.com/docs/user-guide/configuration
  13. Hermes, which file does what: https://hermes-agent.nousresearch.com/docs/user-guide/which-file-does-what
  14. Nous Portal: https://hermes-agent.nousresearch.com/docs/integrations/nous-portal
  15. Channel lists: https://docs.openclaw.ai/channels and https://hermes-agent.nousresearch.com/docs/user-guide/messaging/
  16. Grok Bot: https://docs.x.ai/grok-bot/bots and https://docs.x.ai/grok-bot/overview

AI can make mistakes. Always review outputs before acting. Personas are templates, not legal, financial, tax, medical or HR advice.

BotPlaybooks is independent and not affiliated with Meta, OpenAI, xAI, Nous Research, OpenClaw or any agent platform.

Brian Arrington is an AI consultant and business coach focused on agentic AI. He founded BotPlaybooks and Arrington Consulting LLC. He is an Air Force veteran and the 2025 Georgia Vetrepreneur of the Year.