Hermes vs OpenClaw for always-on customer replies

What they have in common

• Both run on a computer or server you control, and keep running when you are not at the screen.
• Both can answer customers on their own, which chat apps such as ChatGPT, Claude and Grok do not do.
• Both connect to Spaces the same way under the hood: they add the Spaces server and sign in to Spaces in the browser.
• Both read the same Spaces business brief, and are held to the same safe limits on each account.
• Both only answer while the machine is on and online.

Customers cannot tell which one you use. They see normal messages from your own account, paced inside Spaces' limits. The choice is about you: how you like to set things up, and how you want to talk to your agent.

OpenClaw in a nutshell

A self-hosted agent gateway for chat apps. You run it yourself, and you talk to your agent through the chat apps you route to it. You add Spaces once on the gateway, so every channel OpenClaw routes gets the Spaces tools.

That makes OpenClaw a good fit if you like the idea of messaging your own agent from your phone, for example "what came in tonight?", and having it answer your customers through Spaces in the meantime. Keep your own chat with OpenClaw on a different number or app from the business account, so your instructions never land in a customer chat.

Hermes in a nutshell

Nous Research’s open-source Hermes Agent. You add Spaces in one settings file and sign in once. Its Spaces tools show up with names starting mcp_spaces_, which is handy when you want to check what it can do.

Hermes suits owners, or the person who helps them with tech, who are comfortable editing a settings file and running a command or two. Our setup page also shows how to use a Spaces key instead of signing in, which some people prefer for an agent on a server.

Side by side

Both projects change quickly. Their own docs, linked from our setup pages, have the current details of what each supports.

Setting up OpenClaw with Spaces

1. Add Spaces once on the gateway, so every channel it routes gets the tools.
2. Run openclaw mcp login spaces, open the address it prints, sign in to Spaces, then run the command again with the code it gives you.

openclaw mcp add spaces --url https://api.spacesos.com/mcp --transport streamable-http
openclaw mcp login spaces

To check it worked: Run openclaw mcp probe spaces. It lists the Spaces tools.

You sign in on Spaces' own page, so OpenClaw never sees your WhatsApp, Instagram or other logins. Connect your chat accounts to Spaces separately, from the dashboard or by asking the agent for a connect link.

Setting up Hermes with Spaces

1. Add the block below to ~/.hermes/config.yaml.
2. Run hermes mcp login spaces and sign in to Spaces in the browser.

mcp_servers:
  spaces:
    url: "https://api.spacesos.com/mcp"
    auth: oauth

To check it worked: Run /reload-mcp, then hermes mcp test spaces. The tools appear as mcp_spaces_….

Prefer an API key? Use a header instead of auth: oauth.

mcp_servers:
  spaces:
    url: "https://api.spacesos.com/mcp"
    headers:
      Authorization: "Bearer spc_live_…"

If you use a key, keep it private. Anyone with it can act like your agent, so replace it in Spaces if it ever leaks.

Which suits which business

The last row is worth taking seriously. An always-on agent is a small piece of software you look after: updates, a machine that stays on, and the odd restart. If that sounds like a chore, a chat app in a few short sessions a day plus a Spaces routine for after-hours holding replies covers most small businesses well.

What stays the same whichever you pick

The safety nets belong to Spaces and your connected accounts, not to the agent, so they are identical. On WhatsApp, for example, every send is at least 10 seconds apart, with up to 400 messages and 10 new chats a day once the number has warmed up. For the first 2 days after you connect an account, Spaces only sends replies to people who wrote in, and the account then warms up over four weeks.

• Pause everything in Settings stops every send at once, from either agent.
• Ask the agent to hold replies as drafts, and nothing goes out until you approve it.
• Hand a chat to a person, and no agent replies there until it is handed back.
• Your business brief in Spaces is read by both, so switching later keeps your facts.
• A Spaces routine runs on Spaces itself, so a holding reply still goes out if your machine is off.

How an always-on agent should check for messages

Whichever you choose, tell it to wait for new messages rather than checking the inbox over and over. Spaces has a tool for this, wait_for_events, which returns the moment something new arrives and never touches the platform. Listing chats again and again to look for news looks like a bot to the platforms and can get an account restricted. The setup brief on our agent pages already tells your agent this, but it is worth checking your agent's instructions say it too.

Rules for either agent

An always-on agent needs clearer rules than a chat app, because nobody reads its replies as they go out. Here is a starting set to add to its instructions, alongside your business brief:

Read the Spaces business brief before every reply, and follow it.
Wait for new messages with wait_for_events. Never check the inbox over and over.
Answer on your own only: hours, prices in the brief, directions, simple yes/no questions.
Hand the chat to a person for: complaints, refunds, bookings you cannot see, anything not in the brief.
Leave any chat that has been handed to a person or has an owner.
Never start new conversations unless I ask.
If Spaces says to wait, wait. Never retry a refused send.

Keep the machine running

Both agents stop answering when the machine sleeps, loses power or drops off the internet. A laptop that closes at night is the most common reason an "always-on" agent is not. A small computer that stays on, or a server, is the reliable choice for a business account.

Set a Spaces routine as a fallback, such as a holding reply when nobody has answered after a few hours. A routine is not an AI; it sends text you wrote in advance, and it runs even when your agent is down.

Common mistakes

• Picking the agent first and the rules later. The rules matter more than the agent.
• Running both at once with no split, so customers get two answers.
• Talking to the agent on the business number.
• Letting the agent check the inbox in a tight loop instead of waiting for new messages.
• Assuming the agent is answering when the machine has gone to sleep.

Summing up

Hermes and OpenClaw both give you an AI agent that answers customers around the clock through Spaces, with the same limits, pause, drafts and hand-overs underneath. OpenClaw suits owners who want to message their agent from chat apps they already use; Hermes suits those comfortable with a settings file and a terminal. If neither appeals, a chat app plus a routine is a perfectly good way to start.

Pricing

Plans from $19 a month

Which is better for customer replies, Hermes or OpenClaw?

Neither is better for customers, who see the same normal messages either way. Pick the one that fits how you like to set things up and talk to your agent.

Do I need to be technical?

A little. Both are set up with commands or a settings file and run on a machine you look after. If that is not for you, use a chat app such as Claude or ChatGPT and a Spaces routine for after hours.

Can I run Hermes and OpenClaw together?

Yes, as two agents on the same accounts, each with its own permissions and daily share in Your agent. Give each a clear job, such as one answering and one only reading.

Do the limits differ between them?

No. The limits belong to each connected account in Spaces, so they are the same whichever agent sends.

Does either one see my WhatsApp or Instagram password?

No. You connect those accounts to Spaces yourself, and the agent signs in to Spaces. It never gets a platform login.

What happens if my machine is off?

Customers wait until it is back. A Spaces routine can still send a set holding reply, because it runs on Spaces rather than on your machine.

Can I switch from one to the other later?

Yes. Your accounts, business brief and routines live in Spaces, so you add the new agent, check it reads the brief, then remove the old one.

What about Claude Code or Codex?

They can run continuously too, on a computer that stays on. They are coding agents in the terminal, so they suit people who already use them for other work.