I stopped running the company agent gateway myself
Grok Bot vs OpenClaw is a hosting decision: SuperGrok Heavy at $300/mo versus a self-hosted gateway you keep up. For operators who do not want the box.
I used to keep the company agent on a box I owned. First OpenClaw. Then Hermes after the migrate path showed up. Slack plugin, Slack app, tokens, loginctl enable-linger so the process survived logout. That is a real job. Grok Bot vs OpenClaw is not a model bake-off. It is a hosting decision: do you operate the gateway, or do you rent a computer that stays on?
I still hand-deploy when a company wants to own the runtime. I also tell them when they do not want that job.
What OpenClaw and Hermes actually are
OpenClaw is a self-hosted multi-channel gateway from the OpenClaw Foundation, MIT-licensed by Peter Steinberger. You bring the model API key, keep the gateway running, and use the local control UI on port 18789. Slack is a plugin, a Slack app, and tokens you have to keep valid.
That stack is honest. A machine. A user session that stays alive. Credentials. A channel plugin for each place the company talks. The control UI listens on 127.0.0.1:18789. Hermes Agent is Nous Research’s self-hosted agent. The move off OpenClaw is hermes claw migrate. Install is hermes gateway install, then linger so the process does not die when you disconnect.
None of that is a product knock. It is the cost of owning the runtime. If the host reboots and linger is wrong, Slack goes quiet. If a bot token rotates and nobody notices, the plugin is a brick. If the control UI is bound to localhost, someone still has to be on that machine or tunnel to it.
I have run this. It works. It also means the operator is part of the uptime story.
The five-minute setup is the lie
A Slack agent is not installed in five minutes if you own the runtime. You still have a Slack app, bot tokens, a plugin, a gateway process, and a machine that must stay up after you close the laptop.
The demo is always the happy path. Export an API key. Chat in the control UI. Send one message in a test channel. The company path is different. Someone patches the host. Someone notices when Slack expires a token. Someone decides what the agent is allowed to write into a thread or a CRM field.
This is where AI for small business gets practical. The question is not whether a gateway can talk. The question is who gets paged when it stops, and whether the last customer message is still sitting in a queue.
I do not sell five-minute installs. I sell a setup the buyer owns, with a person who can take over. That takes longer than a README.
What Grok Bot changes for a company
Grok Bot is hosted SaaS from SpaceXAI, launched August 11, 2026, still early beta. It is a persistent cloud computer, not a model API. Access is SuperGrok Heavy at $300 a month, Cursor Ultra at $200, or Cursor Premium Teams at $120 per seat.
You need a Cursor account. Enterprise is a waitlist. SuperGrok at $30 is not Bot access. I have watched that mix-up already. Grok 4.6 launched August 12. The API at $2 / $6 per million tokens is the model, not the Bot.
| Setup | Who keeps it up | What you pay |
|---|---|---|
| Grok Bot | SpaceXAI, through Cursor | $120–$300 per month, depending on plan |
| OpenClaw or Hermes | You | Model API key plus a host that stays on |
My take, not a market fact: Grok Bot is easier when a company does not want to operate a gateway. The computer stays on without linger. You are not SSHing in because a reboot dropped the session. That is the product.
I do not claim it replaced OpenClaw or Hermes. I claim it removed the part I was tired of owning when the buyer did not want to own it either.
When self-hosted still wins
Self-hosted still wins when the company wants the runtime, the keys, and the channel wiring under its own roof. A hosted computer is simpler. It is not the same as owning the agent, swapping providers, or inspecting every token the Slack plugin holds.
If legal, IT, or the owner wants Slack tokens on a machine they control, OpenClaw or Hermes is still the shape. You bring the API key. You pick the model. You keep the gateway up. That is ownership, not a bug.
I still hand-deploy when they want that. Telegram is $2k–$4k. Discord is $2k–$5k. The AI Receptionist is $8k. Slack Enterprise is $3k–$6k for companies of 10–250 people. The buyer owns the setup. I do not put a SaaS meter on it.
Grok Bot is the other commercial object. Recurring. Hosted. Fine, if that is the trade they want. I will not pretend a $300 SuperGrok Heavy seat is the same as a runtime they can move.
How this relates to a Slack agent I deploy
A Slack agent I deploy is an owned runtime with escalation, CRM writes, and a person who can take over. Grok Bot can sit beside that for research and drafting. It does not replace a hand-deployed Slack agent unless the company chooses not to own the runtime.
For a company in that 10–250 range, I still start with the Slack Enterprise path when Slack is the operating floor: intake, notes, handoff, and an owner who can pause it. Grok Bot is the easier answer when they do not want a gateway at all. I will say that on a call. I will also say they are renting a computer, not taking ownership of an agent.
If they want a lighter owner console first, the Telegram AI Agent is the other owned shape I actually ship.
If you want the hosting choice mapped to your actual Slack workflow, start with the short audit; I reply with your AI replacement map within 24 hours.
