Keep exploring · more guides
Hermes just got a command that makes your AI agent work while you don't.
You type one line, and it checks on your business every few minutes on its own.
It watches your deploys, your rankings and your long jobs, then tells you the moment something changes.
It runs inside a normal chat, so the updates can land in WhatsApp like a text from an employee.
There is one setting most people will miss, and it decides whether this saves you money or wastes it.
I'll show you exactly which one to use, and when.
Everything on this page comes from Hermes' own docs and source code — plus real sessions I ran against it.
"Each wakeup is a real agent turn: Hermes reads the current state fresh — the latest CI result, the newest queue depth, the file as it is now — does the work, reports back, and goes quiet until the next tick."
— Hermes documentation, Recurring Loops (/loop)
You type a slash command, a time, and a plain-English job. That is the entire setup.
The three parts of the command. Drop the middle one and Hermes picks the rhythm for you — that's the setting most people will miss.
What you're watching: a real Hermes session. One line sets the loop, and /loop status shows the cadence, the safety budget, and the countdown to the next wakeup.
Every wakeup is a real agent turn against the world as it is right now — not a replay of an old answer.
The difference that makes this useful: the agent goes back out and looks, every single time.
What you're watching: the actual wakeup message Hermes injects, word for word from the source code. It orders the agent to re-check everything fresh.
Nothing here is hard work. That's exactly why it's so expensive — it's your attention, not your skill.
You publish something. Then you keep going back to see if it's live.
You start a big job. Then you keep peeking to see if it's done.
You wait on a report. The waiting eats the afternoon.
None of it is difficult.
All of it is you.
That checking is a job — a boring one — and it has been sitting on your plate because there was nobody to hand it to.
The /loop command is who you hand it to.
You type one sentence into a chat window. That's the whole thing.
There's no file to edit, no schedule syntax to learn, and no dashboard to log into.
I run an SEO pipeline across five sites, and most of it is waiting for something to finish or land.
Every gap on that line used to be me, going back to look. Now it's one sentence, and the agent keeps the heartbeat going.
Leave the interval out and the loop paces itself — fast when things move, patient when they don't.
The real backoff, taken from Hermes' own scheduler: 1 → 2 → 4 → 8 → 15 minutes while nothing changes, then straight back to 1 minute the second the answer is different.
What you're watching: Hermes' real pacing engine, driven tick by tick. Five identical replies push it out to the 15-minute ceiling; one changed reply drops it straight back to a minute.
A wakeup is a full agent turn, so a dumb loop hammering away all night is pure waste.
Backoff maths, not a benchmark: a fixed two-minute loop fires 240 times overnight; a self-paced one settles at the 15-minute ceiling and fires around 35.
Not the way I run it. Most everyday work goes to a free local model on my own machine, free API tiers cover more, and the heavy work goes through the CLIs I already pay for — my Claude subscription already includes the Claude CLI, so I'm not paying twice.
Self-pacing exists exactly so the quiet hours stay cheap, and inside the Boardroom there are full token-efficiency tutorials so you stop thinking about it.
My advice: default to self-paced. Only set a fixed time when you're racing an external clock.
This is the part that decides whether you can trust it running while you're not watching.
Five independent brakes. Any one of them ends the loop — which is why you can walk away from it.
Every wakeup teaches the agent to sign off with a private code word when the job is finished or no longer makes sense. It sees the deploy go live, says so, and stops itself.
Add --times 30 and it stops after thirty checks. No surprises, no debate.
Add --until and describe what done looks like. A second model reads each reply and checks it against your condition.
/loop stop ends it. /loop pause parks it so you can pick it up later. Hitting cancel mid-check pauses it instead of killing it.
A hundred wakeups by default, then it pauses itself. Set a loop, forget it exists, walk away — it still can't run forever.
Here's a loop I actually set, watching a deploy until it went live.
What you're watching: a real Hermes session. Two wakeups find the job still building, the deploy finishes, the third wakeup sees it live and the agent ends the loop itself. Replayed at reading pace.
What you're watching: a real run cap firing after exactly three checks, then the pause, resume and stop controls. Real session, replayed at reading pace.
You're not asking the working agent to grade its own homework — a separate judge reads the answer.
The judge is the same one that powers Hermes' goals — and it fails open on purpose, so a broken judge can never freeze a loop forever.
It structurally can't. The backstop budget pauses it after a hundred wakeups even if you forget it exists entirely.
And a real message from you always takes priority — wakeups only fire when the chat is sitting idle.
Hermes is one of four agents wired into the Agent Operating System I give every member — alongside Claude Code, OpenClaw and Free Claude Code, all sharing one memory vault.
Join and you walk out with:
Set up in an afternoon · used in 38 countries · new tools added the week they ship
Hermes already had two ways to make an agent work without you. These three answer three different questions.
Pick the wrong one and you burn wakeups for nothing. Schedule it, chase it, or watch it — that's the whole decision.
Overnight jobs, weekly reports, the standing orders of your business. My competitor-watching agent lives here — it should run whether I'm awake or not.
Fix this until it works. Finish this until it's complete. A judge checks after every turn and keeps it moving.
Polling, monitoring, run-the-tests-and-fix-until-green, tidying up during a long work session.
Claude Code has a /loop command, and Nous says openly that this is their take on it — they even kept /proactive as an alias so Claude Code users feel at home.
The launch post spells out both cadence modes — a fixed five minutes, or leave the interval off and it paces itself.
Notice the last line: the CLI, the desktop, and every messaging channel.
That one line from Teknium is the whole feature: it behaves like a scheduled job, except it lives inside your chat, with your context and everything else you're doing.
That difference — inside the conversation instead of outside it — is exactly why this exists alongside the scheduler.
Live X embeds need a connection to render — offline they show as plain links, which is normal.
A loop you set this morning is still alive this afternoon when you open the chat again.
If the chat gets compressed to save space, the loop migrates across and keeps going. It survives resuming the conversation later too.
No system-prompt surgery, no swapping tools around. That keeps caching intact, which keeps your loops cheap over hundreds of checks.
Set a new one and it replaces the old one. Want several? Run several chats. Want a whole fleet on schedules? That's cron's job.
An active goal owns the session and wakeups wait their turn. Park a goal on something slow and let a loop keep watch on something else.
No — a real message from you always wins. Wakeups only fire when the chat is idle and nothing of yours is waiting.
You're never fighting your own agent for attention.
The terminal, the desktop app, the web dashboard — and every messaging platform you already use.
On messaging platforms the gateway fires the wakeups even between your messages — the loop belongs to the chat itself.
You open the same app you message your family on, type one line about your business, and close your phone.
Updates just arrive. Like a text from an employee who never forgets to report in.
There is nothing to build here. One line of text in a chat window — if you can send a message, you can set a loop.
Plenty of the business owners I work with had never opened a single AI tool before they started, and they run agents like this daily now.
Features only matter when they touch real work, so here's exactly where mine go.
Agents pick up content tasks on a Kanban board. A loop flags when a lane clears or something stalls, so I stop alt-tabbing every ten minutes.
When new pages go out, a self-paced loop watches for them getting picked up — fast at first, then backing off, then snapping back the second something moves.
Big content batches and video renders need a poke when they're done. The old way you babysit it. The new way you set a loop and go for a walk.
"Check this every five minutes and tell me when it changes" is automation. It's also just a sentence.
Wrong: Automation means big scary systems — code, servers, workflows with fifty steps.
Right: The hard part is now knowing what to ask for and which tool to point it at. That's learnable in an afternoon.
Wrong: I should wait until these agent tools settle down before I learn them.
Right: The people figuring this out now will be far ahead when it settles — because every workflow you build compounds.
Wrong: I'd need to be technical to have an agent watching my business.
Right: The gap between "never used AI" and "has an agent watching my business" is now one sentence in a chat window.
Members post their wins every day — agency owners, ecom founders, course creators and solo operators across 38 countries. Real businesses, real numbers, in their own words.
Read the 158-page wins doc →Every tick is a full agent turn, so a loop checking every thirty seconds on something that changes once a day is you paying for the same answer over and over.
My advice: default to self-paced. Leave the interval out unless you have a real reason, like an external clock you're racing.
Six months ago "my agent monitors things for me" meant building something custom. Today it's one line in a chat window.
That pace isn't slowing down, and a loop you set today keeps paying you back every day after.
Schedules, goals and loops. Three questions, three answers — and the whole system hums when you use each for what it's for.
This page shows you what /loop does. The Boardroom saves you the year it took me to build everything around it.
Inside, Hermes is already wired into the Agent Operating System, sharing one memory vault with Claude Code, OpenClaw and Free Claude Code — plus the tools I've added for SEO agents, video and AI avatars. The daily tutorials show which jobs in a real business should become loops, how to write stop conditions that actually stop, and how to run monitoring straight from Telegram or WhatsApp so your agents report to your pocket.
Readers bookmark this page and keep doing the checking by hand. Operators join, wire it in this week, and get that job off their plate for good.
Get the Agent OS → Inside the AI Profit Boardroom · skool.com/ai-profit-lab258 documented member wins · 38 countries · four live calls a week · everything from this guide, pre-wired
An agent that checks, works, reports, and rests.
Go set your first loop.
Point it at the most annoying thing you check by hand.
Then see what it feels like to get that job off your plate.
