Tell your agent once, and it keeps checking so you do not have to — that is the whole promise of the Hermes loop feature. The /loop command re-runs a prompt (or a slash command) on a recurring cadence inside your current session, and every 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 right now — does the work, reports back, and goes quiet until the next tick. This guide covers the whole feature from the official docs: the two cadence modes, every stop condition, where it works, and how it plays with /goal and cron.
📺 Watch: Hermes Loop Engineering is INSANE!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
Quick orientation before the detail. Hermes has three ways to make work repeat, and they answer different questions. /goal is judge-driven: keep working until this objective is achieved. Cron is schedule-driven and lives outside any session: run this at 7am every day, forever, even if my laptop restarted. /loop is timer-driven and lives inside the session you are already in: do this again every N minutes — or whenever it makes sense — until something says stop. It is Hermes's take on Claude Code's /loop, right down to the /proactive alias, which works here too. One disambiguation while we are here: this is not the self-teaching behaviour where the agent improves itself over time — that is a different system I covered in the Hermes agent self-improvement loop guide. /loop is about repetition on a clock.
When A Hermes Loop Is The Right Tool
Four situations, straight from how I actually use it. Polling external state is the canonical one: watch the deploy, watch the CI run, watch the queue, and tell me when it changes. Iterate-until-green is the second: run the tests, fix what fails, repeat until they pass. Third is monitoring during a work session — keeping an eye on error rates or a long job's progress in the same conversation while you get on with something else. Fourth is periodic housekeeping: re-run a lint pass or a status summary every so often during a long session.
Just as important is when not to use it. If the work should run unattended — overnight, on a real schedule, surviving restarts of your terminal — that is a cron job, because /loop lives inside a session and cron lives outside all of them. And if the task is a single objective with a clear definition of done, /goal is usually the better fit. Timer for rhythm, judge for outcomes, cron for schedules.
Quick Start
The simplest loop is one line: /loop 5m check the deploy status and tell me if it's live yet. Hermes accepts the loop, fires the first wakeup five minutes later while the session is idle, and repeats every five minutes until a stop condition fires or you stop it. You can loop a slash command just as easily — /loop 10m /recap — which turns any command you already use into a heartbeat.
If you want loops, goals and crons working as one automation system, the AI Profit Boardroom ships the Agent OS with these patterns already built in and daily tutorials on wiring them together. → Get the automation stack
The Two Cadence Modes
Fixed interval means you set the clock. Give an interval — 30s, 5m, 2h, even 1h30m — and the loop fires on that schedule regardless of what it finds. Use it when the thing you are watching changes on its own timeline: /loop 2m poll the build and ping me the moment it finishes.
Self-paced means Hermes sets the clock, and it is smarter than it sounds. Omit the interval entirely and the loop starts at the floor — one minute by default — then, while the agent's replies stop changing, backs off exponentially: two minutes, four, eight, up to a fifteen-minute ceiling. The moment a reply differs from the last one, the cadence snaps straight back to the floor. Change detection is a local digest comparison that ignores timestamps, so idle waits cost nothing extra. The rule of thumb from the docs is the one I teach: fixed interval when an external clock drives the work; self-paced when the work drives the rhythm.
Stop Conditions
A loop ends when any of five things fires. First, the agent decides it is done — the wakeup prompt teaches it to end its reply with LOOP_COMPLETE on its own line when the task is finished or moot. Second, a run cap: add --times 30 and the loop stops after thirty wakeups. Third, an evidence-based condition: add --until queue depth reaches zero and, after each wakeup, the same auxiliary judge that powers /goal checks the reply against your condition — and it fails open, so a broken judge never wedges the loop. Fourth, you: /loop stop ends it, /loop pause keeps it around without firing. Fifth, the backstop budget: a max-ticks limit (100 by default) pauses the loop so an unattended session can never burn tokens forever — set it to zero only if you genuinely want unlimited.
| Command | What it does |
|---|---|
| /loop [interval] your prompt --times N --until condition | Start (or replace) the loop for this session |
| /loop or /loop status | Show cadence, ticks fired, and time to the next wakeup |
| /loop pause | Stop firing without losing the loop |
| /loop resume | Pick it back up |
| /loop stop | End the loop |
| /proactive | Alias for /loop — Claude Code parity |
Where It Works
Everywhere Hermes runs: the CLI, the TUI, the web dashboard chat, the desktop app — the same desktop that got the preview-rail browser update — and every gateway platform: Telegram, Discord, Slack, WhatsApp. The gateway detail is the underrated one. On messaging platforms the wakeups fire even between your messages, because the loop belongs to the chat's session — so a deploy-watcher you set from your phone pings you in Telegram the moment the deploy goes live, as an ordinary reply. That one behaviour turns /loop from a terminal trick into a pocket monitoring system.
📺 Watch: Hermes AI Agents Just Went Portable
Mixing Loops With Goals
Both /loop and /goal inject synthetic turns when the session is idle, so they follow one clean rule: an active goal owns the session. While a /goal is actively driving, loop wakeups defer. The loop takes the idle time back the moment the goal finishes, pauses, or parks itself on a wait barrier. A parked goal plus a loop is a genuinely useful combo — the goal waits on the big asynchronous thing while the loop keeps a heartbeat on something else. And a real message from you always beats both: wakeups only fire while the session is idle and nothing of yours is queued, so the automation never talks over you.
Behaviour Worth Knowing
- A wakeup is a normal user-role turn — no system-prompt mutation, no toolset swap — so prompt caching stays intact and ticks stay cheap.
- Loop state survives /resume and context compression, the same way /goal does. Long sessions keep their rhythm.
- One loop per session: setting a new /loop replaces the old one. Want several? Run several sessions — or use cron for a fleet of schedules.
- Interrupting a wakeup turn pauses the loop rather than killing it — recoverable with /loop resume, so cancel actually means cancel.
- Token cost scales with cadence: every tick is a full agent turn. Match the interval to how often the state actually changes, and prefer self-pacing for idle waits — the same thinking as my token-usage guide, applied to schedules.
📺 Watch: Grok Bot DESTROYS Hermes Agent?
Configuring The Hermes Loop
Four settings live in the Hermes config file under the loops section, and the defaults are sensible: a 30-second floor for fixed intervals; the 100-tick backstop budget (zero for unlimited); a 60-second starting cadence for self-paced loops; and a 900-second self-paced ceiling. One more detail for tinkerers: the --until judge routes through the same auxiliary goal-judge task as /goal, so if you point that judge at a cheaper or different model, your loop conditions inherit the override too. Which model minds the loop is worth a thought — frequent ticks on an expensive brain add up, and picking a cheap brain that is still good enough is exactly the kind of call my Goldie Bench side-by-sides are for.
Loop Vs Goal Vs Cron
| /loop | /goal | Cron | |
|---|---|---|---|
| Trigger | Timer, or self-paced | Judge verdict after each turn | Schedule, outside any session |
| Lives in | Your current session | Your current session | Its own session per run |
| Ends when | Stop condition, caps, or you | Goal achieved, budget, or you | You remove the job |
| Best for | Polling, monitoring, periodic re-runs | One objective, iterated until done | Unattended, long-horizon schedules |
The Loops I Would Actually Run
For a business, the money is in the boring ones. A deploy watcher during every release afternoon. An iterate-until-green loop on the test suite while you write the next feature. A queue monitor during a launch, self-paced so it goes quiet when nothing moves. An hourly status summary during a long build so you stop alt-tabbing to check. Inside my Agent OS these live alongside the cron jobs that handle the genuinely unattended stuff — loops for the session I am in, cron for the schedule that outlives it, and the team-of-agents layer above both. It is one more reason the v0.20 era of Hermes feels less like a chatbot and more like staff.
Hermes Loop FAQ
What does /loop actually do?
It re-runs a prompt or slash command on a cadence inside your current session. Each wakeup is a full agent turn against fresh state, so the reply reflects what is true now, not what was true when you set the loop.
How do I stop a loop?
/loop stop ends it; /loop pause keeps it around to resume later. Loops also stop themselves via LOOP_COMPLETE, a --times cap, an --until condition, or the 100-tick backstop.
Can a loop run overnight while my machine is off?
No — loops live inside a session. For unattended, restart-surviving schedules, use a Hermes cron job instead; that is exactly the split the docs draw.
Does it work on Telegram and WhatsApp?
Yes. On gateway platforms the wakeups fire between your messages and the results arrive as ordinary replies in the chat — monitoring in your pocket.
What stops a runaway loop burning tokens?
The backstop budget: 100 ticks by default, after which the loop pauses. Plus self-pacing backs off exponentially when nothing is changing, and every stop condition above cuts it earlier.
The Verdict
Most people still use AI the way you use a search box: ask, read, leave. /loop is one of those small features that quietly changes the relationship — the agent starts holding a thread of attention for you. Set the deploy watcher, close the laptop lid on the problem, and let the ping find you. Start with one loop this week: pick the thing you check compulsively, hand it to Hermes with a sensible cadence and a stop condition, and notice how much attention you get back.
If you want an agent that watches, iterates and reports while you do the real work, check out the AI Profit Boardroom — Hermes automation tutorials from loops to crons to goals, the full Agent OS, four live coaching calls a week and 3,700+ business owners running these systems. → Put your agent on watch duty
