Hermes Agent
Hermes Agent cron jobs: schedule tasks and fix misfires
Updated 17 November 2026
A cron job in Hermes Agent is a task that runs on a timer: every morning, every two hours, once in half an hour. For agent-backed jobs, Hermes starts a fresh agent session for each run and delivers the result to a chat or a file. This page covers how to create one, the schedule formats, where results go, why the prompt has to stand on its own, and what to check when a job does not fire.
Facts come from the Hermes Agent 0.23.1 documentation, namely the scheduled tasks reference, the cron troubleshooting guide and the guides on automating with cron and on script-only jobs, all read on 17 November 2026. Calendar events and scheduled jobs are different things; for calendars see Hermes Agent and Google Calendar.
How scheduled jobs run
The documentation says cron execution is handled by the gateway daemon. The gateway ticks the scheduler every 60 seconds and runs due agent-backed jobs in isolated agent sessions. On each tick it loads the jobs from ~/.hermes/cron/jobs.json, checks which are due, starts a fresh agent session for each due agent-backed job, optionally loads the skills attached to the job, runs the prompt to completion, delivers the final response and records the next run time.
The practical consequence is the gateway requirement. A regular CLI chat session does not fire cron jobs by itself. For jobs to fire automatically you need a running gateway: hermes gateway in the foreground, or the installed service.
hermes gateway install # install as a user service
hermes gateway # or run in the foreground
hermes cron status # is the scheduler alive, and when is the next runOn a Linux server the docs also show sudo hermes gateway install --system for a boot-time system service. The documentation also says the Hermes desktop app runs its own ticker, so scheduled jobs keep firing there without you keeping each profile open. If you are new to the gateway, the Telegram setup guide shows it running.
Two limits are worth knowing now. By default, cron-run sessions cannot create, edit or remove cron jobs. Setting cron.allow_agent_scheduling to true in config.yaml enables those actions. Also, by default a job runs on whichever model the main agent is set to at the moment it fires, unless the job or the cron.model setting names another one.
Create a job
The documentation gives three ways to create one. In a chat, use the slash command:
/cron add "in 30m" "Remind me to check the build"
/cron add "every 2h" "Check server status"
/cron add "every 1h" "Summarize new feed items" --skill blogwatcherFrom the standalone command line:
hermes cron create "every 2h" "Check server status"
hermes cron create "0 9 * * 1" \
"Generate a weekly report covering the top AI news, trending ML GitHub repos, and most-discussed HN posts. Format with sections, include links, keep under 500 words." \
--name "Weekly AI digest" \
--deliver telegramOr just ask in plain words. The reference gives this example: “Every morning at 9am, check Hacker News for AI news and send me a summary on Telegram.” Hermes then uses its cron management tool itself, so no command is needed.
A job can have --skill attached (repeat the flag for several, and they load in order), which is how a job reuses a procedure you already installed; see the skills guide. A job can also pin its own timezone, for example hermes cron create "0 9 * * 1-5" --timezone Asia/Hong_Kong "HK market open brief". Without that, the expression is read in your profile’s timezone.
Schedule formats
The reference lists five kinds of schedule, with these examples:
| Kind | Example | Meaning |
|---|---|---|
| Relative delay (one-shot) | in 30m | Run once in 30 minutes (also in 2h, in 1d) |
| Interval (recurring) | every 2h | Every 2 hours (also every 30m, every 1d, every hour) |
| Day and time (recurring) | weekdays at 9am | Weekdays at 9:00 (also every monday 9am, daily at 7am, weekends at 10am) |
| Cron expression | 0 9 * * 1-5 | Weekdays at 9:00 (also 0 */6 * * * for every 6 hours) |
| ISO timestamp | 2026-03-15T09:00:00 | One time, at that date and time |
Times in the day-and-time form accept 9am, 9:30pm, 14:00, noon and midnight. By default a one-shot job runs once, while interval and cron expression jobs repeat until removed. If you want a limit, the job takes a repeat count.
By default, a recurring job missed while the gateway was down catches up once when the scheduler returns. Multiple missed occurrences collapse into one run. Without a per-job override or an explicit grace limit, this follows cron.catch_up_missed, which defaults to true. A job can override that policy with --catch-up on or --catch-up off. For cron expressions, --grace-minutes sets an explicit grace window; --catch-up on still permits one late run beyond that window.
Where results go and how to write the prompt
The agent’s final response is delivered automatically to the job’s delivery target, so the prompt does not need to tell it to send anything. The reference lists these targets:
origin: back to the chat where the job was created. This is the default on messaging platforms.local: saved as files under~/.hermes/cron/output/. This is the default on the command line.- A platform name such as
telegramordiscordfor its home channel, or a specific chat such astelegram:123456. The list also includes Slack, WhatsApp, Signal, Matrix, email, SMS, Feishu, WeCom and Weixin. - Several at once, separated by commas, such as
telegram,discord.
Set it with --deliver. For a monitoring job that should speak only when something changed, tell the agent to reply with only [SILENT] when there is nothing to report. The reference says delivery is then suppressed, the output is still saved locally, and a failed job always delivers regardless.
The prompt must contain everything the agent needs that is not already provided by attached skills. A cron job runs in a completely fresh session, with no memory of your current chat. The docs give a bad and a good example. The bad one is “Check on that server issue”. The good one names the server, the user, the exact command to run and the page that should return HTTP 200. Persistent memory does load, but the guide warns against relying on it for details a job cannot work without. If you want to start from a worked example, the automating-with-cron guide has five patterns, including a website change monitor and a weekly report.
Script-only jobs (no AI)
Some jobs do not need an AI at all, such as a disk or memory alert. Hermes calls this no-agent mode: the scheduler runs a script on schedule and delivers its output as the message, with no model call and no tokens spent.
The script has to exist first. Ask Hermes to create ~/.hermes/scripts/memory-watchdog.sh so that it checks memory use and prints an alert only when your threshold is passed. Once the file is there, schedule it:
hermes cron create "every 5m" --no-agent --script memory-watchdog.sh --deliver telegram --name "memory-watchdog"- The script’s trimmed output is the message. Empty output means a silent tick, so a watchdog only speaks when something is wrong.
- A non-zero exit or a timeout delivers an error alert, so a broken watchdog does not fail silently.
- Scripts must sit in
~/.hermes/scripts/. Files ending in.shor.bashrun under Bash. Other files run under the current Python interpreter unless the job specifies a Python interpreter with--interpreter. The script argument is a file, never a command line. - Scripts do not inherit your provider API keys from the Hermes environment.
A script can also be attached to a normal job, where its output becomes context for the agent. The pattern in the guide is to let the script do the fetching and comparing and let the agent judge whether a change matters.
List, edit, pause and remove jobs
You do not need to delete and recreate a job to change it. You can list, edit, pause, resume, trigger and remove jobs with /cron in chat or hermes cron in the shell, and a job can be referred to by ID or by its name. To stop a run that is already in progress, use hermes cron cancel or ask Hermes to cancel it.
hermes cron list
hermes cron edit <job_id> --schedule "every 4h"
hermes cron edit <job_id> --prompt "Use the revised task"
hermes cron pause <job_id_or_name>
hermes cron resume <job_id_or_name>
hermes cron run <job_id_or_name>
hermes cron cancel <job_id_or_name>
hermes cron remove <job_id_or_name>pause only stops future fires; a run already in progress keeps going until it finishes or times out, and cancel is the command that stops that one run while leaving the schedule alone. hermes pause is a global stop for every scheduled job, lifted with hermes resume. Hermes can also manage individual jobs, so you can say “pause the morning digest” in chat.
Why a job does not fire, and what to check
The troubleshooting guide says most problems are timing, delivery, permissions or skill loading, and to work through the checks in order.
- Is the job there and active? Run
hermes cron list. The state should read active, not paused or completed. A completed job may have used up its repeat count. - Is the schedule what you meant? A misformatted schedule can silently become a one-shot or be rejected. A job that fires once and then disappears from the list was a one-shot schedule such as
in 30m, which is expected. - Is the gateway running? Agent jobs need a running gateway. The ticker lives in the gateway.
hermes cron tickdoes not start a gateway; with none running it skips agent jobs until the next tick afterhermes gateway start. Ifhermes cron statussays the next run is overdue, restart the gateway withhermes gateway restartor run the job now withhermes cron run <id>. - Is the clock right? Cron expressions use the profile timezone unless the job pins its own. Compare the next run time in
hermes cron listwith the intended timezone, and usedateto check the system clock. - Did it run but deliver nothing? The job still runs if delivery fails. Check that the target is spelled correctly and that the platform is set up, for example a Telegram bot token in
~/.hermes/.env, and that the bot may post in that chat. Also check that the reply was not[SILENT]or empty. If a platform handles the wrapper text badly, setwrap_responsetofalseundercroninconfig.yaml:cron: wrap_response: false - Does it depend on a skill or a script? Skill names are case-sensitive and must match the folder in
hermes skills list. Cron management is disabled by default, and the messaging and clarify toolsets are disabled in cron sessions, so a skill that needs to ask you a question will not work there. A script path must exist, for example under~/.hermes/scripts/. - Is there more than one gateway? The scheduler uses a file lock to avoid overlapping ticks, and two gateways can delay or skip jobs. Keep one.
Two smaller points. The troubleshooting guide recommends staggering jobs by a few minutes to avoid delays when several are due together. And the inactivity timeout for a job defaults to 600 seconds (HERMES_CRON_TIMEOUT, with 0 for unlimited); it only fires after sustained inactivity, not while the agent is calling tools. For logs, look at ~/.hermes/logs/agent.log and hermes logs.
Scheduled jobs on ReadyHermes
On ReadyHermes your assistant runs on our servers and stays on, so a scheduled job fires while your own computer is off. You do not need to keep your own computer on for scheduled jobs. You set a job up by asking in chat in plain words, for example “every weekday at 08:00 send me today’s plan”, and the result arrives in the workspace chat or the chat app you connected, such as Telegram.
Some things to know before you plan around it. The workspace has no separate screen for listing or editing schedules, so you ask the assistant to list, pause or remove them. Scheduled runs use AI like any other message, so they count against your included AI or your own key. We have not verified a job limit, a timezone setting screen or script-only jobs on ReadyHermes, so do not assume them. Plans start at US$3.99 a month (Lite) or US$6.99 (Pro), each with US$1 of AI included every month. If you would rather run the gateway yourself, the self-hosting comparison lays out the trade-off, and managed hosting describes the hosted option.
Questions
How do I create a cron job in Hermes Agent?
In a chat, use /cron add with a schedule and a prompt, for example /cron add "every 2h" "Check server status". From the shell use hermes cron create with the same two arguments. You can also ask Hermes in plain words, such as every morning at 9am, and it creates the job itself.
Why is my Hermes cron job not running?
Agent jobs need a running gateway: its scheduler ticks every 60 seconds, and a normal CLI chat does not fire jobs. Also check that the job is active in hermes cron list, that the schedule is what you meant, and that only one gateway is running.
Where does a Hermes cron job send its result?
To the job's delivery target. The default is origin (the chat where it was created) on messaging platforms and local files under ~/.hermes/cron/output/ on the command line. Set --deliver to a platform such as telegram, a specific chat, or several targets separated by commas.
Why must a cron prompt be self-contained?
Each run is a completely fresh agent session with no memory of your current chat, so the prompt must contain everything the agent needs. Name the server, the command and the check to make instead of writing something vague like check on that server issue.
Can a Hermes scheduled job run without using AI?
Yes. A no-agent job runs a script on schedule and delivers its output with no model call, created with hermes cron create and the --no-agent and --script flags. Empty output is a silent tick, and a non-zero exit sends an error alert. On ReadyHermes we have not verified script-only jobs.