Hermes Agent Scheduled Tasks: Your AI That Works While You Sleep

The Problem With AI That Only Works When You Ask

Most AI tools are reactive. You ask a question, you get an answer. You close the chat, the interaction is over. Nothing happens until you come back and ask again.

That model works for search. It does not work for the things you actually want an assistant to handle: checking for updates, monitoring changes, sending reminders, compiling reports, and repeating any task that matters on a schedule.

Hermes Agent solves this with scheduled tasks. You describe what you want, when you want it, and the agent handles the rest. No scripts, no cron syntax memorization, no external automation tools. Just a natural-language instruction and a timeframe.

What Scheduled Tasks Can Actually Do

Hermes scheduled tasks (built on the cronjob tool) support several real workflows that go beyond simple reminders.

Morning Briefings

Tell Hermes to compile a summary every morning and send it to you on Telegram:

Every morning at 8am, check Hacker News for top AI stories, check the weather for Madrid, and send me a short briefing on Telegram.

Hermes reads your instruction, uses its web search tool to gather the data, formats a briefing, and delivers it to your messaging platform. You wake up to a useful summary instead of opening five apps.

Content and Feed Monitoring

If you follow specific blogs, RSS feeds, or topics, you can schedule Hermes to check them periodically:

Every 6 hours, check these feeds for new posts about "AI agents" and "LLM deployment" and send me a digest.

The blogwatcher skill is designed for exactly this. You attach it to the cron job and Hermes loads the skill's feed-checking logic before running the prompt.

Server and Infrastructure Checks

For anyone running services or managing projects:

Every 2 hours, check if my website returns a 200 status code. If it doesn't, send me a Telegram message immediately.

This is the kind of task that should run on autopilot. Hermes executes the check, evaluates the result, and only alerts you when something needs attention.

Recurring Research and Reports

Every Monday at 9am, search for news about "Hermes Agent" and "Nous Research" and compile a weekly report with links.

This is useful for founders, researchers, and anyone tracking a competitive landscape. Instead of doing manual searches every week, you get a compiled document delivered to you.

How to Set Up a Scheduled Task

Hermes gives you three ways to create scheduled tasks. All of them end up using the same cronjob tool internally.

Option 1: Natural Language in Chat

The simplest method. Just tell Hermes what you want:

Every day at 9am, send me a summary of my upcoming calendar events for the day.

Hermes parses the schedule and the task, confirms the details, and creates the job. You do not need to know cron syntax.

Option 2: The /cron Command

For more explicit control:

/cron add "every 2h" "Check server status and alert if down"
/cron add "0 9 * * *" "Morning briefing with AI news" --skill blogwatcher
/cron add "every 1h" "Check feeds and summarize new items" --skill blogwatcher --skill find-nearby

The --skill flag attaches one or more skills to the job. Skills are loaded before the prompt runs, giving the cron job access to specialized knowledge and workflows.

Option 3: The CLI

From the terminal:

hermes cron create "every 2h" "Check server status"
hermes cron create "0 9 * * *" "Morning AI news briefing" --skill blogwatcher

This is useful when you want to set up jobs programmatically or include them in a setup script.

Managing Running Jobs

Once jobs are created, Hermes gives you full lifecycle management:

List jobs: hermes cron list or ask "show me my scheduled tasks"
Pause a job: hermes cron pause <job_id> or "pause my morning briefing"
Resume a job: hermes cron resume <job_id>
Edit a job: Change the schedule, prompt, or attached skills without deleting and recreating
Trigger manually: Run a job immediately without waiting for the next schedule
Remove a job: Delete it permanently

This is not a fire-and-forget system. You have ongoing control over every scheduled task, and you can adjust them as your needs change.

How Skill-Backed Cron Jobs Work

The most powerful pattern is attaching skills to scheduled tasks. A skill is a reusable knowledge document that the agent loads on demand. When a cron job has skills attached, the agent loads those skills before executing the prompt.

For example:

cronjob(
  action="create",
  skills=["blogwatcher", "find-nearby"],
  prompt="Look for new local events and interesting nearby places, then combine them into one short brief.",
  schedule="every 6h",
  name="Local brief",
)

When this job runs, Hermes loads the blogwatcher skill (which knows how to check RSS/Atom feeds) and the find-nearby skill (which knows how to search for local places). The prompt combines both capabilities into a single output.

Skills are loaded in the order you specify. The prompt becomes the task instruction layered on top of those skills.

Chain Jobs Together With context_from

Sometimes one cron job is not enough. You might want a cheap script to poll an API every few minutes, a periodic job to enrich that data, and a daily agent job that takes both as input. Hermes supports this directly through the context_from parameter.

context_from tells the scheduler to prepend the output of one or more prior jobs into the prompt of the current job, so the agent sees their results as context.

cronjob(
  action="create",
  schedule="0 8 * * *",
  context_from="price_poll_job_id,news_sweep_job_id",
  prompt="Using the price snapshot and news context above, write the morning digest.",
  name="Morning digest",
)

This lets you split work across small, focused jobs instead of one expensive prompt. The cheap poll can run every 30 minutes for cents, the periodic step can run hourly with a single skill, and the agent step only fires once a day when it actually has something to say.

Silent Watchdog Mode With [SILENT]

By default, every cron run delivers its output to wherever the job was created. That is fine for digests and reports. It is the wrong default for watchdogs, where you only want to hear about problems.

Prefix a cron output with [SILENT] and Hermes suppresses delivery for that run. The local audit log still records what happened, but no message is sent. Failures always deliver regardless of the prefix, so you cannot accidentally silence a real alert.

cronjob(
  action="create",
  schedule="every 2m",
  prompt="Check https://my-site.com. If status is 200, output [SILENT]. \
          Otherwise write a one-paragraph diagnosis: DNS, port, TLS, response body.",
  name="Site watchdog",
)

The result is a clean inbox. You hear nothing for days, then a real diagnosis arrives the moment something breaks.

Script-Only Jobs With no_agent

For tasks that do not need reasoning, the LLM is dead weight. A heartbeat check, an API poll, a file sync - these run faster and cheaper as plain scripts. Hermes supports this with no_agent=True.

When you create a cron job with no_agent=True, the scheduler runs your script on schedule and delivers its stdout directly. The agent never wakes up, no tokens are spent, no model call happens.

cronjob(
  action="create",
  schedule="every 30m",
  no_agent=True,
  script="crypto/fetch_prices.sh",
  name="Price poll",
)

This is the right mode for the cheap, frequent half of any pipeline. Pair it with an agent-driven job that uses context_from and you get an LLM where it adds value, and plain code where it does not.

Where Results Get Delivered

Cron jobs deliver results based on where they were created:

Created in Telegram: Results arrive as a Telegram message
Created in Discord: Results arrive in the Discord channel
Created in CLI: Results are logged locally

You can also configure delivery targets explicitly when creating jobs, so a job created on the CLI can still deliver to Telegram or another platform.

What Cron Jobs Cannot Do

Hermes sets a few guardrails:

Cron jobs run in fresh agent sessions. They do not inherit your current conversation context.
Cron jobs cannot create more cron jobs. This prevents runaway scheduling loops.
Jobs have access to the normal static tool list, not every tool available in an interactive session.

These are intentional safety constraints. The goal is reliable, repeatable execution without side effects.

Why This Matters for Non-Technical Users

Scheduled tasks are where Hermes crosses the line from "chatbot" to "assistant." A chatbot answers questions. An assistant handles things proactively.

If you are not a developer, the key insight is this: you do not need to write scripts, set up AWS Lambda functions, or learn Zapier. You describe the task in plain language, pick a schedule, and Hermes does the work.

The most common non-technical use cases are:

Daily briefings (news, weather, calendar)
Content monitoring (blogs, feeds, competitor sites)
Reminders and follow-ups delivered to your messaging app
Recurring research (weekly reports, market updates)
Health checks (is my website up? did my deployment succeed?)

All of these run automatically, without you opening a terminal or remembering to check something.

The Hosting Question

Scheduled tasks only work if Hermes is running. If the agent process stops, the cron jobs stop too. This is one of the strongest arguments for managed hosting: your scheduled tasks should not depend on your laptop staying awake or your VPS staying online.

Hermify keeps your agent running on persistent infrastructure, so your scheduled tasks execute reliably at the times you set. If you want the automation benefits of scheduled tasks without managing a server, that is the trade Hermify is designed to make.

Getting Started

If you already have Hermes running, try creating your first scheduled task right now:

In 30 minutes, remind me to check the build status.

That one-line instruction will prove the concept. From there, you can build up to daily briefings, weekly reports, and full automated workflows.

If you do not have Hermes running yet and want the scheduled-task experience without the infrastructure work, get started with Hermify and have your agent live in under 60 seconds.