♻️ Project 5: 24/7 Background Agents for $0
Up to this point, you talked with the agent in real time. In this project, it works on its own: you schedule tasks with hermes cron, they run in the background all the time — 24 hours a day, 7 days a week — and since everything is local, the cost is $0. You’ll run it in Vault (private) and learn to monitor the agent’s health with hermes status e hermes doctor.
🎯 Goal: agents running all the time
Until now, the agent only worked when you were there in front of it, typing. A background agent and the opposite: it runs in the background, on its own, at times you set — summarize emails every morning, scan a repository every night, organize your notes at the end of the day. You leave; it keeps going. And since the model is local, leaving this on 24/7 doesn't generate a bill.
hermes cron) is how you create these scheduled tasks.📋 What you’ll deliver
- •A task scheduled with
hermes cron. - •Confirmation that it runs on its own, without you there.
- •Everything in Vault — private, with nothing leaving the machine.
- •The agent monitored with
hermes statusehermes doctor.
Prerequisites: Hermes agent already installed and connected to the local model (Projects 2 and 3). Here we use the real Hermes CLI—the commands hermes cron, hermes status e hermes doctor are real; what doesn't have its own command (like turning on Vault) is described as an action in the interface.
Key concepts
An agent that runs in the background, without you at the keyboard.
Available all day, every day.
Where you see each scheduled run that was triggered.
Running locally doesn’t incur a per-run charge.
⏰ Step 1 — Schedule with hermes cron
The command that creates the schedule and the hermes cron. It opens Hermes's recurring task manager, where you specify what the agent should do and when. From then on, Hermes runs that task at the scheduled time, even when you're away from your keyboard.
New here? "Cron" is a classic computing term: it’s the system that triggers a task recurring at set times ("every morning at 8 a.m.", "every Monday"). The hermes cron brings this idea to the agent: instead of you asking every time, it remembers to do it on its own.
🎯 Command objective
Open the Hermes scheduler to create a recurring task that the agent will run on its own.
hermes cronHermes opens the scheduling assistant. Follow the interactive prompt (or the dashboard UI) to define the task — for example: "<your task>" = "summarize my notes in folder X every day at 7 p.m." You choose the task, recurrence, and local model that will run it.
How to verify: when it’s done, the schedule appears in the list (run hermes cron again to see the list) and the task then appears in the panel Activity as soon as you trigger it for the first time.
🔎 How to verify this step
Create a test schedule for next (e.g., in a few minutes). When the time comes, check the dashboard Activity that the task ran—that’s the sign that the hermes cron is actually triggering the agent on its own.
Key concepts
Command that opens the scheduler.
Something that repeats at a set time.
The assistant where you decide what happens and when.
Hermes runs without you asking again.
💸 Step 2 — Why this costs $0
Here's the trick behind local background agents. In a cloud service, leaving an agent running 24/7 would be expensive: every run, every token generated, shows up on the bill — and a loop that runs all day turns into a scary bill at the end of the month. Locally, after downloading the model, each run is free. There’s no meter. You only pay for the electricity your computer would already use.
O cron clock triggers the local agent, which runs the task and saves the result to the Activity; then the cycle starts again at the next scheduled hour. Notice the badge $0: every turn of that loop is free because the intelligence is on your machine — the opposite of the cloud, where every one of those turns would show up on the bill.
✓ Local 24/7
- ✓$0 per run — run it as many times as you want.
- ✓No rate limits: no "you’ve reached your limit".
- ✓Predictable bill: always zero.
- ✓You can run “token-expensive” loops without worrying.
✗ The same in the cloud
- ✗Each scheduled run charges per token.
- ✗A 24/7 loop can turn into an expensive bill.
- ✗Rate limits can block scheduling.
- ✗You end up "rationing" the agent to save money.
💡 Practical tip
Since there is no cost per run, it's worth starting generously: schedule several small tasks and see which ones actually help. The only limit is your machine's CPU/GPU time — not your credit card.
Key concepts
Running locally doesn’t cost per task or per token.
A 24/7 loop in the cloud multiplies the bill.
No usage limits imposed by third parties.
Machine time, not money.
🗄️ Step 3 — Run in Vault (private)
An agent running on its own 24/7 handles your data without you supervising every step. That’s why the right place for it is the Vault — the air-gapped mode you saw in Track 1: everything happens locally, nothing goes out over the internet. This lets your background agent scan sensitive notes, emails, or repositories with the assurance that this data never travels outside.
Reminder: "Vault" is Hermes’s private/airgapped mode (literally like unplugging the internet cable). It doesn’t have its own command in the CLI list — you enable it through the Hermes interface (the mode selector / privacy toggle). Leave it on before the scheduled task runs.
Confirm the local model
The scheduled task needs to run on your Ollama model (e.g., qwen3-coder-64k), not in the cloud.
Turn on the Vault in the UI
In Hermes, enable mode Vault (airgapped) before letting the agent run on its own.
Let the schedule trigger
When cron runs, it executes in Vault—private end to end.
✓ Good for running in Vault 24/7
- ✓Summarize emails or notes containing personal data.
- ✓Scan proprietary code in a private repo.
- ✓Organize client / financial files.
- ✓Any task subject to compliance (healthcare, legal).
✗ Considerations
- ✗Forgetting to turn on the Vault: the task may use the cloud.
- ✗Accidentally point the agent to a cloud model in the schedule.
- ✗A task that requires internet access (web search) won't run in pure Vault mode.
- ✗Machine off = the agent doesn't run (you are the "server").
🔎 How to verify this step
With the Vault enabled, disconnect from the internet and let the schedule run (or force a run). If the task finishes offline and the result appears in the Activity, you have a 100% private background agent.
Key concepts
Airgadded mode: everything local, nothing leaves.
How to disconnect the internet cable.
Private even when you’re not supervising.
The scheduled task should use Ollama, not the cloud.
🩺 Step 4 — Monitor the agent's health
An agent that runs on its own needs to be lightly monitored: you want to know if it's up and healthy. Two real CLI commands do that. The hermes status shows the state the agent’s current status (on, connected, which model). The hermes doctor does a health check, pointing out what is wrong and how to fix it.
🎯 Commands objective
Confirm that the agent is up and diagnose any issues before they disrupt scheduled tasks.
hermes status
hermes doctorhermes status = "what's the agent's status right now?" (up, model, connection). hermes doctor = "is anything wrong?" — it checks the installation and health and suggests a fix.
How to verify: hermes status should report the active agent and the correct local model; hermes doctor should finish without critical errors (or list exactly what to adjust).
Check the status
Run hermes status and confirm: agent is running, local model is connected.
Run the doctor
hermes doctor diagnoses system health and suggests adjustments (PATH, connection, missing model).
Check Activity
Cross-check with the panel Activity: are the scheduled runs actually happening?
💡 Practical tip
Make it a habit to run hermes status every now and then — it’s cheap and alerts you early if the agent goes down. If anything seems odd, the hermes doctor and the first place to look before changing any configuration.
🔎 How to verify this step
hermes status shows the agent active and using the local model, e hermes doctor ends no critical errors. With both OK, you know the background agents will run reliably.
Key concepts
Current status: online, model, connection.
Health diagnostics with a suggested fix.
Know what the agent is doing without having to guess.
The log confirms that the scheduled runs executed.
✅ Step 5 — Result: free and private automation
If you made it this far, you have a agent working on its own: it runs at the time you scheduled, runs in Vault (private), costs nothing per run, and you know how to monitor its health. This is where local AI stops being "a chat window" and becomes infrastructure that works for you in the background, 24/7.
🏁 Expected result
- ✓A scheduled task with
hermes cronruns automatically at the scheduled time. - ✓It appears in the dashboard Activity.
- ✓Runs in Vault, with nothing leaving the machine, at $0.
- ✓
hermes statusehermes doctorconfirm that the agent is healthy.
⚠️ If something doesn’t match
- ✗Task didn't trigger? Run
hermes cronand check that the schedule exists. - ✗Agent “offline”? Check
hermes statusand thenhermes doctor. - ✗Is data leaving? Confirm that Vault is on and the model is local.
- ✗Nothing in Activity? The machine may have been turned off at the scheduled time.
Optional self-check: Why does a 24/7 background agent cost $0 when it runs locally?
🎯 Project summary
Next module:
3.6 — Project 6: Switch between Vault, Connected, and Cloud