Hermes Agent: 7 automations that run for me every week (and the real time they save)
While I’m recording a video, helping the kids with homework, or watching an NBA game, jobs are running on my server. They write an article, translate it, publish it, weave internal links and watch my Search Console. This is not a coach’s promise: it’s the exact list of my 7 Hermes Agent automations, with their real schedules.
I run mintavocado.com, a trilingual site, with Hermes Agent, an open source AI agent developed by Nous Research. For each automation: what it does, the mechanism behind it, and the real time cost. No invented numbers: everything comes from my crons, my audits and the articles already published on this site.
What these automations actually change
- 7 cron jobs run every week: writing, publishing, EN/ES translations, internal linking, watch, site improvements and article ideas.
- Every article first goes through a draft and a 12-hour validation window before publication.
- My real time: under 30 minutes of supervision for a 2000-word article, about 5 minutes of daily validation for the watch, a 20-minute retrospective per week.
- Hermes Agent is free and open source; the infrastructure runs on a VPS at around €5 per month.
If you’re looking for an “autopilot that turns you into a millionaire”, keep scrolling. What I describe here is an assistant, not a boss: human control stays everywhere.
1. The daily article gets written during my breakfast
From Monday to Thursday, at 8:30 a.m. Singapore time (00:30 UTC), a job writes an article and drops it as a draft. The mechanism: the job loads my writing skill and its three reference files (persona, style, WordPress guide), checks through the WordPress API that no existing article cannibalizes the topic, then creates a draft. Never a direct publication.
If no qualified topic is in the backlog, the job cancels itself: the real cadence is 3 to 4 articles per week, not “one per day at any cost”. This pattern, a conductor delegating to workers, is detailed in the multi-agent editorial factory and the subagents pattern.
Real time: a 2000-word SEO article goes from raw idea to publication in under 30 minutes of my time (review, tweaks, GO), as I document in my competitive watch stack.
2. Publishing kicks in after a 12-hour delay
Every 6 hours, a Python script scans the drafts queue. Any draft older than 12 hours without my “GO” is revalidated then published. I can speed things up with a single Discord message.
This worker contains no LLM call at all. The 12-hour rule is hardcoded, a pre-publication validation runs before every release, and already published posts are never touched.
Real time: zero minutes per day, the site lives on even when I’m not looking at it.
3. EN/ES translations follow the FR status
Every weekday evening at 21:30, a job syncs the English and Spanish versions of the day’s article. The principle: FR is the source of truth, translations follow. An FR rewrite triggers an EN/ES resync, an FR draft stays a draft on the translation side.
On the technical side: same featured image, same Gutenberg block structure, EN internal links pointing to EN posts (never to FR), and a single Polylang call ties the FR/EN/ES trio to generate hreflang. Latest audit: 39 complete articles in 3 languages, 0 orphan, 0 duplicate.
Real time: two translations a day without lifting a finger.
4. Internal linking weaves itself, then gets switched off
A daily job analyzed my internal link structure and added relevant links until it reached a target score of 8/10, spelled out right in the job’s name. Every run: content backup before editing, report of what changed.
A detail I own: this job is now paused, goal reached. An automation has a lifespan, it’s not a subscription you forget to cancel.
Real time: zero minutes during the campaign, and a cleaner link structure for Google.
5. The Monday morning watch: GSC quick wins
On Mondays at 9:30, a job reviews my Search Console: pages in decline, titles and excerpts to touch up. It hands me a short list of quick wins to validate.
It builds on my watch setup: 99 X accounts followed, Product Hunt and Hacker News monitored, 5 to 10 proposals a day with at least 4 actionable. I’m on a topic within hours of it showing up.
Real time: about 5 minutes of validation per day, that’s the name of the step in my process.
6. The smart weekly: one improvement per week, rotating day
Once a week, a job applies one concrete improvement to the site. Two mechanisms: the execution day is computed by an ISO rotation (Tuesday, then Thursday, then Friday…) so it never lands on the same slot, and the theme rotates across 5 workstreams: performance, GSC, internal linking, indexing, refresh of older articles.
Every intervention ends with a 20-minute retrospective: what broke, what do we improve next?
Real time: 20 minutes a week, and no more “I’ll deal with it later” projects dragging on for months.
7. The idea pool that refills twice a month
On the 1st and the 16th at 9 a.m., a job delivers a top 3 of ideas (keyword, angle, affiliate, category), taken from the past week’s watch, and adds them to the backlog automatically. Every Monday, a reminder lists the articles waiting for their affiliate links.
Real time: a few minutes of reading twice a month, and not a single Monday spent wondering what to publish.
The recap in one table
| Automation | Frequency | My real time |
|---|---|---|
| Daily draft writing | Mon-Thu, 8:30 a.m. | Under 30 min of supervision |
| Publishing after validation | Every 6 h | 0 min (LLM-free script) |
| EN/ES translation sync | Mon-Fri, 21:30 | 0 min |
| Internal linking | Daily, 3 UTC | 0 min (paused, goal reached) |
| GSC quick wins | Monday, 9:30 a.m. | 5 min of validation |
| Smart weekly | 1×/week, rotating day | 20 min retro |
| Ideas + affiliate reminder | 2×/month + Monday | A few min of reading |
What I keep under human control
In late August, I shut everything down. All jobs paused, site frozen for two weeks. Why? To take quality back into my own hands.
On September 14, I restarted a reduced process: mandatory draft, 12-hour window, technical validation before publication, a hard ban on touching already live posts. What the agent does not decide alone: validating an angle, picking an image, judging a risky title, committing the brand or money. My experience with what works when you delegate to an AI agent is clear: the more a decision commits the site long term, the more it stays human.
The agent prepares, I make the call. Less spectacular than the hype, but it doesn’t break.
Set up your first automation in 30 minutes
Where to start? Here is my path.
- Understand what an agent can do for a solopreneur: my AI agents pillar lays the foundations, no jargon.
- Set up the infrastructure, detailed in my AI agents setup: installing Hermes Agent takes 10 to 15 minutes on a VPS at around €5 per month.
- Launch ONE cron job, recurring and low-risk: the watch or the draft writing. Allow 30 minutes to 1 hour for a first workflow; the official Hermes Agent documentation covers installation step by step.
One job, two weeks of observation, then the next one: that’s how I went from a single watch job to these 7 automations.
Common questions about Hermes Agent and automations
Do you need to know how to code to run these automations?
Not to get started: skills, delegation, cron and human validation are enough for the 7 jobs described here. Code becomes useful later, to glue specific pieces together like the publishing worker, a pure Python script.
How much does a setup like yours cost?
Hermes Agent is open source and free. The cost sits in the VPS (around €5 per month) and in the language model API keys, billed by usage. No license, no hidden subscription.
What happens if the agent gets it wrong?
It doesn’t publish directly: draft first, a 12-hour window, then technical validation before going live. And if the system really derails, a full shutdown takes one afternoon: that’s what I did in late August for two weeks.
Can you let Hermes Agent publish without review?
Yes, and that’s my deliberate choice today. It rests on three guardrails: mandatory draft, a 12-hour delay you can interrupt with one message, and a hardcoded ban on touching published posts.
Where to start if you’re a solopreneur in a hurry?
With a single recurring, low-risk automation: the watch or the drafts. The full path is in my AI agents pillar: about thirty minutes for installation, under an hour for a first workflow.
The bottom line
These 7 automations don’t make me rich, they make me available: for my videos, for my kids, for the decisions that matter. It’s the only return on investment I take seriously.
If you want to build your own step by step, my free guide gives you the foundations to stop suffering AI and start building with it.
Leave a Reply