01What the Herdr Hermes Agent Setup Is and How It Works
The Herdr Hermes Agent setup puts a whole AI team on one screen, with Hermes Agent as the boss and Herdr as the office. You say one sentence, Hermes hands out the jobs, and Herdr tells you the moment a helper finishes, gets stuck or has a question.
Herdr is a free, open-source app built for AI agents. It runs on Mac, Windows and Linux, and in this episode Julian Goldie notes that the newest version came out on the 8th of November. Julian also says that, as far as he can tell, nobody had shown Hermes running a team inside Herdr before.
Here is the basic layout:
- Hermes sits in the first pane as the team lead.
- Next to it are three Claude Code helpers: the scout, the writer and the checker.
- Each helper is a real terminal with one job.
The helpers keep running even if you close the window, because Herdr keeps them alive in the background. That means the team can keep going while you work on something else.
02Main Features: The Lead, the Three Helpers and the Board
The power of this setup comes from giving each agent one clear role. Hermes is the team lead, and it never does the work itself. It only hands out jobs.
The three helpers each have a single task:
- The scout scans what customers are asking right now and scores the winners.
- The writer makes the answer page and publishes it.
- The checker reviews the page and sends fixes straight back to the writer. It is not allowed to edit anything. It can only read and send notes.
On top of the team, you get a few extra tools:
- A Kanban board: every job lands as a card and slides to done when it is finished.
- Alerts from Herdr: pings when a helper is done, blocked or needs you.
- A gallery: everything the team made shows up in one place at the end.
The agents also talk to each other. A note from one helper lands in another helper's pane, so work moves between them without you passing messages along.
03How to Set Up Herdr and Hermes Agent Step by Step
Julian walks through five steps to build this team yourself. Herdr is free, so you can start right away.
- Install Herdr. Go to herdr.dev and download it, or paste the one-line install command into your terminal.
- Install Hermes Agent with its one-line command. Then run Hermes Kanban init to create the board.
- Open Herdr. Press the plus in the tab bar, start Hermes in a new space and give it the team playbook.
- Add the helper templates. Each template is a small file with a name, the program to run and the job. You need three: scout, writer and checker.
- Turn on alerts. Open settings, then alerts, and pick the sounds for done and blocked.
Once that is in place, you say the sentence: find today's topic and write the page. Then you watch the panes light up as each helper starts work.
The team playbook is one page of plain English. The key part tells Hermes that it is the team lead. Hermes reads the playbook and starts the three helpers with one command each, giving each helper its name and its job.
04Giving Orders With Hermes Jarvis
You do not have to type your orders. Julian uses Hermes Jarvis, a voice tool, to give the team its job. He opens Jarvis and says it out loud: find today's topic and write the page.
That sentence goes straight to Hermes inside Herdr. From there, the chain of events looks like this:
- Hermes reads the team playbook.
- Hermes starts the three helpers, each with its own job.
- The moment Hermes hands out a job, a card appears on the Kanban board.
- Each helper works in its own pane.
- As each helper finishes, its card slides across to done.
The nice part is that you do not need to chase anyone for an update. Julian points out that he did not ask a single helper how it was going. The board and the alerts told him.
The rest of the time, he worked on something else.
At the end, Jarvis also reads out a summary of what the team did. So you can start the job by voice and hear the result by voice, with the board showing every step in between.
05Walkthrough: How the Scout Finds Today's Topic
The scout is the first helper to start. It runs Julian's content ideas skill. Here is what it does, in order:
- It reads the business file, so it knows what the business does.
- It checks two real sources: recent videos on the topic and the questions people type into Google.
- It scores every question out of 100.
- It picks a winner.
In the run shown in this episode, the scout ran five searches, read 40 recent videos and found 74 questions. Every one of those questions got a score, and the top one became the topic for the day.
When the scout finishes, it flips from working to done. Its card on the Kanban board slides across, and Herdr can ping you so you know that stage is complete.
This is a good example of why one job per helper works well. The scout does not write anything. It only finds and scores topics.
That keeps its work focused, and it gives the writer a clear starting point instead of a vague request.
06Walkthrough: The Writer and Checker Working Together
Once the scout picks a topic, the writer makes the page and publishes it. Then the checker steps in. This is where the agents talk to each other.
The checker reads the page and writes notes. Each note lands in the writer's pane as if you typed it yourself. In this episode, the checker sent three fixes.
Each fix pointed to a claim the page could not back up.
One example shows how careful the checker is:
- The page said matching details build trust.
- The checker asked for a safer line.
- The new line said matching details make you easier to recognise.
The writer applied the fixes, published again and told Hermes it was done. Hermes then had a finished, checked page without Julian stepping in.
The rule that the checker cannot edit matters here. It can only read and send notes. That keeps a clean split between the agent that writes and the agent that reviews, so changes always go through the writer.
07Results: The Gallery and What the Team Produced
When the run is over, everything the team made lands in one gallery. In this episode, the gallery held three things:
- The trend scoreboard from the scout.
- The published page from the writer.
- The test site the page went on.
You click one card and it opens full size, so you can check each piece without hunting through folders. Then Jarvis reads you the summary.
Looking back at the full run, the team did the following from one spoken sentence:
- Found and scored 74 questions from five searches and 40 recent videos.
- Picked a winning topic.
- Wrote and published a page on it.
- Checked the page and fixed three claims it could not back up.
- Republished and reported back to Hermes.
Every one of those jobs showed up as a card on the Kanban board and slid to done. That is the main result: a team that runs a full content job while you watch the board, instead of you doing each step yourself.
08Tips and Limits for Your Herdr Hermes Agent Team
Julian shares two honest notes about his setup that are worth copying.
- The checker asks before it runs commands. He set this on purpose, so you could see a real needs-you ping from Herdr. It also means you stay in control of anything risky.
- His test site is a local one. Point the Rank Engine at your real site only when you trust the pages the team makes.
A few more tips from how this team is built:
- Give Hermes the lead role and tell it never to do the work itself.
- Give each helper one job only.
- Keep the checker read-only, so it can only send notes.
- Turn on alerts for done and blocked, so you do not need to watch every pane.
If you would rather copy this than build it from scratch, Julian shares the Herdr and Hermes team setup, the playbook, the three helper templates, Hermes Jarvis, the Kanban board, the content ideas skill and the Rank Engine skill inside the AI Profit Boardroom, all in the Agent OS with a guide for each one.