Hermes Agent Just Learned to Clock Itself In

Your calendar never forgets a client meeting. It has, however, never once sent an invoice on your behalf.

It is 11:14 p.m. In Austin, and Priya Shah is still at her laptop. Her six-person growth marketing studio landed two new retainers this month, which is the good news. The bad news is what that actually means on a Tuesday night: three client status emails she meant to send after lunch, a proposal draft for a lead who is going quiet, and a $3,200 retainer invoice for a client that is now eleven days late because sending it always slides to tomorrow.

None of this is strategy. None of it needed her ten years of experience. It is just the small stuff, and the small stuff does not stop arriving because she is tired.

You Are Not Behind on Strategy. You Are Behind on the Boring Stuff.

Ask most small business owners what is actually eating their week and you rarely hear about the big decisions. You hear about the status update. The follow up. The invoice. The brief nobody wrote down. One at a time, none of it looks urgent. Stacked up every single week, it is the whole job.

This is where most people reach for a chatbot. They open ChatGPT, ask it to draft the invoice or the email, copy the answer out, and close the tab. It helped, for four minutes. Then Wednesday arrives and they are back at square one, retyping the same context about the client, the retainer amount, and the tone they use, because the chat forgot everything the second the tab closed.

A chatbot answers. It does not remember. It definitely does not check in on its own. That gap, between something that responds when asked and something that keeps a job running without being asked, is the entire story of what shipped in the AI agent space this week.

Hermes Agent Just Learned to Clock Itself In

Hermes Agent is a free, open-source AI agent built by the research lab Nous Research. It has been climbing fast this year, and this week it shipped a feature called Recurring Loops. Shubham Saboo, a Google AI product manager and one of the most followed builders in the AI agent space, put it plainly when he posted it: give Hermes a task and it repeats. Leave the interval off and it picks its own pace. Like cron, he said, except it runs inside your session with full context.

Nous Research followed a few days earlier with Bot Mode: your agent profiles become named Bots, each with its own role, model, memory, and skills, so one Hermes install can run a support Bot and an admin Bot side by side instead of forcing everything through one chat.

None of this is a scheduling app with a calendar grid and a save button. It is a sentence.

Isn't This Just Another Scheduling App?

No, and the difference matters. A scheduling app runs the same fixed action every time: send this email, post this update. Hermes running a Recurring Loop is closer to handing the job to a person who happens to be available at 7 a.m. Every day. It can check three different sources, use judgment about what counts as urgent, write the summary in your voice, and send it wherever you actually read your messages.

It also does something a scheduling app cannot: it gets better at the job. Every time you teach Hermes a task, from writing a proposal to drafting an invoice, it saves that process as a reusable skill in its own library. The next time a similar job comes in, it is faster, because it already knows the shape of the work. Teach it once, and it keeps the playbook forever.

How Do You Set Up a Recurring AI Agent for Your Business?

Here is the pattern, pulled from Nick Puru's hands-on walkthrough of running his agency almost entirely through Hermes. It holds regardless of what kind of small business you run.

1. Install the desktop app. Go to the Hermes desktop page, download the version for your operating system, and open it. No terminal, no code. A few minutes later it tells you it is ready.
2. Connect it to Telegram. This is what turns it from "a program on my desk" into "a coworker I can text from the coffee shop." You create a bot inside Telegram, paste one API key into Hermes, and from then on you can message it from your phone exactly like a contact.
3. Train it on your business, once. Give it your business details, your rates, your tone, the names of your regular clients. This is the part people skip, and skipping it is why so many people end up with an expensive chatbot instead of an employee. The training is the whole difference.
4. Give it one real job. Not a test question. An actual task: draft this month's invoice for a specific client, using our standard retainer amount, laid out as a clean document. Watch it do the work and quietly save that process as a skill.
5. Turn the job into a loop, in one sentence. Something like: "Every Monday at 7, check overnight client emails, my calendar, and open deal status, and send me a brief on Telegram." That one sentence is the entire setup. No settings screen. It is now a job in Hermes's own list of recurring tasks, sitting alongside anything else you have handed it.
6. Let the skill library compound. Every job you teach it stacks. Proposal writing becomes a skill. Client status checks become a skill. Invoicing becomes a skill. Month over month, the list of things it can already do without retraining just keeps growing.

What This Looks Like on a Monday Morning

Priya set up two loops after her late night. The first checks her inbox, calendar, and open deals every morning at 7 and lands one clean brief in her Telegram before she is awake. No client email waits more than a day for a reply flag again. The second runs on the first of the month: it drafts the retainer invoice for each client from memory, using the rate on file, ready for her to glance at and send through the accounting tool she already uses.

Neither loop needed a developer. Neither needed a new subscription. They needed one sentence each, and a business owner who was willing to spend twenty minutes training the agent properly before asking it to run itself.

What You Have After This

A week in, nothing falls through a crack that used to fall through a crack every single week. A month in, the skill library has quietly picked up four or five of your most repeated jobs, and new ones take less time to teach because the pattern is familiar. Six months in, you are not doing less work because you got faster at typing. You are doing less work because a good chunk of it no longer requires you to be in the room.

Same team. Same client list. A completely different Tuesday night.