What Is Hermes Agent, and Should Your Business Use It?

Scott Hewitson gave his parents an employee who never asks for a raise.

Two Retirees, One Telegram Chat, and a Business That Runs Itself

His parents are in their sixties. They run a service business with no CRM and no back office, just a notebook, a phone, and hours of paperwork every single night after the last job of the day. Scott Hewitson built them a Hermes Agent and pointed it at Telegram, the app they already had open anyway to talk to their kids.

Ten days later, according to Nous Research, the company behind the agent, that Hermes Agent was estimating new jobs, creating Stripe payment links, reading the inbox, analysing the month's financials, pitching marketing ideas, and buying its own software with a real credit card when it needed a tool it did not have. Nous Research picked the build as a runner-up in its own hackathon and called it exactly what it is: an agent that runs a real small business through the one app two non-technical people already trust.

Notice what did not happen. Nobody hired a developer. Nobody bought new hardware. Nobody learned to code. A person who understood the business sat down, described it once, and let an agent that remembers everything take the paperwork.

That is the part worth stopping on, whether you sell plumbing, run a boutique, manage a clinic, or ship products from a warehouse. The bottleneck was never the business. It was the eight hours between closing time and bed that used to disappear into invoices.

What Is Hermes Agent, Exactly?

Hermes Agent is an AI agent built by Nous Research, the open AI research lab whose Hermes work is led by the engineer known online as Teknium, that installs on your own computer, a small cloud server, or Nous Research's own managed cloud, and then answers you from wherever you already work: Telegram, WhatsApp, Slack, Discord, Signal, email, or a plain terminal. According to Nous Research's own product page, it is one agent with one memory across every one of those surfaces, so a question you asked it on Telegram this morning is still remembered when you message it on Slack tonight.

Two things separate it from a chatbot tab you already have open. First, it is model-agnostic: you can run it on Claude, ChatGPT, Grok, Kimi, GLM, or a free local model, and swap between them without waiting for a vendor to add support. Second, it writes its own playbooks. Nous Research says it "learns your projects, auto-generates skills, and never forgets how it solved a problem," so the fifth time you ask it to draft a quote, it is not starting from zero, it is reusing what worked the first four times.

For most of this year, the loudest open-source agent story was a rival project called OpenClaw. That changed fast. On a widely watched walkthrough with Greg Isenberg, longtime OpenClaw builder Alex Finn said flatly: "I believe this is the moment Hermes overtook OpenClaw." That single video has pulled more than 138,000 views, a sign this is not a niche argument between two hobby projects, it is where a real chunk of the AI-agent audience is currently looking.

Why Everyone Is Talking About It This Week

Two things happened close together. Nous Research shipped an official companion skill that lets Hermes drive Unreal Engine through Epic's own MCP connector, turning plain-English prompts into finished 3D scenes, and creators reacted immediately.

At the same time, Hermes Cloud, the managed hosting layer for people who do not want to run a server themselves, has been rolling out more broadly. Nous Research describes it plainly: "Pick a name and model, your agent is online in seconds. No servers, no DevOps, no YAML," and it bills for active work rather than a flat subscription, scaling to zero the moment your agent is idle. That single change removes the one real barrier that used to keep non-technical owners like Scott Hewitson's parents out of this entirely: somebody technical setting up a server first.

Isn't This Just a Chatbot With Extra Steps?

No, and the difference is worth being precise about. A chatbot answers a question and hands the follow-through back to you. An agent finishes the follow-through. Ask a chatbot to help with a customer refund and it drafts an apology you still have to send and a credit you still have to issue. Ask Hermes Agent the same thing inside a workflow it already knows, and it can check the order, draft the message, generate the Stripe refund link, and log what happened, in one pass.

Hermes does this with two building blocks that are easy to confuse. A profile is a separate Hermes agent with its own skills, its own personality file, and its own memory, built for a role that needs a distinct skill set, your quoting agent and your customer-reply agent should probably be different profiles. A sub-agent is a temporary copy of one agent, spun up to run the same skill on several things at once, five sub-agents each building one feature of the same small tool in parallel, for instance. Different roles get different profiles. The same job repeated many times at once gets sub-agents.

How to Actually Set This Up for Your Business

Skip the temptation to hand it everything on day one. Every teardown of a working setup, including the walkthrough that convinced Alex Finn to leave OpenClaw, follows roughly the same order.

1. Pick the channel you already live in. WhatsApp, Telegram, Slack, or email, whichever one you already check without thinking. Hermes meets you there instead of asking you to learn a new app.
2. Brain-dump your business once, then ask it to write its own instructions. Describe your services, prices, policies, and the way you actually talk to customers in one long message. Then, instead of guessing at the perfect prompt yourself, ask the agent: "Based on what you now know about my business, what is the best way to set this up?" This one habit, called reverse prompting, consistently produces better cron jobs and templates than owners writing instructions from scratch.
3. Set exactly one scheduled job before anything else: a daily brief. Unread messages, unpaid invoices, and tomorrow's bookings, delivered every morning at a fixed time. Confirm it actually fired before you build anything on top of it.
4. Give it one boring, high-value job, not everything at once. Late invoice follow-up is a strong first choice for almost any business: draft a polite reminder at day seven, a firmer one at day fourteen, and a call to your attention at day twenty-one, all logged, none of it landing on your desk unless a client actually needs you personally.
5. Keep separate sessions for separate topics. One thread per subject keeps each message slim, and slim messages are the single biggest lever on your bill. A single sprawling thread that mixes invoices, marketing ideas, and personal questions is the most common way people accidentally triple their costs.
6. Decide self-hosted or Hermes Cloud. If you already have a spare computer or a small server and someone willing to keep it patched, self-hosting is free beyond the AI usage itself. If you want zero setup, Hermes Cloud is built for exactly that trade.

One more habit worth borrowing before you connect anything to real money. A separate hackathon entry, called Custodian, put a small approval layer between the agent and any payment: the agent proposes a purchase or a payout, a rules-based check approves or blocks it, and only then does money move. You do not need custom software to copy the idea. A simple rule such as "never send a payment above this amount without asking me first" inside the agent's own instructions does the same job.

What You Have After This

A week in, you have stopped forgetting to chase the invoice that has been sitting there since the ninth. A month in, your morning brief tells you what actually needs a human before you have finished your coffee, and everything else has already been handled. Three months in, you notice you are the one deciding whether to take on a new client, not the one drafting the seventeenth follow-up email of the week.

Same business. Same customers. A completely different Tuesday.

Frequently Asked Questions

Common questions people ask when they first hear about Hermes Agent.