Hermes Agent is a free, open-source AI agent from Nous Research. It runs all day on a rented server or a computer you own, takes requests from your team in Slack, Microsoft Teams or another chat app, does the work in the tools you connect, and writes down what it learns so the next job goes better.
We run VantaSoft's own team of 20 AI agents on Hermes Agent and set it up for customers as a managed service, so this explainer uses our own numbers as well as the documentation. Facts about the software come from the official documentation and the project's release notes, checked again on November 15, 2026.
Who makes Hermes Agent?
Nous Research makes Hermes Agent. The company describes itself as "a pioneer in open AI training and research," and it publishes the agent's code in the open on GitHub, where anyone can read it, copy it or suggest changes. The license is MIT, which allows business use at no charge.
The project's GitHub page dates to July 2025. On November 15, 2026, it showed more than 263,000 stars, which are bookmarks from people interested in it rather than a count of businesses using it. A better sign of how much work goes into it is the November 5 release notes, which credit 666 people with contributions since the October 13 release.
Nous Research also offers a hosted version, Hermes Cloud, through its paid Nous Portal plans, for people who would rather not run the machine themselves. What Hermes Agent costs a small business to run has the current plan details.
What does Hermes Agent do?
Hermes Agent carries out tasks as well as answering questions. It can read and write files, look things up on the web and use more than 60 built-in tools, plus the business apps you connect. It keeps working when nobody is talking to it, on a timetable you set, and it remembers what it learned between conversations.
Three things in the documentation matter most for a business:
- It gets better with use. After a task, it writes up how it did the work as a reusable procedure, which it calls a skill, and improves that procedure on later runs. A job it did last month goes more smoothly this month.
- It is not tied to one AI company. It works with AI models from Nous Portal, OpenRouter, OpenAI and most other sources, and moving to a newer model is a settings change rather than a rebuild. Since the October 13 release it can also run free AI models entirely on your own computer, with nothing leaving it, if the computer is strong enough.
- It does not depend on anyone's laptop. The project's own page says it can run on a $5-a-month rented server, on powerful dedicated hardware, or on pay-as-you-go cloud services that cost almost nothing while it sits idle.
What does Hermes Agent look like in a working business?
Each part of Hermes Agent matches something an owner already knows: a coworker you message, a list of recurring chores, a short notebook, written procedures, a task board and a sign-off rule. The table shows each part as the documentation describes it and what it looked like in our company on November 15, 2026.
| Part | What it does | At VantaSoft on Nov 15, 2026 |
|---|---|---|
| Chat apps | Takes requests in more than 20 chat apps, Slack and Teams among them | Each of our 20 agents has its own name in Slack |
| Separate agents | Runs several agents on one machine, each with its own logins, memory and skills | One agent per job, each with only the logins it needs |
| Scheduled jobs | Runs a task on a timetable and posts the result to a chat | 167 jobs switched on across 15 agents; 59 use no AI |
| Memory | Two small notebooks of about 2,200 and 1,375 characters | Kept short; standing rules go in each agent's job description |
| Skills | Written procedures it creates after a job and reuses | Our content agent keeps one for each website it publishes to |
| Task board | A shared board where agents hand each other work | 962 tasks in 30 days, 863 finished, across 17 agents |
| Approvals | Asks a person before risky commands, and blocks them in scheduled jobs by default | Owners approve each email reply to a customer and every payment |
Scheduled jobs are a big share of the work. The 167 that are switched on have run more than 139,000 times between them since the oldest was set up on July 5, and 111 of them ran in the 24 hours before we counted. Much of what our agents do is not a conversation at all. It is a chore that runs every morning or every half hour, and a person reads the result. If your first idea for an agent is a chatbot, look at your recurring chores first.
Memory is small on purpose. Each agent's notebook holds facts it picked up on the job. Anything that must always hold, such as who approves what, sits in a written job description the agent reads at the start of every conversation.
What each of our agents does lists the jobs one by one, and our Hermes Agent review after running our own team on it since July scores where it is strong and where it falls short.
What changed in recent Hermes Agent updates?
Hermes Agent changes quickly. Nous Research shipped ten tested releases between October 13 and November 9, 2026, and since November 13 it has also put out a test build each night. Two of the ten were large releases. This table covers their main changes and the fix that followed, from the official release notes.
| Release | Date | What changed | What it means for a business |
|---|---|---|---|
| v0.22.0 | Oct 13 | A new desktop install opens on a short setup chat that runs on a free trial model | You can try it before you sign up with any AI company |
| v0.22.0 | Oct 13 | Connectors: sign in once to apps such as Linear and Notion and the agent can work in them | Fewer technical steps to connect an app, though each connection is still a login you choose to hand over |
| v0.22.0 | Oct 13 | One-click setup for AI models that run entirely on your own computer | An option for private work, if your computer is strong enough. The release rates each model against your hardware |
| v0.22.0 | Oct 13 | Tested stable releases that installs follow by default, and signed desktop apps, including in the Microsoft Store | Fewer surprise breakages from untested updates |
| v0.23.0 | Nov 5 | The desktop workspace can open in a web browser, served from your own machine. It is off until someone turns it on | Your team could use it from any browser without installing anything, once someone sets it up with a sign-in |
| v0.23.0 | Nov 5 | One copy of the agent now owns each conversation, wherever you open it | Fixes messages that were written twice or went missing when people used it from several places |
| v0.23.0 | Nov 5 | Price tags and spending limits on paid add-on tools | A cap you can set on what paid tools may spend |
| v0.23.1 | Nov 9 | Fixes a fault that, from November 9, kept every install from seeing new add-ons | Installs left on the older version keep the problem until someone updates them |
The pace is what an owner should plan for. A business running Hermes Agent gets a new version most weeks, and some fixes arrive only when someone applies them. New features also bring new settings someone has to decide on, like the browser workspace, which stays off until a person turns it on and protects it. Our review describes what keeping up has taken for us.
Hermes Agent vs the alternatives you have probably heard of
| Option | What it is | Who runs it | Learns from past work | What you pay |
|---|---|---|---|---|
| Hermes Agent | A free AI agent that learns and works from your chat apps | You, or a provider you hire | Yes: skills, memory and a picture of you | Nothing for the software; you pay for the AI model and hosting |
| Claude Code and similar | Coding assistants for software developers | A developer, on their own computer | Mostly the project in front of it | The AI company's own pricing, tied to its models |
| n8n and other workflow builders | Automation that follows steps you set, with AI as one of the steps | The vendor's cloud, or your own server | No; every run follows the same steps | Paid hosted plans, or host it yourself |
| Hosted agent platforms | A vendor's agent builder in your web browser | The vendor | Varies by vendor | Monthly, per seat or per agent |
Workflow tools fit work whose steps are known in advance and rarely change. An agent like Hermes fits work that needs judgment and conversation, especially when you want the assistant to learn how your business does things.
If you already use OpenClaw and are thinking about moving, your choice is narrower than this table. We sorted that decision by reason in what to switch to when you leave OpenClaw, and what has to move with you, and the head-to-head is in what actually decides Hermes Agent against OpenClaw.
What does Hermes Agent cost?
The software is free. You pay for what surrounds it:
- A computer to run on. A small rented server is enough for one agent, and more agents or heavier work cost more. If you would rather not run the computer at all, the companies that host Hermes Agent and OpenClaw for you publish their prices.
- The AI model. The AI company you pick bills you for the text the agent reads and writes, and what an AI agent costs works through those numbers.
- Setup and upkeep. Someone has to install it, connect it to your tools and your chat app, decide what it may touch and keep it updated. That costs your team's time or a service fee. AI agent integration covers which of your systems are ready to connect.
In our experience the last line is the one owners underestimate. What Hermes Agent costs a small business to run puts today's published prices against each of these lines and gives you a worksheet for your own monthly number.
When should a business run Hermes Agent?
It is a good fit when:
- You want an assistant your team can message in Slack or Teams that does real work, such as reading a shared inbox, updating a task board, running a report, drafting a document or checking a system.
- You have chores that come around every day or every week and need some judgment each time.
- You want it to get better at your business instead of starting from zero each time.
- You want to pick which AI company's model it uses and keep the option to switch when a better one comes out.
It is a poor fit when:
- The job is one fixed sequence of steps that runs at high volume with no judgment in it. A workflow tool is simpler and cheaper for that.
- Nobody on your side will decide what it may reach and check its work. An agent that can act in your systems needs deliberate limits on what it can access.
How do you get started with Hermes Agent?
There are two routes, and the right one depends on who you have on staff.
If someone on your team is comfortable installing software, the installation guide covers Mac, Linux, Windows and Android, and the desktop app now starts with a setup chat. Installing takes minutes. Writing down what the agent should do, what it must never do and who reads its output takes much longer. A realistic first project is one task in one chat app for one month, with a named person reviewing everything the agent does. How to implement an AI agent in your business walks through that rollout.
If nobody on your team will own it, the other route is to have someone build, host and maintain the agent for you. That is what VantaSoft does. We design the work with you, run Hermes Agent on our managed setup, connect it to your systems and keep it up to date. The build starts at $1,000 one time, and hosting and maintenance start at $475 a month.
Frequently asked questions
Is Hermes Agent free?
Yes. It is open source under the MIT license, so the software costs nothing, including for business use. You pay for the AI model it uses, the computer it runs on, and the time or service it takes to set it up and keep it running.
Who made Hermes Agent?
Nous Research, an AI research company. It publishes the code on GitHub and puts out new versions most weeks, with release notes for each.
Is Hermes Agent open source?
Yes. Its full code is public under the MIT license, so anyone can read it, run it or change it, including for a business.
What models does Hermes Agent use?
Whichever AI model you connect it to, including those available through Nous Portal, OpenRouter and OpenAI. Changing models later is a quick settings change, and since October 13 it can also run free models on your own computer.
Does Hermes Agent run locally?
Yes. It can run on a computer you own or on a private rented server, and the October 13 release added one-click setup for AI models that run on your own machine with nothing leaving it. A desktop app is available for Windows and Mac.
How often is Hermes Agent updated?
Often. Nous Research shipped ten tested releases between October 13 and November 9, 2026, plus nightly test builds since November 13. Whoever runs your agent should plan to apply an update most weeks.
Is Hermes Agent secure enough for business data?
You choose where it runs, so the agent itself sits on machines you control, though what it reads still goes to the AI service it uses. How safe it is after that depends on the limits you set: which systems it can reach, how its logins are stored and who reviews what it does. Most of the security work lies in those choices. Whether Hermes Agent is safe for a business covers its own security defaults, its public flaw record and the settings to check first. The same questions decide the answer for OpenClaw, and the eight to settle before either one gets a real login work for both.
Scope your first Hermes Agent workflow
If you want Hermes Agent working in your business without setting it up yourself, bring the one task you would hand over first and we will scope it with you.




