Hermes Agent is an open-source AI agent you can run yourself or use hosted, reached through messaging channels and able to run scheduled tasks.

Hermes Agent is an open-source AI agent from Nous Research. It runs on macOS, Windows, or Linux, and there is also an optional hosted cloud agent. You can message it from apps such as Telegram, Discord, Slack, WhatsApp, Signal, email, or a command line, and it can keep persistent memory and reusable skills. The software is free under the MIT license; model providers and optional hosted services have their own pricing. We have worked with Hermes Agent and set it up for clients. This page explains what it is, what the official product actually does, and when we recommend it. Product facts are taken from hermes-agent.nousresearch.com and the official docs, current as of November 10, 2026.

What is Hermes Agent?

Hermes Agent is software you install so an assistant can work with you over time, not only answer a single prompt. Official pages describe an agent that lives across several chat surfaces with one memory, remembers projects, can schedule work in natural language, and can hand pieces of work to isolated subagents. It can also search the web, drive a browser, and run in an isolated sandbox.

That is a different shape from a chatbot you visit when you have a question. The point is an assistant that can keep context and do work, including unattended tasks, on a computer you control or on an optional hosted agent.

Nous Research publishes desktop apps for Mac and Windows and a terminal install for Linux and other setups. The official FAQ says you connect a model provider after install, and that you can use your own provider or connect to Nous Portal for model and tool credits. Cloud hosting is optional. We do not resell Nous Portal, and we do not treat any one model provider as required.

What can it do?

These capabilities are stated on the official site and in the official user guide:

  • It can meet you in more than one place. Official messaging and access surfaces include Telegram, Discord, Slack, WhatsApp, Signal, email, and the CLI, with a desktop app as another way to run and talk to it.
  • Persistent memory. Official docs describe bounded, curated memory that persists across sessions, so the agent can remember preferences, projects, and things it has learned.
  • Skills. Skills are on-demand knowledge the agent can load when needed. Official docs say the agent can also create and update its own skills, and that third-party skills are scanned before install.
  • Scheduling. The official site describes natural-language scheduling for reports, backups, and briefings that can run unattended through the gateway.
  • Delegation. Isolated subagents can run with their own conversations and terminals, which is useful when you want one job not to crowd out another.
  • Web and computer use. Official features include web search, browser automation, vision, image generation, text-to-speech, and multi-model reasoning.
  • Sandboxing. Official docs describe several backends, including local, Docker, SSH, Singularity, and Modal, with container hardening when you use an isolated backend.

You do not have to turn all of that on. A first business setup is usually smaller: one channel, one or two workflows, and tight limits on what the agent may change without asking.

How do we set it up?

We have worked with Hermes Agent and set it up for clients.

The first step is not the install command. It is a conversation about the work you want off your plate, the tools you already use, and how comfortable you are supervising an agent. If Hermes Agent is a fit, we install it on a machine you control, or we help you decide whether an optional hosted cloud agent is the better operational choice. We connect the chat channel you will actually use, hook it to a model provider you can maintain, and set permissions so the agent is not wide open on day one.

Official security docs describe a defense-in-depth model: who is allowed to talk to the agent, approval for dangerous commands, file-write protections, optional container isolation, and scanning for prompt injection in context files, among other layers. The default approval mode is designed to catch dangerous commands and ask a person; official docs also warn that “YOLO” modes turn those prompts off and should be used only in trusted environments. In a business setup, we keep approvals on for anything that can delete data, send a message to the outside world, or change account access.

We walk you through how to talk to the agent, how to approve or deny a risky step, and how to tell when it should stop. Setup is paid work after a free 30-minute discovery call. We quote before anything starts.

When is Hermes Agent a good fit?

Hermes Agent is often a good fit when you want an open-source agent on your own computer, you care about official sandbox and approval controls, and you might want an optional hosted cloud agent later without changing product families. It can also fit people who already live in Telegram, Discord, Slack, WhatsApp, Signal, or email and want one agent they can reach from there.

It may be less of a fit if you want no machine and no gateway to operate at all. A hosted teammate such as Grok Bot is built around a cloud computer that keeps running when your laptop is closed. If you want a local, foundation-stewarded assistant with no hosted subscription tier, compare OpenClaw. The longer comparison is Hermes Agent vs OpenClaw.

As with any agent, fit also depends on the work. An inbox draft helper and a fully authorized browser agent are not the same project. We would rather start with a job you can see and correct than with every feature at once.

How is security handled?

Official Hermes Agent security guidance is published at hermes-agent.nousresearch.com/docs/user-guide/security. Among the points we rely on when we set it up:

  • Control who can talk to the agent, with allowlists and DM pairing rather than open access.
  • Keep dangerous-command approval turned on for real work. Official docs treat disabling it as a special case.
  • Prefer an isolated terminal backend for a gateway that will run unattended.
  • Be careful with secrets, environment variables, and third-party skills. Official docs describe scanning for third-party skills and filtering credentials in several places.

Our broader, tool-agnostic advice is in Securing an AI agent: basics. If official docs and this page ever disagree, follow the official docs.

Official sources

Facts on this page are taken from those sources as of November 10, 2026. We do not invent features, client names, case studies, or numbers.

Questions

Is Hermes Agent free?

The official FAQ says Hermes Agent is free and open source under the MIT license. Model providers and optional hosted services have their own pricing. You can use your own provider or connect to Nous Portal for model and tool credits; cloud hosting is optional. Our setup work is quoted separately after a discovery call.

Does Hermes Agent run only on my computer?

No. Officially it runs on Mac, Windows, or Linux, and you can also deploy an optional hosted cloud agent. Stopping a local agent ends the process but does not delete saved conversations, memories, or skills, according to the official FAQ. Hosted agents are managed separately.

What apps can I message it from?

The official site lists Telegram, Discord, Slack, WhatsApp, Signal, email, and the CLI, and it ships desktop apps as well. We connect the channel you will actually use, not every channel at once.

How is Hermes Agent different from OpenClaw?

Both are open-source agents you can run on your own machine and reach from chat apps, with memory and skills. Hermes Agent also documents an optional hosted cloud agent and a detailed official security and sandbox model. OpenClaw is stewarded by a 501(c)(3) foundation and, officially, has no hosted subscription tier. See Hermes Agent vs OpenClaw for a longer, dated comparison.

Do you only set up Hermes Agent?

No. We review your stack and recommend what fits. We have also worked with OpenClaw and Grok Bot and set those up for clients.