Hermes Agent: the open-source AI agent, explained for businesses.

What Hermes Agent is

Nous Research released Hermes Agent in February 2026 as an open-source project (MIT licence) with a simple pitch: "the agent that grows with you". Unlike a chatbot that forgets you when the window closes, Hermes keeps a persistent memory of your projects, turns the ways it solved a problem into reusable skills, and can be reached from the messaging apps you already use.

• Memory and skills: it learns your projects, generates its own skills and, in its makers’ words, never forgets how it solved a problem.
• Messaging platforms: Telegram, Discord, Slack, WhatsApp, Signal, email and the command line, with more being added.
• Scheduling in plain language: recurring reports, backups and briefings.
• Subagents: isolated helpers with their own conversations, terminals and Python scripts for bigger jobs.
• Tools: web search, browser automation, vision, image generation, text-to-speech and reasoning across several models.
• Sandboxes: it can run its work locally or inside Docker, over SSH, in Singularity or on Modal, so a mistake stays contained.

How it differs from Claude Cowork and n8n

Details from each product’s own site in September 2026.

They are not rivals. At ContentJet, Hermes keeps the folders in order, Cowork runs hiring, and n8n-style workflows connect the rest. See Claude Cowork for business and n8n vs Make.

How ContentJet keeps its client folders in order with it

A content agency accumulates files: briefs, videos, versions, approvals, one folder per client and per campaign. Left to people, the structure drifts within weeks. For ContentJet we put Hermes Agent in charge of the client folders: it keeps every client folder organised automatically, as one of five AI tools we built for the team, next to creator hiring on Claude Cowork, a Slack agent connected to Monday.com, Gmail and Drive, an AI strategy platform and an AI quality check on every video. Details and numbers are in the ContentJet case study.

How to install it

• Requirements: macOS 12 or newer, Windows 10 or 11, or any Linux, per the official site. A small server (VPS) is the usual home for an agent that must be always on.
• Install with the one-line command from hermes-agent.nousresearch.com (it downloads and sets up everything; no sudo needed).
• Connect a model provider: the Nous Portal, which gives access to 200+ models with a free tier and paid credit plans, or your own API keys from a provider.
• Connect a messaging platform (Telegram is the quickest to test) and, for always-on use, run it as a system service on the server.
• Start with one job in a sandbox (Docker is the safe default), give it only the folders and accounts that job needs, and read its logs for the first weeks.

What it costs to run

• The software: free, MIT licence.
• The models: billed by usage through your own provider keys, or through Nous Portal plans that run from a free tier to $20 to $200 a month with monthly credits, at the time of writing.
• Hosting: a small VPS if it must run around the clock; nothing if it lives on a laptop.
• The setup and supervision: the real cost, and what an agency takes on. Our builds start at $2,500, fixed before we start; see the AI agent pricing guide.

Rules we follow when deploying it for a client

• Least access: the agent gets the folders, accounts and channels its job needs, nothing more.
• A sandbox: Docker or an isolated server, so a wrong command cannot touch anything else.
• Logs and a human: every action is logged and someone reads them until the job has run cleanly for weeks.
• No secrets in prompts: credentials live in the environment, not in instructions the model can repeat.
• Your infrastructure, your keys: the agent runs on your server with your provider account, so you can turn it off or take it over at any time.