Hermes Agent setup: install the self-improving AI agent, the full guide.

What is Hermes Agent? Key features

Hermes Agent is an open-source AI agent built by Nous Research. Unlike a chatbot in a browser tab, this AI assistant runs as a service you own, connects to the channels your team already uses, and acts. The headline features: it's a self-improving AI agent, the agent learns and grows across sessions, saving new skills as it works. It's multi-channel, Telegram, Discord, WhatsApp, Slack, and the terminal. It does tool-calling for web search, file operations, image generation, and text-to-speech (TTS). And it handles multi-step, scheduled tasks so it can run a job every morning without you.

Because Hermes Agent is an open-source project (MIT-licensed), you can download Hermes, read the code, and run it free. Hermes uses whatever LLM you point it at, so you're never locked to one vendor. This guide walks the full Hermes Agent setup from prerequisites to a running agent.

System requirements and prerequisites

Before you install, you'll want a machine running Linux, macOS, or Windows via WSL, with Python, Node.js, and Git available. You also need access to an LLM, either a hosted provider (an API key) or a local model. Hermes Agent requires a model with a usable context window; a model with at least 64k tokens of context is the comfortable floor so the agent can hold a conversation plus its skills and memory. For a development setup, your laptop is fine. For 24/7 operation, a small VPS is the move. Docker is optional but makes the install cleaner.

How to install Hermes Agent

There are two common paths. The fast path: run the one-line installer from the project's GitHub, which pulls the code and launches a setup wizard that walks you through config. The manual path: clone the repo with Git, install dependencies with your package manager (Python and Node), and copy the example config. Either way, the CLI is how you'll interact with the installer and, later, run the agent. On a fresh Linux box or via WSL on Windows, a basic install hermes agent run takes about 30 minutes. The full Hermes agent setup, the part that makes it useful, is the configuration that follows.

Configure an LLM provider (API key or local model)

Hermes Agent requires a model, and you choose where it comes from. The easiest option is a hosted provider through OpenRouter: create an account, generate an API key, and paste it into your Hermes config. That gives the agent access to a range of frontier models behind one API. If you'd rather keep everything local, point Hermes at an Ollama install running a local model on your own hardware, no data leaves the box, at the cost of needing a beefier machine. Set your chosen hermes model in the config, save your api key as an environment variable, and the agent is ready to think.

Connect Telegram, Discord, and WhatsApp

This is the step that makes Hermes feel alive, you message the agent from your phone. To connect Telegram, create a bot with BotFather, copy the bot token into the config, and restart; the agent now answers in Telegram. Discord works the same way: create an application, add a bot, paste the token. WhatsApp and Slack follow the same gateway pattern, a token plus a short pairing step. Hermes routes every channel through one tool gateway, so the same agent, memory, and skills are available no matter where you message it. On Android, a Termux guide lets you run the whole thing from a phone if you want.

Run Hermes Agent 24/7 as a service

Running Hermes in your terminal is fine for testing, but for a real assistant you want it always on. Deploy it to a VPS, run it under Docker or a process manager, and use cron for scheduled tasks, the "every Monday at 7am, pull last week's invoices" kind of job. A lightweight dashboard shows you what the agent has been doing. Once it's running as a service, Hermes handles messages, scheduled jobs, and multi-step work around the clock without your laptop being open.

Using Hermes: skills, tools, and automation

With the setup done, the agent earns its keep. Ask Hermes to do something new and it figures out the steps, then saves the pattern as a skill, that's the self-improving part. Built-in tools cover web search, file operations, a cloud browser, image generation, and TTS. Because it's an agent that learns and an agent that grows, the value compounds: the Hermes you use in month three knows your tools, your terminology, and your data far better than the one you installed on day one. Tool-calling plus memory across sessions is what separates a real AI agent from a chat window.

The terminal and the gateway: how Hermes is wired

Two pieces sit at the center of Hermes. The terminal is where you run and watch the agent during setup, a few bash commands start it, check its status, and tail its logs, and the CLI stays your control panel for the whole agent. The gateway is the layer that routes messages: every channel (Telegram, Discord, WhatsApp, and the terminal itself) flows through one tool gateway, so the agent's memory and skills are identical no matter where a message comes from. Understand those two, the terminal you drive it from and the gateway that connects it, and you understand most of a confident full setup.

A typical Hermes Agent setup, end to end

Putting it together, here's the full Hermes agent setup most people follow. Spin up a Linux VPS. Install the prerequisites, Git, Python, Node. Run the installer from GitHub and let the setup wizard write the config. Add your OpenRouter api key, or wire up a local Ollama model if you want everything on your own hardware. Start the agent from the terminal with a bash command and confirm it answers in the CLI. Connect Telegram with a bot token so you can reach the agent from your phone, then add Discord and WhatsApp the same way through the gateway. Finally, put it under Docker or a cron-managed service so it runs 24/7, and set up your first scheduled tasks. That's nothing to a self-improving AI assistant in an afternoon.

What businesses actually build with Hermes

Once the setup is done, the useful part begins. Small businesses use Hermes Agent for a handful of jobs: an agent that drafts and sends follow-up emails on a schedule, a Telegram bot that answers staff questions from internal docs, a Discord agent that posts a daily numbers summary, and a terminal workflow that reconciles invoices every Friday. Each is a skill the agent keeps and improves. Because Hermes uses your chosen model and runs on your infrastructure, the same setup scales from one skill to dozens without new per-message licensing, an open-source AI agent that grows with the business.

Troubleshooting common setup issues

Most Hermes setup problems fall into three buckets. Model errors, usually a missing or wrong API key, or a model whose context window is too small; double-check the key and pick a model with at least 64k context. Channel not responding, almost always a bad bot token or a pairing step that didn't complete; re-copy the token and restart. Crashes on start, typically a missing dependency (Python, Node, or a system library); a clean Docker install sidesteps most of these. The GitHub issues and contributing guide are the best reference when you're stuck.

Or have us run the full Hermes Agent setup for you

The DIY install above is the easy 30 minutes. The reason businesses hire us is everything after the install: connecting Hermes to your CRM, accounting software, and calendar; building the skills for your actual workflows; configuring memory with your business context; and training your team to use it. That's the 21-day Installation, and it's backed by the 10-Hour Guarantee. You get the same open-source Hermes Agent, configured by people who've done it before, running against your real work on day 14.