Wiki · Start here
Frequently asked questions
Short answers to what people ask most — with a link to the page that covers it fully.
Before you start
What is Hermes Agent?
A program that runs on your server and works as an assistant you brief in plain language. It can run commands, work with files, search the web, remember context and run on a schedule. More in What is Hermes.
Do I need to know how to program?
No. You write briefs in ordinary language, and any command that has to be pasted into the server comes ready-made — you paste it and press Enter. The agent verifies each step. Start with installation.
What does it run on?
A Linux server of your own (a VPS). Your own computer works too, but a server has an advantage: it is always on, so the agent can reach you at night and scheduled jobs do not stop when you close the laptop. See Hosting.
How long does it take to set up?
The slowest part is creating an account with a model provider and pasting the key (minutes). The installation itself is a handful of steps, walked through one at a time, each of them verified.
Money and data
How much does it cost?
The program itself is free and open source. You only pay for using a model — that is, for how much you talk to the agent. With a model of your own on your own machine, you pay nothing. Prices and recommendations: choosing a model.
Is my data sent anywhere?
Requests go only to the model provider you choose yourself. There is no telemetry or
usage collection, and conversations, memory and skills stay on your server in ~/.hermes/.
The only thing that leaves is the text you ask about.
Can I use it offline?
Yes — with a model running on your own machine (Ollama, for instance). The agent is pointed at your address and needs nothing from outside. Expect a local model to be slower and less capable than paid cloud models, though.
Using it
How do I control it from my phone?
Through Telegram or Discord — the agent behaves the same as in the app, so you can write from your phone while away from home. Guide: Telegram and Discord.
Can several people use one agent?
Yes. Through connected platforms several people can talk to the same agent at once; who may write is decided by an allowlist of accounts, plus pairing for direct messages. Details in Security.
Can it work on its own while I sleep?
Yes — that is scheduled tasks. A brief can arrive in the morning, a check can run at night, a website can be watched every hour.
Memory and knowledge
Does it remember what we did together?
Three things differ: the live conversation (everything you have written), memory (a short list of essentials the agent always carries) and past conversations (searchable without limit). Explained in Agent memory.
What is the difference between memory and a skill?
Memory holds facts — who you are, how things work for you, what you dislike. A skill holds a procedure — how to do something step by step. Memory is always in play; a skill is pulled out when a fitting task appears: Skills.
Does it learn anything on its own?
Yes. From work that went well it can write itself a procedure and reuse it next time. You can also say
/learn and hand it documentation or a web page to write the procedure from.
Safety and operation
Is it safe?
The agent carries six safeguards — from the account allowlist to approving dangerous commands. Before running something that deletes or overwrites, it asks you. The full overview is in Security.
What if something breaks?
First find out what happened (hermes logs) and have the installation checked
(hermes doctor). Most trouble has one of five causes — see
When something breaks.
How do I stop it completely?
Scheduled jobs stop with hermes pause (hermes resume starts them again).
To stop the agent entirely, stop the service on the server — running conversations end, while saved
conversations, memory and skills remain.
Where do I ask when I am stuck?
In our community Discord — that is where answers come fastest. The official documentation lives at hermes-agent.nousresearch.com/docs.