Hermes Agent: Features and Tips for Non-Technical Users

This guide picks up where the Hermes installation guide leaves off. If you have not installed Hermes yet, start there — or if the terminal is new territory for you, see our guide on using ChatGPT to walk you through the install. Once Hermes is running, the features below are what separate a basic setup from one that genuinely adapts to you.

Give your agent a memory of you

Hermes keeps notes on your conversations automatically in a file called memory.md. This updates as you use the agent, recording key details from your sessions so it does not have to ask the same questions twice. But there is a second file — soul.md — that you write yourself, and it is the single most impactful thing you can do to improve the quality of Hermes's responses.

soul.md is a plain text file that you save in your Hermes folder. It can contain anything you want the agent to know about you as standing context: your name, location, what you do for work, your current projects, how you prefer to communicate, whether you want the agent to challenge your thinking or just carry out instructions. Think of it as a briefing document that Hermes reads before every session.

A minimal soul.md might look like this:

Name: Sam
Location: Edinburgh, UK
Work: independent marketing consultant, mainly B2B SaaS clients
Communication style: direct, no filler, happy to be pushed back on
Current focus: building a repeatable content process for client onboarding

With this in place, when you ask Hermes about scheduling a client call or what to prioritise this week, it has background it would otherwise have to ask for each time — or worse, answer without.

Connect meetings, notes, and email

Hermes supports integrations — connections to external services via MCP (Model Context Protocol) connectors. Three connections have an outsized effect on how useful the agent becomes day to day.

Your notes (Obsidian or similar). If you use Obsidian to take notes, you can give Hermes a path to your notes folder. Once connected, you can ask questions like "what did I write about the Henderson project?" or "pull out all my notes on client onboarding" — and Hermes searches your actual notes rather than guessing. You can also connect other local note systems; the key is giving Hermes a folder path it can read.

Your meeting recordings. Tools such as Granola, Fireflies, and Fathom produce transcripts of your meetings. Connecting one of these to Hermes means you can ask "what did we agree in Tuesday's call?" or "who said they would send the proposal?" without digging through a recording yourself.

Your email. Gmail and other email services can be connected via MCP connectors, including through tools like Zapium. This is where the security note below applies — read it before setting up email access.

Background tasks and scheduling

By default, sending Hermes a new message while it is working on something stops the first task. The /background command lets you start a second task that runs in parallel without interrupting the first. Both tasks run in separate threads and report back independently when done.

This becomes useful when you want Hermes to research two unrelated things at the same time, or when you want to keep a long-running task going while asking a quick question about something else.

For scheduling, Hermes accepts natural language: "every morning at 8 a.m., send me a brief with today's weather, my calendar for the day, and three things I should focus on." You do not need to know how scheduled tasks work technically — Hermes handles the setup. These scheduled tasks can become increasingly useful over time because they draw on everything Hermes has accumulated about your work and goals.

The /goal feature — set a destination, not just a task

The /goal command is what makes Hermes an agent rather than a chatbot. Instead of asking a question and receiving a single reply, you give Hermes a goal and it works autonomously until it achieves it — checking its own output, calling tools, iterating — without requiring you to keep sending follow-up messages.

The critical rule for using /goal well: be specific. "Help me grow my business" will produce vague output. "Research the top three pricing models used by freelance designers in the UK, write a one-page summary of the pros and cons of each, and save it as pricing-research.md" will produce something you can actually use.

For longer goals that involve multiple stages — some handled by Hermes, some requiring your input — you can structure the goal as a numbered list of steps. For example: "1. Research five competitors and list their main features. 2. Pause and wait for my review. 3. Based on my feedback, write a positioning statement." The numbered structure tells Hermes where it works and where it hands back to you, which prevents it from racing ahead past a decision that needs a human judgement.

Use /stop at any point to halt a running goal. Use /resume to pick up a previous session. Use /steer [instruction] to redirect Hermes mid-task without stopping it entirely.

Switch models for different jobs

Hermes can connect to multiple AI model providers and switch between them with the /model command. This matters both for cost and for quality, because different models have different strengths.

The cost argument for switching: sending a simple research task to an expensive model when a cheaper one is adequate burns money unnecessarily. Hermes lets you tag the right model for each task either by using the /model command or by specifying the model directly in your prompt — "use Grok to search for recent discussion of this topic on X" or "use MiniMax for this because the document is very long."

You can also create named tools — sometimes called a Pantheon in the Hermes community — where each tool has a defined model, system prompt, and purpose. This is useful if you want a "deep reasoning" tool, a "quick drafting" tool, and a "research assistant" tool, each pre-configured and accessible by name.

Back up your agent

Everything that makes your Hermes setup useful — your soul.md, conversation memory, custom tools, and task history — lives in a local folder on your computer. If you get a new machine or your drive fails, that is gone unless you have a backup.

The standard approach is to back up your Hermes folder to a private GitHub repository. GitHub is a code storage platform, but you do not need to be a programmer to use it for this purpose. A private repository is a secure, off-device backup location where nothing is publicly visible.

You can ask Hermes to set this up: "Create a daily backup job that commits everything in my Hermes folder to my private GitHub repository." If you do not have a GitHub account, create one at github.com — the free tier is sufficient. Hermes will walk you through the authentication steps.

Ten commands worth knowing

Most of what Hermes does can be triggered in natural language without knowing any specific commands. But these ten are worth having to hand:

For everything related to keeping your Hermes setup secure — where to run it, what file access to grant, and how to handle API keys — see the Hermes Agent security guide.