This is the third in a three-part series on reducing the cost of AI-powered marketing workflows. Part 1 introduced the issue of token consumption and the need to keep the context in check. Part 2 explained the architecture behind this approach. This walks you through installing Hermes Desktop, connecting to an AI model, creating your first skill, and running your first task.
Hermes Desktop works on macOS, Windows and Linux. Download the latest version from Agent Hermes website and install it as you would any desktop application. The installer includes the desktop interface and agent runtime, so there’s nothing else to install before you get started.
The first decision Hermes asks you to make is where to store your local context.
The context store contains conversation history, tool outputs, embeds, and other information that the agent accumulates over time. By default, Hermes creates it in your home directory, but you can change the location during setup or later in the application settings.
For individual use, the default position is generally sufficient. If you work with a team, you might choose a synced cloud folder, shared network drive, or dedicated local storage so your data is backed up and easier to manage.
Unlike most AI chat applications, this information remains under your control. The LLM provider receives only the information necessary to respond to each request. Your complete work history remains in your archive.
10x your SEO with Semrush for business.
The most powerful SEO platform in the world, created specifically for businesses.
Request a demonstration
Connect a template provider
Next, connect Hermes to an LLM provider. Hermes is vendor agnostic, meaning it can work with OpenAI, Anthropic, Google, Meta templates, self-hosted templates, or any compatible API endpoint. During setup, you will choose a provider and provide the appropriate API credentials.
The simplest option is OpenRouter. An OpenRouter API key grants access to dozens of templates through a single account, allowing you to switch between providers without changing your workflow. Create an OpenRouter accountgenerate an API key, then paste it into your Hermes provider settings. The free tier is sufficient for experimentation.
If you already have an API key for OpenAI, Anthropic, Google Gemini, or a self-hosted model like LLaMA, you can enter those credentials instead. Once the connection is established, select the model you want Hermes to use as default.
Create your first skill
Skills are one of the central concepts at Hermes.
A skill is a reusable reference document that the agent can consult. Typical marketing competencies include brand voice guidelines, editorial style guides, messaging structures, campaign briefs, audience personas, competitive positioning documents, and standard operating procedures.
There are two ways to create a skill.
The simplest is to use the /learn command within the chat interface. Provide Hermes with the reference material you want it to learn—for example, upload your brand’s voice guide or paste its content into the conversation—and ask it to create a skill. You can write a message like:
/learn Create a reusable skill from these brand voice guidelines for future email campaigns.
Hermes analyzes the material, converts it into a structured skill and automatically saves it in the skills folder.
The second option is to create the skill yourself. Write the reference material as a Markdown document and place it directly in the Skills folder of your file system. The next time you launch Hermes, or after refreshing the Skills panel, the new skill becomes available for use.
In the Desktop application, the Skills panel displays all available skills. Enable the ones you want the agent to use for the current task and disable the ones that aren’t relevant.
Over time, your skill library becomes one of your most valuable assets. Instead of repeatedly copying brand guidelines or campaign instructions in each tip, Hermes reuses the appropriate expertise each time it applies.
Run your first task
Once your template provider is connected and at least one skill is enabled, you are ready to use Hermes.
Open the chat interface and enter a simple marketing task, such as:
Create a draft five subject line email for our next product launch using the Brand Voice skill.
Hermes loads the relevant skill, combines it with your request, sends the prompt to the selected template and returns the completed draft.
Continue to refine the output through regular conversation. Because conversation history is stored in the local context store, Hermes maintains context throughout the session without requiring repetition of previous instructions.
Hermes can do more than generate text.
Depending on the tools you have enabled, you can ask it to:
- Search the web.
- Parse CSV files.
- Generate images.
- Read local documents.
- Call external APIs.
- Run the Python code.
- Automate multi-step workflows.
The experience is the same regardless of the activity. Interact through a single chat interface while Hermes coordinates the necessary tools behind the scenes.
Check where your data is stored
An advantage of Hermes is that your work data remains accessible.
Open the selected folder as a context archive and you will find saved conversations, tool outputs, embeds and other supporting files. The skills folder contains reusable reference documents that you have created or imported.
Because these resources reside on your system rather than within a provider’s application, changing template providers does not affect them. If you decide to move from OpenRouter to Anthropic or from a hosted model to a self-hosted LLaMA deployment, your skills, conversation history, and context remain intact. You just need to update your provider settings.
Scalability beyond the desktop
Hermes Desktop is designed to help people get started quickly, but the architecture itself is scalable for larger deployments.
The runtime can also work via a command line interface, inside Docker containers, or as a remote API server serving an entire team. The skills, context store, and workflows created in the desktop application are transferred directly into those environments.
For most marketers, the desktop application is the easiest place to start. Once you’ve built a useful library of skills and experienced how local context reduces repeat requests and unnecessary token usage, you’ll have a solid foundation for deciding whether broader deployment makes sense for your organization.
