# Questions and answers about Agent Herman

What can Agent Herman help with, and how does it work? Here you will find plain answers about everyday life with him — and more on the technology if you are curious.

Canonical URL: https://agent-herman.se/en/faq/

## Getting started with Agent Herman

You do not need technical knowledge to get started. Here we take the most common questions from the beginning.

### What is Agent Herman, really?

Agent Herman is a digital colleague with specific responsibilities in your organisation. He can prepare a news item for your website, compile information or transfer information between systems. We bring the tools together and adapt the setup to the way you work. You decide what he should help with and which information he can access.

### Do I need to be technical to use Agent Herman?

No. You should be able to describe what you need help with in plain English. We help with installation, settings, connections and getting started. We can also go through the way you work and suggest tasks that are worth simplifying or automating. You need to know your business. We take care of the technology.

### What can he help us with in practice?

The website, research, customer information, documents and marketing, for example. He can prepare news items and social posts, compile material or put together a report for the morning meeting. With the right connections he can also move information between your systems. We start with a few clear tasks that make a difference for you and build from there. What he can do depends on the tools and access in your installation.

### How does it work when I want to change something on the website?

With our Google Chrome extension you can point at part of the website and describe the change in your own words. For example: “Change this heading and add our new opening hours.” Agent Herman then receives the instruction together with the place you have marked. The work happens through the connection we have set up to the website, and publishing takes place with human approval.

### Do we have to replace the programs we already use?

No. We start from the tools you already have and look at what can be connected. That might be the website, the accounting system, the customer records or files in OneDrive, for instance. Sometimes a connection already exists; sometimes we need to build one. And if you want to change something, we can help you find open alternatives that suit the business.

### Does Agent Herman decide for himself what he is allowed to do?

You decide which tasks he has and which systems he may use. Some tasks can run on an agreed schedule, while others should be left as suggestions for review. Publishing and sensitive changes need clear approval. He can also misunderstand things or make mistakes, so we adapt the way of working to the task and how important it is. It is a good idea to start by letting him prepare material.

### Is Agent Herman meant to replace ChatGPT or Claude?

No. Do carry on using ChatGPT, Claude or other tools just as before. Agent Herman instead has agreed tasks in the business and can work in the background even when nobody is at the computer. He can, for example, compile a report overnight or prepare the next publication. You can chat with him whenever you want to give him a job or follow up on the work.

### Does all the work have to live in a chat?

No. The chat is an easy way to give instructions and ask follow-up questions, but results can be saved where you actually work. A report can go into your shared documentation, a draft onto the website and customer details into the customer records. That makes the work easier to find and use, including for colleagues who were not part of the conversation.

### Can we decide where our data is processed?

Yes. We help you choose where Agent Herman runs and which services he uses. An AI model can run on your own hardware if you want to keep the AI processing itself in-house. If you use external AI services or other cloud services, information may need to be sent to them. We review the whole setup with you, based on the information you work with.

### What are we paying for?

You pay for our help and the services you choose: installation, adaptation, operation and support, for example. Servers, AI usage and certain external services may also cost money. We help you choose a reasonable level for your needs and budget. What is included in an existing Webbproffs package is set out in [the offer](/en/#oppenhet) and in your agreement.

### Do we have access to our own installation and our data?

Yes. You can have full access to your own server environment and retrieve and use your information yourselves. Agent Herman can also help export the information he can access, using the connected tools. We add no restrictions of our own on access or export. The programs and connected services remain subject to their respective licences and terms.

### Why do you build on open technology?

Because many people and companies contribute improvements that others can benefit from. We can draw on developments across several projects and choose what suits your organisation. As new possibilities emerge, we do not have to build every feature ourselves. That makes it easier to develop your solution over time and gives you more freedom to choose your tools.

### How do we know where to start?

Start with something that takes time or tends to be left undone. That might be updating the website, looking up information or compiling a recurring report. We go through the way you work and choose a first clear task together. You do not need a finished plan or to know which programs are required. That mapping is part of the help we can give you.

## The technology behind Agent Herman

Want to know how the parts fit together? Here we take a closer look at the agent, integrations, AI models and control over your environment.

### What parts is Agent Herman built on?

The basic recipe combines Hermes Agent as the executing agent with LibreChat as the user interface, usually in an Ubuntu Server environment. In between sit the adapter and integration parts that the installation needs. Search can be done with SearXNG, and the business systems are connected as required. The AI model is a separate part. That makes it possible to choose model, interface and business tools without treating everything as a single locked product.

### Why do you use both Hermes and LibreChat?

Hermes carries out tasks using the AI model and the connected tools. LibreChat gives users a shared interface for instructions and conversations. Our integration connects them and handles how requests and responses pass between the two. Keeping these roles separate makes it easier to adapt the user experience and the workflows independently, while benefiting from development in both projects.

### How is Agent Herman connected to our systems?

Through the interfaces the systems actually offer: APIs, MCP servers or a custom adapter, for example. MCP is a shared way of making tools available to the agent. For code-based websites, a code repository and a build pipeline can be the right route. Every connection needs the right authentication, access and error handling. We therefore distinguish between a finished integration and a possible connection that first has to be built and verified.

### Can we create our own skills and workflows?

Yes. Skills are reusable instructions that describe how the agent should carry out a particular kind of task. That could be following your publishing routine or compiling a report in a fixed format. You can write your own or get help from us. A skill does not, however, automatically give access to a system: the tools and permissions also have to be in place. External skills should be reviewed and tested before they are used in the business.

### Can he also work with web design and web development?

Yes. With the right tools and access, he can work with code, content and layout and prepare new features. We choose an approach suited to the platform, such as Joomla, WordPress, Astro or Shopify. The Chrome extension sends the instruction and context from the selected part of the page. The change itself is made through the website's connected tools. Code changes should go through version control, a test environment and review before publication.

### Can Agent Herman help with system administration?

Yes, an installation can be equipped to read logs, check services, compile operational information and carry out limited actions. For example, he can investigate why a local model service is not responding and suggest the next step. Exactly what he may change needs to be governed by accounts, permissions and workflows. We distinguish between investigating a problem and being authorised to restart or change a production environment.

### Are we tied to a particular AI model?

No. We build so that you can choose between models through the providers and model interfaces that the current installation supports. That could include GPT models, Claude, DeepSeek, GLM or a locally run model, for example. The choice affects cost, answer quality, response time and how well tool use works, among other things. A shared interface is no guarantee of equivalent function; we test the model against the tasks you actually want carried out.

### Can the whole solution run with local AI?

It is possible to build a solution where the model and the central services run on your own infrastructure. The model, the hardware and the tool support need to suit the tasks. That does not, however, make external integrations local: web search, cloud storage or social media can still involve external calls. For a workflow that has to run without external services we need to go through all the parts and their dependencies. Local operation also has costs for hardware, energy and maintenance.

### What do you mean by a separate environment and control over data?

That the agent and its information can be managed in a separate installation for your business, rather than everything having to sit in our shared product environment. Access to files, accounts, integrations, logs and backups still has to be configured. A separate environment is therefore an architectural choice, not a general promise that no data leaves it. We map out which parts run with you and what information any external services process.

### How does Agent Herman work with internal knowledge and research?

Internal documents, routines and project information can be made available through approved connections to a documentation space or wiki, for example. For current information on the web we can use SearXNG or other search services. Searching and fetching the full content of web pages are separate functions and can require different tools. Results can be written back as reports with sources in the business documentation, so that the knowledge remains outside the chat.

### How do we benefit from development in the different projects?

We keep Agent Herman's own code as small as possible and let features grow through Hermes, LibreChat and connected tools. As those projects improve, we select, test and introduce changes that suit your installation. We build our own components for clear needs, such as the Chrome extension, and keep them separate and optional where possible. This simplifies maintenance and lets your solution grow with your organisation. Versions and integrations still need to work together; upstream changes are not applied automatically.

### Is everything open source, and can we manage it ourselves?

Hermes and LibreChat are open source under the MIT licence. Other components can have different terms: LibreChat's database, MongoDB Community Server, uses the SSPL, while Outline uses the BSL. Those two licences are not OSI-approved open source licences. You can manage your own installation within the applicable licences and service agreements. We document the components and how they fit together, giving you a clear basis for operation and further development. See the [building-block overview](/en/#din-Agent Herman) for more details.