Hermes Agent is an open-source AI agent developed by Nous Research. It reads files, runs scripts, finds information, and completes tasks from start to finish on the platforms your team already uses. This article explains what Hermes is, how it works, and six use cases where businesses can put it to work. For each one, you see what it needs to run. For installation and support, see our page about the AI agent for your business.
Hermes Agent is an open-source AI agent. It works through chat channels or the desktop app and remembers project context between sessions. It saves procedures as skills and can run tasks at a set time. Instead of only answering questions in a chat, it can take action: read and edit files, run scripts, search the web, and verify results. Its code is open, so you are not tied to one vendor and can inspect how it works.
What happened in earlier sessions stays saved: decisions, project context, and agreed ways of working. That keeps later tasks consistent. For a closer look at the distinction, start with what an AI agent is or our comparison of Hermes vs ChatGPT and Copilot.
Hermes works with projects, knowledge, and tools. You open a project, give it a task, and the agent works out the steps it needs to take. You choose which tools and access it gets. Its actions stay within the permissions you configure.
| Step | What Hermes does | Result |
|---|---|---|
Read context | The agent reads the project's files and context | Hermes knows what the project does and how it is organized |
Use tools | The agent only uses tools you make available | Actions stay within the permissions you configure |
Complete the task | The agent completes the task and verifies its own result | You get a finished result, not a suggestion |
Remember decisions | The agent remembers decisions for later tasks | Later tasks become more consistent |
Hermes Agent also uses skills: reusable instructions the agent loads when a task calls for them. When the agent works out a new procedure, it saves the steps as a skill for next time. It also updates existing skills as it learns. Want to review that first? Turn on the approval step, and new or changed skills wait for your approval. With scheduled tasks, it runs a job at a set time, such as a daily summary or a weekly report. The result goes to your team's channel.
Hermes works on the platforms your team already uses, such as Telegram, Slack, Discord, WhatsApp, or Buzz. Individuals can also use Hermes Desktop, with the full toolset on their own workstation. For teams that need a shared environment, a shared Hermes installation often makes more sense: everyone talks to the same agent, with separate profiles and permissions.
Six practical business use cases, drawn partly from the official user stories and partly from our own use:
The agent reads an issue, investigates the codebase, writes the change, runs tests, and opens a pull request for human review. In the official user stories, a solo developer describes how Hermes handles most of the work on an entire app and files its own tasks and deploys. In the developer's words: "I mostly review things over Telegram."
Needs: access to the repository and the test setup. Trigger: an issue or a message from the team. Human check: a person reviews the pull request before it is merged.
Scheduled tasks read sources you choose each day, from market figures to industry publications, and deliver a short summary through your team's channel. In the same user stories, someone working in construction had Hermes summarize 82 messages from eleven WhatsApp groups into three lines.
Needs: read access to the sources you choose. Trigger: a daily schedule. Human check: a person decides what to do with the summary.
Weekly reports, status updates, and KPI summaries do not have to stay manual. The agent retrieves figures from the sources you connect, such as spreadsheets, a CRM, or project management tools, and puts them in the same format each time. We explain what that looks like in practice in automating recurring tasks with AI, and how the agent collects the figures in retrieving marketing data with an AI agent.
Needs: read access to the spreadsheet, CRM, or project tool. Trigger: a weekly schedule. Human check: a person checks the figures before the report goes out.
Hermes searches documentation and earlier tickets, drafts responses, and flags anything that needs confirmation. This keeps first-line support moving without letting the agent send messages to customers unchecked. For a guide to setting up least-privilege access for this work, read our article on AI agents in customer service.
Needs: read access to the documentation and earlier tickets. Trigger: a new support request. Human check: a person approves every answer before it is sent.
After completing work, the agent updates the documentation: decisions, procedures, and pitfalls are recorded in the project itself. That keeps knowledge from disappearing into separate chats and email threads. Our own Hermes installation at Voltti works the same way: the agent's work ends up in a readable project record. Read how to set one up in our article on an AI knowledge base the agent maintains.
Needs: write access to the project documentation only. Trigger: a finished task. Human check: the team reads the changes in the project record.
For research tasks, the agent searches multiple sources, compares claims, and produces a summary with references. Our own agent has read-only access to Search Console, but rewrites articles directly in the CMS and sometimes publishes them itself. We review the work afterward and make corrections where needed. Read more about that process in SEO with an AI agent.
Needs: web access and the sources you trust. Trigger: a question or a schedule. Human check: a person checks the references before the summary is used.
Do not start with the most impressive use case. Start with a task that already comes up often and has a result you can check. Think of a report that needs the same format every week, an inbox that needs a summary, or a check you now do by hand. Start small, track what the agent delivers for a week, and only then expand. The more an action can affect customers or money, the more approval steps you build in. A first pilot usually takes 2 to 4 weeks, from the first conversation to the evaluation. Read how that works in our approach from pilot to ongoing support.
Hermes Agent works best on work that keeps coming back and that you can check. For some tasks, an agent is more than you need, or it takes more work than you expect.
A one-off question or a text you only rewrite once is faster in a chat interface. ChatGPT or Copilot often handle that better. Setting up an agent pays off when a task comes up often.
An agent that can write to your systems needs upkeep: scoped permissions, updates, and someone who keeps reviewing the results. Factor that in before you start.
Tasks where every decision needs human judgment, such as a negotiation or a sensitive customer conversation, belong with a person. The agent can do the groundwork, but the decision stays with you.
Output quality and cost depend on the language model the agent runs on. For how those costs add up, read what an AI agent costs for a business.
Running everything locally is possible, but a local language model needs powerful hardware. In that case, first read which computer you need for local AI.
Hermes can do a lot out of the box, but a business environment needs more than an installation. Voltti provides:
Installation on a system at your business premises
Project, knowledge, and access configuration
Encrypted backups to a separate location on your network, and updates
Support for your team in Belgium and the Netherlands
Read more about what Voltti does. Want to try Hermes yourself first? Our guide to installing Hermes with the desktop app or the terminal walks you through both routes step by step.
Yes, the agent itself is open source and free to use. You pay for what surrounds it: installation, support, and the language model it runs on. A cloud model costs money per use; a local model needs hardware that can run it. Voltti provides installation and support as a service, and we always prepare a custom quote for your business.
No. ChatGPT provides a chat interface and can also perform tasks. Hermes is an agent you configure with tools to read files, run scripts, take actions, and verify results. It also remembers decisions and project context between sessions, and it runs in an environment you control. For the full comparison, read Hermes vs ChatGPT and Copilot.
The Hermes environment runs locally, on your own system. For the language model, it can use an external model through an API. Fully offline use is possible with a local model, but that needs powerful hardware, especially for larger models. For more on the choice between local and hybrid setups, read what stays local in a hybrid AI setup.
It is, with the right setup. The agent only gets the access a task needs, with minimal permissions per tool. Risky actions, such as sending messages to customers, wait for human review. Encrypted backups go to a separate location on your network, not to the system Hermes runs on. Read more about our approach to security and privacy.
Yes. With scheduled tasks, you tell Hermes in plain language when a job should run, such as every weekday at 8 a.m. or once a week. Each run starts in a fresh session and can deliver its result to a chat, a file, or your team's channel. That is how use cases 2 and 3 in this article work.
