01What Is Hermes Computer Use Agent?

Hermes computer use agent is a free, open-source feature inside the Hermes agent tool built by Luce Research. It lets an AI model operate your Mac computer just like a person would—clicking, typing, scrolling, and dragging—all in the background.

This means you can keep working or doing something else while the agent handles emails, notes, files, and admin tasks for you. It does not move your cursor, change your keyboard focus, or switch spaces. You and the agent use the same machine at the same time, like a team.

02Main Features of Hermes Computer Use

  • Background operation: The agent works without interrupting you—your cursor and keyboard stay untouched.
  • Any AI model: You can use Claude, ChatGPT, Gemini, or even a free local model like Nemetron or Kimi K2.6.
  • Self-improving: Hermes tests its own abilities and learns, as seen when it verified permissions and enabled computer use itself.
  • Multi-layer guardrails: It asks for approval before risky actions like deleting files.
  • Two ways to use: Run the dedicated CLI command hermes computer use chat or use the regular Hermes CLI, even from your phone.
  • Mac-only: As of this update, it is only available on macOS.

03How to Set Up Hermes Computer Use

Setting up is straightforward. You don't need coding or technical experience. Here are the exact steps shown in the episode:

  1. Open a terminal and run hermes update to get the latest version and skills.
  2. Run hermes computer use install to install the CLI command.
  3. Alternatively, enable the tool set interactively inside Hermes.
  4. Grant OS permissions by going to Accessibility settings and allowing your terminal.
  5. Let Hermes configure computer use—it may even test itself to confirm it works.
  6. Choose a model that supports vision. Text-only models won't work because they can't see screenshots.

Once installed, you can start giving the agent instructions. The episode showed Hermes even pasting the setup instructions into itself to make the process easier.

04How to Use It: Real Examples Tested

The episode tested several real tasks. First, Julian asked Hermes to open the Notes app. It did so in seconds, while he was still talking.

Then he asked it to open Notes, create a new note, and journal about the best ways to save him time. Hermes brainstormed, opened the app, and created a note with 10 personalized ideas—all in the background.

Next, Julian tried a more advanced task: organizing his Obsidian knowledge base. He asked Hermes to add details about him, organize folders, add emojis, and improve the knowledge graph. The agent struggled and moved slowly, so Julian stopped it and switched to simpler tasks.

Finally, he had Hermes open Terminal, start a Hermes conversation, and say hello—which it did successfully, even getting a reply.

Julian used a framework for best results: give the agent a full goal (not a one-off task), hand it over to the computer, oversee rather than operate, and review the output. He also highlighted that you can switch models by typing model in the Hermes CLI or set up new models with hermes setup.

05Results and Performance in the Tests

The results were mixed but promising. For simple tasks like opening Notes and creating a journal note, Hermes worked flawlessly and quickly—Julian said he'd never seen computer use work that fast. The note was personalized based on his context, including his work in AI SEO.

For complex tasks like organizing Obsidian, it was notably slower and had difficulties. The episode mentioned that screenshots use a lot of tokens, so heavy usage can hit API limits. In fact, the Codex model ran out of usage during the tests, so Julian switched to Kimi K2.6.

Another result: Hermes successfully opened Terminal, launched a Hermes session, and typed "hello," getting a response—effectively talking to itself. Despite the limitations, the feature impressed Julian for everyday tasks.

06Tips, Limits, and Mistakes to Avoid

Here are key tips and warnings from the episode:

  • Start simple: Use it for quick actions like opening apps, writing notes, or summarizing emails. Avoid huge, complex jobs until the feature matures.
  • Give full goals: Instead of one-off tasks like "write an email," say "finalize emails, summarize what they need, draft a reply, and put it in drafts." Multiple actions in one prompt work better.
  • Watch API limits: Computer use consumes many tokens, so choose models with generous limits or use free local models.
  • Be safe: It's an experimental feature. Grant permissions only if you're comfortable, and rely on the guardrails that ask for approval on risky actions.
  • Don't expect perfection: It's still beta—slow on complex tasks. Oversee and review, don't just let it run wild.
  • Know the limits: It's Mac-only and requires good performance. Text-only models won't work because they lack vision.

07Who Should Use Hermes Computer Use?

This is for anyone who wants to automate repetitive computer tasks without being a coder. Julian emphasized that if you can type a sentence, you can give the agent a task. It's particularly useful for content creators, SEO professionals, and business owners who deal with lots of emails, notes, and files.

Julian also addressed common limiting beliefs: you don't need to be a tech expert, the agent won't take over your computer (it asks permission), and you're not late—most people haven't heard of this new release yet. Being early gives you an advantage.

If you want to see more, Julian mentioned a guide with 100 prompts, a 30-day roadmap, and a step-by-step operating procedure available inside the AI Profit Boardroom community. But the core tool itself is free to use, especially with a free local model or a free API like Quen3 from Nous Portal.

08Why Hermes Computer Use Matters

This update matters because it moves AI from giving answers to doing actual work. Most people use AI by copying and pasting back and forth, which is slow and keeps you at the keyboard. Hermes computer use changes that by letting the AI do the clicking, searching, and filing itself.

In the episode, it saved time within the first five minutes—opening an app, creating a note, and filling it with ideas while Julian did nothing. This is a major step up from before, as the new computer use models are far more capable than the previous version.

It also matters because it's free and open source, works with many AI models, and levels the playing field. You don't need expensive tools or technical skills to get started. Just install it, grant permissions, and you can start automating your day.