Video: "Hermes Agent: Swarms + Computer Use + New FREE Model" by Julian Goldie on YouTube.
What computer use adds to an AI agent
Most AI agents work through APIs. They can write to files, call web services, run code — but they cannot interact with graphical applications that do not expose an integration point. Computer use changes that. Hermes watches the screen (with your permission), interprets what it sees, and takes actions: clicking buttons, filling forms, scrolling through pages, switching between applications.
In practice that means tasks that previously required a human click — logging into a web console, copying data between tools that have no shared API, running a workflow inside a desktop app — can now be delegated to Hermes. To be fair, this works best on well-structured, predictable interfaces. A novel UI or a flaky web form will trip the agent up more often than it would trip a person. But for stable, repeatable sequences it is genuinely useful.
The background mode is what makes it practical
The specific feature Julian Goldie highlights is that computer use runs in a background session while you work on something else. The agent operates on a separate desktop window, so your screen is not taken over. That distinction matters: if computer use required your full attention to supervise, you could just do the task yourself. Running in the background makes it genuinely autonomous — set a task, get on with your own work, and come back to a completed job.
This is currently available on Mac. Windows support is in early development and is listed as a future release rather than something available now. If you are running a Windows-only setup, this particular feature is not ready for you yet.
Combining computer use with swarm mode
Swarm mode lets you run multiple Hermes agents simultaneously on coordinated tasks from a shared Kanban board — a planner, a builder, a reviewer, each handling their defined role in parallel. Add computer use to that and you can run multiple agents across different desktop tasks at the same time: one agent handling web-based research via browser, another processing documents in a desktop application, a third working through a checklist in a business tool.
The practical limit is your machine's memory and how many application sessions it can sustain without slowing down. On a modern Mac with 16GB or more of RAM, several concurrent background sessions are workable. Worth knowing: the new free model integration means you can run this stack without a per-token API bill if you route through OpenRouter's free tier — useful for testing the setup before committing to a paid model for production use.
The use cases that benefit most
For a small business that relies on web-based tools without good APIs — older CRMs, industry-specific portals, government services — computer use opens automation paths that were previously out of reach without custom browser extension development. Repetitive admin tasks are the best fit: the agent handles the click sequence while you work on something that actually needs judgement.
It also removes one common objection to AI agents in regulated environments. Tools that cannot expose an API for security reasons can still be automated by an agent that operates through the UI, in the same way a person would — without needing code-level integration.
Where this connects to NordSys
Setting up Hermes Agent's computer use feature takes more than hitting install. You need to map the desktop interactions, set permission limits, and test the background mode against your actual workflow — otherwise it will automate the wrong things. NordSys configures Hermes Agent for teams as part of our AI Agents service, including computer use where it makes sense for the business.
See our AI Agents service →