Video: "Hermes Agent: Automate Anything!" by Julian Goldie on YouTube.
What Hermes actually is
Hermes is a free, open-source AI agent from Nous Research. You run it locally, point it at whatever model you prefer — Gemma 4, Hermes 3, or anything else available via Ollama — give it a goal, and it works through the task in a loop. Plan, execute, check the output, correct, move on. It maintains a skills library so it reuses approaches it has already solved. It logs every step so nothing happens inside a black box. At free, local, and transparent, the combination is genuinely rare in this space.
The v0.13 Tenacity release added the /goal command, which keeps Hermes on a single objective across a long session rather than drifting to adjacent problems when the task gets complicated. That change made the tool meaningfully more reliable for multi-step jobs.
What the automation covers in practice
The tasks Goldie demonstrated fall into a few clear categories. Content: give it a keyword cluster and a content structure, it writes drafts and saves them to file. Research: give it criteria, it scrapes, filters, and writes results to a spreadsheet. SEO maintenance: give it a site URL, it checks internal links, flags thin pages, and lists suggested fixes. These are actual business tasks, not toy examples. The catch is that each one works reliably only when the goal prompt is specific enough — vague goals produce wandering outputs, same as any AI tool.
Worth noting: running Hermes with a free local model like Gemma 4 via Ollama removes the API cost entirely. That makes it viable for high-volume, repetitive tasks where you'd otherwise be watching a token meter tick up.
Where it earns its keep
Repetitive, structured work suits Hermes well. A nightly run that compiles competitor price changes from a set of URLs, a weekly content digest assembled from a set of sources, a bulk meta description rewrite across a product category — these fit the loop-based execution style naturally. The work gets done while you're doing something else. That is a real and useful shift, particularly for small teams where nobody has time to do the routine SEO housekeeping that accumulates over months.
Where it still needs a human
Hermes is not a replacement for knowledge-worker judgement. Quality control on content, decisions about what to publish, anything requiring client communication — these still need a person in the loop. Running it unsupervised on tasks with external consequences (sending emails, modifying live databases, updating published pages without review) is not advisable at this stage. The logging and observability tools have improved considerably in recent versions, but review before anything goes live remains the right approach. In practice, the sensible split is: let Hermes do the grinding, let a person make the calls.
Where this connects to NordSys
We configure Hermes Agent for clients who have a clear backlog of repeatable tasks and want to take the manual effort out of them. That means the full setup — install, model selection, skill configuration, and the goal prompts that match your actual workflow, not generic tutorial examples. If you have tasks that are eating time every week and follow a consistent structure, they're worth discussing. See our AI Agents service for what that looks like.
See our AI Agents service →