Video: "Hermes Agent + Codex Browser Use Updates!" by Julian Goldie on YouTube.
What browser use actually means
Browser automation is not new. Selenium has been around for twenty years and Playwright is a staple for developers. What is new is tying browser automation to an agent that decides what to do next rather than following a script you wrote in advance.
The Codex browser use integration gives Hermes a live browser window it can control. It can navigate to a URL, read the page content, click a link or button, fill in a search box or form field, and then decide what to do with the result — all within a single agent run. You give it a goal, and it works out the web interactions needed to get there.
The kinds of tasks this opens up
The obvious use cases are research and data gathering. An agent that can search Google, read a competitor's pricing page, pull the numbers into a spreadsheet, and repeat that across twenty competitors is genuinely useful — and previously required either a developer building a custom scraper or you doing it manually.
Beyond research, the browser layer also enables form-based workflows. Submitting directory listings, posting to web interfaces that don't have APIs, filling in supplier portals — these tasks are awkward to automate with traditional tools but straightforward for a browser-capable agent. In practice, anything a person does inside a browser tab, an agent can now attempt.
How Codex fits in
Codex here refers to OpenAI's execution environment rather than the older code-generation model. It provides the sandboxed browser instance Hermes uses to run the web interactions. The combination means Hermes handles the reasoning and task management while Codex handles the actual browser execution — a clean split of responsibilities.
Worth knowing: you need an OpenAI API key to use this via Codex. If you are running Hermes entirely on free open-source models, this particular integration adds a paid dependency. There are also open-source browser automation options (Playwright, browser-use) that the Hermes documentation covers as alternatives.
Where it still needs watching
Browser automation is reliable until a website changes its layout, adds a CAPTCHA, or requires a login session you haven't configured. Hermes will attempt to handle these — sometimes successfully, sometimes not. For tasks where the source websites are stable and well-structured, the success rate is high. For consumer sites that change frequently or aggressively block bots, you will need to supervise more carefully.
Rate limits and politeness settings also matter. An agent that can loop through a hundred pages will do so as fast as the target server allows — which is not always what the site owner expects. Building in delays and respecting robots.txt is on you as the operator.
Where this connects to NordSys
Browser-capable AI agents are one of the more useful additions to a small business's automation toolkit, but getting them to work reliably on real tasks takes more setup than the demos suggest. Configuring the right model, scoping the tasks sensibly, and deciding which workflows are worth automating — that's the work we do with clients. If you want to explore what Hermes or similar agents could actually do for your business, our AI Agents service is the right starting point.
See our AI Agents service →