01What Is the Hermes Agent Update?

Hermes Agent update is a major release of the open-source AI agent framework that adds new features to make AI agents more powerful and easier to use. In this episode, Julian Goldie explains how the update lets you reuse your existing AI subscriptions, use a huge memory, and control your desktop with multiple AI models.

The update changes how AI agents work by turning your subscriptions into real developer tools. You no longer need separate API keys for every tool. Instead, Hermes connects your existing Claude Pro, ChatGPT Pro, or SuperGrok accounts to all your coding and automation tools.

02Top Features in the New Update

The update packs several big features. Here are the main ones Julian highlights:

  • Local OpenAI-compatible proxy – Type hermes proxy and any tool that expects an OpenAI API can use your existing subscriptions.
  • XIGROK support – Get a 1 million token context window, so the model remembers your entire codebase or long work session.
  • Computer use for multiple models – GPT, Gemini, and Grok Vision can now control your desktop, not just Claude.
  • Vision upgrade – The model now receives actual images, not text descriptions, for better accuracy.
  • 180x faster browser tool – Keeps a persistent connection to Chrome, so agents don't reconnect every time.
  • Microsoft Teams integration – Reply in channels, send DMs, and use Microsoft Graph APIs.
  • Quality-of-life additions – Native Windows support, one-command install, slash handoff, and native X search.

03How to Set Up the Hermes Agent Update

Getting started is simple. Julian shows that you can have everything running in about 30 seconds. Here are the steps:

  1. Install Hermes Agent with one command: pip install hermes-agent.
  2. Run the proxy command by typing hermes proxy in your terminal.
  3. Hermes creates a local endpoint that looks exactly like the OpenAI API.
  4. Connect your existing AI subscription (like Claude Pro, ChatGPT Pro, or SuperGrok) to that endpoint.
  5. Point your tools (Ada, ClientContinue.dev, Codex CLI, etc.) to the local endpoint.
  6. Done. All your tools now use your subscription without extra API keys.

For Windows users, the update adds native support, so no workarounds are needed. The install command works the same way.

04How to Use the New Features in Real Workflows

Julian walks through practical examples of using the update. With the proxy, you can have Ada open for coding help, Client for file edits, and Codex CLI for terminal tasks. All these tools talk to the same AI through your existing subscription, with one setup.

The 1 million token context changes debugging. Instead of re-explaining your code, you dump the whole project into the AI. It sees everything and connects dots between files you didn't think were related.

For computer use, you can tell an AI agent to open your browser, find data, copy it, paste it into a spreadsheet, and format it. You can switch models mid-session with the /handoff command, starting with one model and moving to a more powerful one without losing context. The browser tool now runs 180 times faster, so web scraping and automation are quicker.

In Teams, you can have an agent in a channel that answers questions, pulls data, summarizes threads, or starts tasks when someone sends a specific message.

05Results and Comparisons from the Episode

Julian notes several performance improvements. The browser tool is now 180 times faster because it keeps a persistent Chrome connection instead of reconnecting every time. He compares the old and new vision handling: before, images were converted to text descriptions; now the model gets the actual pixels, which improves reading text and understanding layouts.

He also compares the old memory limitations with the new XIGROK 1 million token window. Most AI models forget context after a few minutes, but with XIGROK, the model holds an entire codebase or multi-hour session in memory, reducing repetition and the need to re-explain things.

For computer use, the update expands beyond Anthropic. Now GPT, Gemini, and Grok Vision can control the desktop, offering flexibility to pick the best model for each task.

06Tips, Limits, and Mistakes to Avoid

Watch out for these points from the episode:

  • Don't rely on the proxy for everything – It mimics the OpenAI API, so tools expecting that connection work, but check compatibility with your specific tools.
  • Understand context limits – Even 1 million tokens isn't infinite. For huge projects, you still need to manage what the model sees.
  • Use the handoff command wisely – Switching models mid-session is powerful, but make sure each model is suited for the part of the task you hand to it.
  • Consider security – Computer use lets AI control your desktop; be careful with sensitive data on your screen.
  • Stay updated – The browser tool improvement is huge, but persistent connections can fail if Chrome crashes or updates.

Julian emphasizes that these features stack, so test them together to get the most benefit. Start with the proxy, then add XIGROK and computer use gradually.

07Who the Update Is For and Why It Matters

This update is for developers, business owners, and anyone who uses AI tools daily. If you already pay for Claude Pro, ChatGPT Pro, or SuperGrok, the proxy saves you from buying extra API credits for each tool.

For agencies and teams, the Microsoft Teams integration is a big step. It turns Hermes from a personal tool into a team-wide system. A single AI agent in a channel can help everyone, not just the person who set it up.

The 1 million token memory and computer use make it possible to handle large codebases and automate complex desktop tasks. Julian says this is a big jump in one update, from an interesting side project to something that can run your daily workflow end to end.

If you want to automate time-consuming tasks, improve coding efficiency, or manage team communication with AI, this update gives you the tools to do it.