01What is GLM 5.2 Hermes agent?

GLM 5.2 Hermes agent is a combination of the GLM 5.2 AI model and the Hermes agent system. It lets you create teams of AI agents that work together on tasks like writing, editing, and publishing content. In this episode, Julian Goldie shows how GLM 5.2 plugs into Hermes to build a multi-agent content machine.

You can set up separate profiles for GLM 5.2 inside Hermes. You can use it with the agent operating system, including a voice-activated version called Javis. You can also use the CLI directly.

The system lets you visualize all your projects and come back to them later.

02Main features of GLM 5.2 and Hermes

GLM 5.2 is a powerful model that Julian compares to Claude Opus 4.8. It scores 74.4 on Frontier SWE, while Opus 4.8 scores 75.1. It also performs well on Terminal Bench and is great for agentic tasks.

  • Multiple agent profiles – You can have different profiles for different tasks.
  • Kanban board integration – Agents work on cards like a Trello board.
  • Automated quality control – A judge checks every draft and sends weak work back.
  • Auto-publish – The system can deploy blog posts and videos to your website directly.
  • Low token costs – Using GLM's coding plan is cheap; Julian says he hasn't run out of tokens.

03How to set up GLM 5.2 in Hermes

Setting up GLM 5.2 with Hermes is simple. You can do it from your terminal or through the Hermes interface.

  1. Open your terminal.
  2. Type Hermes model to see available models.
  3. Go to ZAI and add the coding plan for GLM 5.2.
  4. Or go to your Hermes agents, click manage, then models.
  5. Change the model to GLM 5.2.
  6. Create a separate profile for GLM 5.2 if you want.

You can also set up a chat dashboard to switch between models for different APIs. Julian shows a dashboard with GLM via Olama and GLM 5.2 separately.

04How to use the content machine with GLM 5.2 Hermes

The content machine works like a relay. You drop in a topic, and agents handle the rest. Here's the flow Julian shows:

  1. Drop in a one-line topic onto the Kanban board.
  2. The keyword researcher finds the angle.
  3. The writer drafts a blog post.
  4. The editor sharpens it.
  5. The judge scores it and sends it back if it's not good enough.
  6. When it passes, the video builder creates a video.
  7. The system publishes both to your website.

Julian shows an example blog post published to aoprofitborn.com. The post itself talks about running GLM 5.2 in Hermes. He also shows a video that was created and edited entirely by agents using the free skill Hermes plus hyper frames.

The judge checks for facts and blocks fabrication, so weak work never gets published.

05Results and comparisons from the episode

Julian compares GLM 5.2 to Claude Opus 4.8. On Frontier SWE, GLM 5.2 scored 74.4, and Opus 4.8 scored 75.1. That's within 1.5 percent, which is very close.

He also notes GLM 5.2 is strong on Terminal Bench and good for agentic tasks.

The biggest difference is API cost. Julian says there's a huge difference in pricing, but you can avoid high costs using the coding plan. He built an entire agent operating system with a terminal, music player, and paint app using GLM 5.2 and Hermes.

He also built an open world game and other projects. The content machine produced a live blog post and a video. The blog post was fully designed and published automatically.

Julian shows website traffic that improved after using this system. He says SEO is working better than ever because of better quality control.

06Tips, limits, and mistakes to avoid

One big mistake is not using a judge. Without a judge, agents can produce generic or fabricated content. The judge blocks that and sends drafts back until they pass.

  • Don't skip the judge – It checks accuracy and prevents hallucinations.
  • Don't worry about token cost – With GLM's coding plan, it's cheap and you likely won't run out.
  • You don't need to be technical – The Kanban board is just like Trello; agents handle moving cards.
  • Old way is slower – Manually researching, writing, and editing in separate tabs leads to mistakes. The agent team does it all on one board.
  • Expect iteration – Videos and posts may not be perfect first time; agents keep iterating until ready.

07Who is this for and why it matters

This is for anyone who needs content for SEO, social media, or a new blog. Julian says now is a great time to set up a new website because SEO is working better with better agents and quality control.

The system lets you drop in a topic and get a checked post and video done automatically. You can go do something else while the agents work. You don't need to be a coder.

Julian emphasizes that if he and others can do it, you can too.

The full content machine, including the GLM 5.2 agent crew, the judge, the Kanban board, and auto-publish, is pre-built inside the agent OS in the AirPuffer boardroom. It comes with coaching calls, daily tutorials, a prompt library, and a community of 3,600 operators.