01What Chinese AI Agents Are and How They Work

Chinese AI agents are artificial intelligence models developed in China that can build and automate tasks. In this episode, Julian Goldie covers three of them: GLM 5.2, Kimi K2.7, and N2. These agents can create websites, games, and videos, and they work inside a system called an agent operating system.

These agents are not just chatbots. They can look at their own work, critique it, and improve it. GLM 5.2 and Kimi K2.7 are open source, while N2 is free through an API on OpenRouter for a limited time.

Julian shows how to use them together in a team to get better results than using a single model.

02Main Features of These Three AI Agents

Each agent has unique strengths. GLM 5.2 is the most powerful among the three. It can build full websites, 3D games, and even an operating system called GLM OS that saves your work automatically.

It also supports a judge agent that reviews and critiques output for quality.

  • Kimi K2.7 has a fast mode for quick coding and can create videos. Julian shows a fully AI-generated video that was scripted, edited, and produced by this agent.
  • N2 has a 200k context window and is free as an API for about two weeks. It works with free Claude Code and can be voice-controlled to build tools and apps.

03How to Get Started with These Chinese AI Agents

To start using these agents, you need access to them and a system to run them. Here are the steps Julian recommends:

  1. Get the API key for N2 from OpenRouter (free for a limited time). For GLM 5.2 and Kimi K2.7, download open-source versions from their official sources.
  2. Set up an agent operating system (Julian uses the one from the AI Profit Board community, which includes pre-wired profiles for these agents).
  3. Create agent profiles for each model. In the system, you can name them GLM 5.2, Kimi K2.7, and N2, and assign them to separate workspaces.
  4. Plug the agents into a Kanban board to organize tasks. You can create teams, like a writing agent, research agent, and judge agent.
  5. Use the Fusion API from OpenRouter to combine multiple agents. This method uses a judge to fuse answers from all models into one best output.

04How to Use These Agents in Your Projects

Julian shows several real-world uses. For coding, you can give an agent a prompt to build a website or a 3D game. The agent will create the code, and you can preview the result.

With N2, you can even use voice control to describe what you want, and the agent builds it.

For video creation, Kimi K2.7 can script, edit, and produce a full video. Julian shows an example where the AI made a video better than what most of his team could do. For quality control, set up a judge agent that reviews the output and tells the team to iterate until it is done properly.

You can also use the idea factory feature: agents in a group chat drop ideas into a pipeline, which generates a plan. You approve or reject it, and the system builds it.

05Results and Tests Shown in the Episode

Julian shows several test results. He built a full operating system called GLM OS, which created multiple apps and saved data automatically. He also created an open-world 3D game with GLM 5.2.

With Kimi K2.7, he made a video that was fully AI-generated and better than human-made ones.

For benchmarks, Julian notes that the Fusion API method combining all three agents beats Claude Sonnet on the Draco score. He says you should test it yourself. He also mentions that GLM 5.2 has the best performance among the three, with a larger context window than Kimi K2.7, which has a smaller one.

06Tips, Limits and Mistakes to Avoid with These Agents

Julian warns not to use these agents directly in a separate tab. Without an operating system, you lose track of what you built and cannot coordinate teams. Use a system like the agent OS to keep everything organized.

He also notes that Kimi K2.7 has a much smaller context window than GLM 5.2, so pick the right agent for your task. For the Fusion API, do not trust the benchmark alone—test it yourself. Finally, N2 is only free for a limited time, so use it before it expires.

Julian recommends starting with GLM 5.2 if you can only pick one, or N2 if you need a free option.

07Who Should Use These Chinese AI Agents and Why

These agents are for anyone who wants to build and automate tasks without deep coding skills. Julian says people in his community with no coding experience have set up these agents successfully. The agents are powerful enough for creating tools, websites, games, and videos, which makes them useful for entrepreneurs, content creators, and developers.

The key reason to use them is the combination of open-source models and a free API, which reduces cost. With a team of agents, you can get quality control and better results than using a single model alone. Julian emphasizes that having an organized workspace and a pipeline of ideas lets you reuse past work and scale up your automation quickly.