Original video in Chinese.
Key Takeaway
- Real-time information fully managed: Kimi Claw automatically collects AI news every 8 hours (core summary + background + what’s worth paying attention to), scans designated YouTube channels every 24 hours (core summary + Chinese and English subtitles), structurally generates Markdown, and pushes it to GitHub, automatically syncing to local Obsidian to achieve zero-intervention external information handling.
- Perfect local + cloud loop: locally, OpenCode + the newtype Profile plugin are responsible for deep creation and knowledge management; in the cloud, Kimi Claw handles information gathering and automated tasks; through GitHub relay synchronization, it is both securely isolated and seamlessly connected.
- Three-layer personal AI production system: knowledge base (local accumulation of historical content), skills (structured personal experience), real-time information (cloud Agent automatic crawling). I’ve fully built mine, and I strongly recommend ordinary people plan their systems around these three layers.
I’ve started handing over all external information gathering to Kimi Claw. I’ll just wait to read the “AI briefing.”
Every 8 hours, Kimi Claw collects the newly released AI news once. It organizes the content into three sections: “core summary,” “background information,” and “what’s worth paying attention to.” Among them, “background information” means it expands the search based on the news content and fills in the full context, so that when we read the news, it’s more complete. And “what’s worth paying attention to” is its own judgment of the news, identifying what in it deserves notice. This can sometimes be a pleasant surprise.
Every 24 hours, Kimi Claw scans the YouTube channels I specify. If it finds a new video, it extracts the subtitles and organizes them into three sections: “core summary,” “Chinese translation,” and “English original.”
Whether it’s news or video content, once Kimi Claw has processed it, it generates a Markdown document and uploads it to the corresponding folder in the GitHub repository. And on my local machine, whenever I open Obsidian, it automatically downloads the documents from the GitHub repository. In this way, I can read and further process them locally.
This is just the beginning. Next I’ll also arrange for Kimi Claw to periodically scan posts on X, filter out the high-traffic AI-related posts, turn them into a daily report, and push it to me.
Why do this?
A couple of days ago, I said in Knowledge Planet that if you want to integrate Agents into your personal production system, there are three levels we need to plan and build ourselves:
First, the knowledge base level. These are the things we’ve accumulated over time, such as personal notes, historical data, and so on. These are all things AI can’t obtain externally, or even if it can, they require complex processing.
Second, the skills level. We’ve accumulated a lot of experience in our minds, but that experience is unstructured and can’t be taught to AI. So we must proactively structure and output that experience, and package it into Skills.
Third, the real-time information level. The first two levels are all things from the past. But the world keeps moving forward. So we need to capture and process real-time information. For example, the news and new videos I mentioned at the beginning.
I started building the knowledge base level two years ago, and it’s basically done now.
I started making the skills level last year, and this year I even built the newtype Profile multi-Agent orchestration system, all of which are about structuring and outputting my experience.
So now only real-time information remains. Before, what we used were chatbots, whose capabilities were very limited. For example, you could only set up one scheduled task to collect news. Now with OpenClaw, like the Kimi Claw I’m using, all of that can be solved.
I’ve currently configured three tools for Kimi Claw.
For information retrieval, there are two tools, both called through APIs. One is Tavily. Kimi Claw uses it to search for information, such as AI news. The other is Supadata, which is responsible for extracting subtitles from YouTube videos.
For document synchronization, I use GitHub CLI. Obsidian also has a GitHub plugin. After configuration, Kimi Claw pushes the documents to the GitHub repository, and then they sync to the Obsidian note repository.
In the previous video, I introduced the method of synchronizing via email. Now I’ve switched to this approach, with the goal of keeping an archive on GitHub. The reason I chose Obsidian is that its notes are actually Markdown documents stored locally. These documents can later all serve as material for other AIs to reference.
Synchronizing the cloud and local machine like this might seem redundant, but in fact it’s entirely for security reasons.
A lot of creators hype OpenClaw endlessly for traffic, but completely fail to mention the word “security.” Honestly, only individual developers would dare to launch this product; large companies absolutely would not dare to release it this way. There are too many pitfalls inside.
For example, recently the alignment lead at Meta Labs got hit—her emails were deleted by OpenClaw.
Even professionals like these can’t avoid it, so ordinary users definitely shouldn’t be stubborn. Before OpenClaw matures, the safest way to use it is certainly in an isolated environment, such as in the cloud.
At this point, my AI production loop has been formed.
Locally, I use the newtype CLI I developed myself to open the content repository, and all idea discussions and finished outputs are completed here.
In the cloud, Kimi Claw automatically helps me gather all the information and syncs it back locally. If there’s any temporary research need, I’ll issue instructions directly.
Today’s AI tools are already powerful enough and abundant enough. But to use them well and get results, you still need to think it through yourself. Of the three levels I mentioned earlier, how many have you set up now? Feel free to share in the comments.
OK, that’s all for this episode. If you want to learn about AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next episode!