It's currently Sunday evening, January 18, 2026.
Today, I worked on a summarization feature, handing all articles from the blog to an AI Agent for summarization. Managing the context was a bit troublesome, so I directly integrated the OpenCode Agent. This thing can use MiniMax M2.1 for free and even spin up a Web Server. It's incredibly useful. I immediately switched sides—goodbye, whatever that Claude Code thing was, tossed it out! I reworked things a bit, and this little code is just way better than the current DeepSeek. Of course, once DeepSeek v4 comes out, that might change.
Completed some features for CZON.
- Dark Mode (0.4.3): Automatic/manual dark mode switching, supports switching based on system theme, and includes basic dark mode adaptation for typography, images, and Mermaid diagrams.
- Sitemap (sitemap.xml) generation, requires the additional CLI parameter
--baseUrl(0.4.2) - Crawler rules (robots.txt) generation (0.4.2)
- Removed the unused
-t, --templateparameter (0.4.2)
Apart from some special logic for dark mode (Mermaid, Tailwind dark mode adaptation) that required manual adjustment, MiniMax M2.1 basically nailed all other requirements on the first try. A pleasant surprise.
Planning to integrate OpenCode into CZON for some in-depth analysis features. This will make the tech stack heavier, but the benefit is being able to directly use OpenCode's Agent functionality, saving me a lot of hassle. Developing an Agent myself is simply impossible and unnecessary.
Wanting to further develop CZON into an online product, allowing users to use it without installation, while also needing to reduce our operational costs.
- Host the product's static website for free via GitHub Pages to save on server costs. Use Cloudflare to proxy traffic, ensuring good access speeds for users globally.
- Use GitHub to host user data, saving on database costs. (This introduces some complexity, but I think it's acceptable. Otherwise, we'd need to build an authorization system and a full database, and the initial user base can likely accept this method.)
- Use GitHub OAuth to obtain user authorization for read/write operations on document repositories, bypassing PRs and pushing directly to the user's repository's main branch. (This essentially means users' data repositories are fully managed by us, but users can take them back at any time.)
- The product's static website will deploy an online Markdown editor that loads the user's repository. Users can edit documents online and save them directly to their GitHub repository via GitHub OAuth.
- When users need to call CZON's features, they will use OAuth to trigger GitHub Actions to run CZON's CLI functionality, saving the results back to the user's repository. (This process will have some latency, but I think it's acceptable since it's free hosting.)
- Users access their documentation websites via GitHub Pages, with the website content pulled directly from their GitHub repository. (This means users' website content is also fully managed by us, but they can take it back at any time.)
- Integrate OpenCode and other AI services within GitHub Actions. Users will need to manually configure Secrets so they can use their own OpenAI Key or other AI service keys. (This avoids us bearing high API call costs, and users get better privacy protection, though OpenCode currently offers many free models.)
It sounds like we don't need to spend a single cent! Fantastic, thanks to GitHub, Cloudflare, and OpenCode! I'm freeloading off everyone, amazing! Let's call this product CZONE (czone.zccz14.com)! It will likely use Vite + React + TailwindCSS for the frontend, be serverless, and directly use OAuth to call the GitHub API.
Ahem. Today, C1 seems to have used LegionMind to fix some lagging bugs in Yuan. I haven't been involved with the Yuan project at all lately; I have my promotional work to do. Gotta trust the power of the team.
