5 Overlooked Packages Running My AI Directory Stack

The best parts of a project are not always the AI model or the hosting platform.

I spent time reading the source code for five dependencies in my package files. These tools are not trending. They are load-bearing.

My stack uses Astro 5, Turso libSQL, GitHub Actions, and Claude Haiku 4.5. I run three sites: Top AI Tools, Find Games Like, and Open Alternative To. The infrastructure is solid. This lets me focus on content instead of fixing bugs.

Here are the five tools:

  • tsx I use this to run every ETL script. It works without extra configuration. It uses esbuild, so startup is fast. It does not perform type checking. I use pnpm for type checking during CI. This tradeoff makes my scripts run faster.

  • Pagefind This runs after my build step. It crawls my HTML and creates a compressed index. It works on static hosts like Vercel or Cloudflare with no extra setup. The index stays small. I use the JS API directly to control how results look in my Astro components.

  • Crawlee I have not used this yet, but I plan to. I currently use manual parsing for my data. Crawlee adds request queues and rate limiting. It is the right tool for when I start scraping sites without APIs.

  • yaml This parses the frontmatter in my articles. It is small and has zero dependencies. It handles complex objects well. It can also turn objects back into YAML while keeping comments. This helps when I want to update files automatically.

  • @libsql/client This connects my scripts to Turso. I rely on the batch API. One call can run multiple statements in one network trip. This is fast when seeding tables. It also works locally without a network. This saves my API quota during development.

These packages are the quiet infrastructure. They let the AI parts of my stack do their work.

Source: https://dev.to/morinaga/five-overlooked-packages-running-my-ai-directory-stack-1lem