I have a Mac Mini M4 Pro with 24GB. I wanted to run the best local coding model I could, with enough context to be useful. TurboQuant — Google's KV cache compression — is supposed to make impossible m
LLMs are great at generating code. That's the problem.
Start your script with:
LLMs default to Svelte 4 patterns. If you don't give them context, you'll get export let, $: reactive statements, and on:click instead of Svelte 5 runes and modern syntax.
Six patterns you'll actually use, with minimal examples.
Model Context Protocol (MCP) lets Claude access local tools, files, and APIs. Instead of pasting context into chat, you give Claude direct access to what it needs.
Add Tailscale to any Docker Compose project as a sidecar container. Your app gets a stable hostname on your tailnet with automatic HTTPS -- no port forwarding, no reverse proxy config.
Scripts go in ~/.scripts, which I add to my path.
Tailscale makes it easy to connect all my devices. (I have too many(/gear)).
UV replaced pip, virtualenv, pipenv, poetry, and pyenv. One tool. Installs in seconds. Runs everything faster.
After years of development, I've launched my dev blog, powered by Svelte.
Run your SvelteKit app in Docker with secure Tailscale access. Template: github.com/mrgnw/svelte-docker-tailscale
⭐️ GL-iNet travel router <a href="https://link.gl-inet.com/menubarTRMT3000" class="no-underline" target="blank"><span class="text-current/20">↗</span></a>