Bolt.diy

🛠️ The Setup

The coolest part? It can load local models. 🏠 I decided to run Qwen2.5-coder-14b (about 8GB) using LM Studio as the server.

• Context Length: 8192 🧠
• Performance: It completely maxed out my 8G GPU memory! 📈

One reason I prefer LM Studio over Ollama for this setup is the visibility—you can monitor every single token and request on the console in real-time. 🖥️🔍

📝 The Test Drive

I asked bolt.diy to build the simplest To-Do app using only pure JS, HTML, and CSS. 🎨 The generation was definitely slower compared to the online version, but honestly? It was still acceptable for a local run. 🐢➡️🐇

🏁 Final Thoughts

And there it is—a fully functional To-Do app working right in the web container! 🌐 I even tried using it as my “copilot” in OpenCode. 👨‍💻

While the speed isn’t quite at a “daily driver” level yet, it’s getting there. Who knows? Maybe one day this will be my go-to workflow when I’m stuck offline. ✈️🔌