A local chatbot, built in about ten minutes
A local chatbot, built in about ten minutes
A quick experiment: how far can you get with a local LLM and an agentic coding tool in one sitting?
I wanted to see how quickly I could get a chatbot running against a model on my own machine, so I set a loose timer and recorded the whole thing. I used CloudCode.ONE to do the coding and pointed it at a local model. Roughly ten minutes later I had something I could actually talk to. The recording below is the unedited run — it's silent, since my screen capture didn't grab audio.
The unedited ~10-minute run (no audio) — click to watch on YouTube.
The setup
Nothing exotic. It all runs on hardware I already had:
- GPU — NVIDIA RTX 5060
- Model — Bonsai 27B (a 1-bit LLM)
- Runtime — llama.cpp
- Coding — CloudCode.ONE
- Time — roughly ten minutes
- Cost — under 20 cents
The model runs on the machine itself, so the conversation stays local and there's no per-message API bill. A 1-bit model like Bonsai 27B is small enough to sit comfortably on a mid-range card, which is really what makes the ten-minute version possible — you're not waiting on a big download or a beefy GPU.
What I'd flag
This was a quick experiment, not a finished product. Ten minutes gets you something that works, not something polished — no real error handling, no UI to speak of, and a 1-bit model trades some quality for its small size. The point wasn't to build the best chatbot; it was to see how little friction there is now between "I want to try this" and having it running in front of you. Turns out: not much.
If you try something similar, I'd be curious how it goes.