This is a repository for my vibecoding experiments. This project explores the actual capabilities of vibecoding, pushes the boundaries of this approach, and investigates how AI technology can be integrated into my development workflow.
- Backend: llama.cpp (Vulkan backend) on Sapphire Pulse RX 7900 XTX (24G VRAM), managed via llama-swap, on Debian Trixie
- GPU Tuning: LACT is used to limit GPU clocks to 1929 MHz and fan speed to 40% for reduced noise
- AI Agent: OpenCode
- GitHub Pages Landing Page — Root
index.htmlserving as a subprojects menu, hosted via GitHub Pages.
- AGENTS.md - Agent guide for the repository structure and subprojects navigation.
| Subproject | Model | Description |
|---|---|---|
bonds-profitability/ |
qwen3.6:27b | Bond Profitability Calculator single-page web application |
cue-builder/ |
qwen3.6:27b | CUE File Builder web application |
snake-game/ |
qwen3.6:27b | Classic Snake Game single-page web application |
tldr/ |
qwen3.6:27b | Linux TLDR Commands single-page web application |
games-ontology/ |
qwen3.6:27b | Games Ontology interactive knowledge graph visualizing game relationships |
flash-cards/ |
qwen3.6:27b | Chinese language learning flashcards for HSK vocabulary |
arkanoid/ |
qwen3.6:27b | Classic Arkanoid (brick breaker) single-page web application |
text-adventure/ |
qwen3.6:27b | Click-choice text adventure set on a damaged space station |
This repository follows a hands-off file editing approach:
- Ideas & Prompts: All project ideas and prompts are my own, originating from old unfinished or dropped projects, or even projects I deemed not viable to research into at all.
- File Editing: All code files are edited exclusively by the AI model via OpenCode. This includes this README.md.
- Filesystem Management: I manually manage the filesystem from time to time, including adding example files to subproject directories (e.g.,
cue-builder/examples/). - AI-Assisted Analysis: AI is used in my workflow to extract summaries and metadata from example artifacts (e.g., analyzing CUE output, generating example inputs, or documenting patterns).
- Review: I perform a thorough review of everything that is generated before it is committed.
- Git Management: I manually manage the Git repository (pushes, branches, signing, etc.), but commit messages are generated by the AI model.
- Deployment: Automated via GitHub Actions on every push to
master(deployment configuration also created by AI). - Local Environment: Everything runs locally and self-hosted. No cloud services or external APIs are used.
- Open-Source Focus: This project exclusively uses open-source solutions for both models and tooling.
This project is released under the Unlicense.