A technical documentation website for setting up and optimizing CachyOS with Ollama and Nous Research Hermes Agent. Built as a lightweight, zero-dependency static site.
Select your GPU vendor (NVIDIA/AMD/Intel/CPU-only), processor instruction set (x86-64-v3 or v4), and primary workload — the configurator generates a tailored shell script with the correct Ollama package, systemd overrides, and driver configurations. Performance metrics (latency reduction, throughput boost) update in real time. Features a Dynamic Script Downloader to download generated setup scripts directly.
Select model parameters (8B, 14B, 32B, 70B, Custom), quantization bit levels, and target context length to calculate VRAM overhead for weights and KV cache. Displays color-coded safety indicators and a Segmented VRAM Allocation Diagram outlining system headroom relative to a 24GB RTX 4090.
Step-by-step optimization guide and automated script builder for GPU-accelerated Docker containers (CUDA / ROCm) on Arch Linux. Contains pre-configured tabs to copy docker-compose.yml boilerplates for Open WebUI, ComfyUI, and Ollama.
Explains optimizations such as x86-64-v3/v4 target compilation, BORE CPU scheduler, and ZRAM. Offers advanced setup guides for:
- Multi-GPU Isolation (
CUDA_VISIBLE_DEVICES) - Custom Context Configurations (
num_ctxin Modelfiles) - Kernel Dirty Page Allocation (
vm.dirty_ratio, etc.)
Comprehensive setup details for running virtualized Windows environments inside Virt-Manager with VirtIO SCSI storage, UEFI Secure Boot, vTPM 2.0 emulator devices, and driver mounts.
- PipeWire quantum overrides (
default.clock.quantum = 128) and virtual loopback audio sinks for voice agents. - Custom CachyOS ISO compilation using
cachyos-archisoprofiles preloaded with CUDA, pip packages, and startup scripts.
Detailed specification and structure sheets for registering host automation skills (e.g. system info inspections, resource monitoring) inside the Hermes Agent workspace. Includes a Live Telemetry Dashboard Canvas drawing animated wave tracks of CPU, VRAM, and token inference rates.
Includes a Live Search & Tag Filter to instantly filter troubleshooting accordion cards (e.g. GPU, Virt-Manager, PipeWire, Kernel) by category and text query.
| Layer | Technology |
|---|---|
| Structure | Semantic HTML5 |
| Styling | Vanilla CSS (custom variables, grids, flexbox layouts, keyframe animations) |
| Logic | Vanilla JavaScript (no frameworks, no build step, HTML5 Canvas, SVG animation) |
| Fonts | Inter + JetBrains Mono |
| Icons | Inline SVG (no external library dependencies) |
Any static file server works. The simplest option:
# Python (built into most Linux distros)
python3 -m http.server 8080
# Then open http://localhost:8080.
├── index.html # Main page — layout containers, SVGs, FAQ, and inputs
├── styles.css # Design system — variables, dark/light themes, animations, media queries
├── app.js # State manager — configurators, VRAM calculator, FAQ search, canvas loops
└── assets/ # SVG logo and dashboard assets
- Dark/Light theme toggle with
localStoragepersistence - Global Toast Notification Manager sliding copy confirmations into view
- Dynamic shell syntax highlighting in configurator screens
- Scroll spy navigation with a Sliding Indicator vertical line tracking sidebar anchors
- Floating "Back to Top" navigation shortcut triggering smooth page returns
- Responsive design fully optimized for mobile, tablet, and ultra-widescreen monitors