Uses
What I build with.
The kit I reach for every day. The lab hardware that runs the agents has its own datasheet on /infrastructure — this is the layer above it: the machine I type on, the tools I build with, and the services that keep the sites and the home lab wired together. Everything here is already documented somewhere on this site; consider it a living list.
Hardware the lab runs it — full specs on /infrastructure
Daily driver
- Anvil — M4 Mac Mini, 16GB — the machine I actually type on; runs the Claude Code workstation and the ARIA Relay daemon
The lab
- Furnace, Crucible, Bellows — two AMD Strix Halo boxes on a Thunderbolt-5 link plus an out-of-band KVM — the compute behind the agents
Build & edit
Agent
- Claude Code (Opus) — the pair-programming agent this whole site — and most of the lab — is built with
Version control
- Git + GitHub — every project lives in its own private repo
Languages
- TypeScript / Node — the sites, Nexus, and the agent workers
- Swift — the ARIA iOS companion and the macOS Relay daemon
- Python — scripts, image/video generation, data plumbing
This site how niclydon.io is made — see /colophon
Framework
- Astro 5 — static build, zero JS shipped beyond a 40-line keyboard-shortcut handler
- MDX content collections — projects, infrastructure, and the build log
- Plain CSS — design tokens in :root, no framework runtime
Type
- Instrument Serif — display
- JetBrains Mono — everything monospace — served from Google Fonts
Delivery
- Vercel — static hosting at the edge
- Cloudflare — public DNS (DNS-only, grey cloud)
- Let's Encrypt — TLS via Vercel
Data & infra
Data
- PostgreSQL 16 + pgvector — the ARIA / Nexus store and every embedding search
- SQLite — request logging and edge caches
- AWS S3 — cold media archives
Inference
- llama.cpp — the primary local runtime
- vLLM — batch and bulk work
- Forge — an OpenAI-compatible gateway over every local model
Network & ops
- Tailscale — zero-trust tailnet across every machine
- Caddy + CoreDNS — wildcard TLS and *.niclydon.io resolution
- systemd — how every service actually runs
- Prometheus + Grafana — metrics and dashboards
- AWS Secrets Manager — one place for credentials, pulled at deploy
Models & services
Local models
- qwen3 family via Forge — reasoning, vision, embeddings, code — running on the lab GPUs
Cloud models
- Claude, GPT-4o, Gemini, DeepSeek, Grok — reached per-request when a specific voice or a fallback is needed (see Broadside)
Publishing
- Blotato + Resend — multi-platform distribution and newsletter sends for Broadside
Sourced from what's already published here — the colophon, the infrastructure datasheet, and the project pages. If something's missing, it just hasn't been written down yet.