Kanade
Private AI workstation. Guided onboarding, local provider lanes, Forge handoff, and memory that survives sessions.
What it runs
- desktopprivate development build
- memorysession ledger + vault context
- bridgeForge, Discord, Obsidian
homelab / notebook / archive
I build the tools I wish existed.
Projects, services, generated media, and field notes from the systems I keep close enough to understand.
selected work
The front door is the work with the strongest evidence: private AI tooling, agent handoff, and media systems that run on my own stack.
Private AI workstation. Guided onboarding, local provider lanes, Forge handoff, and memory that survives sessions.
What it runs
Spec → branch → review loop. Codex, DeepSeek, and Claude as accountable agents inside one task board.
What it ships
BYOK anime sommelier. Plus is in testing with accounts, AniList write-back, and an Overseerr request flow.
What it proves
more work
Group trip planning with shared candidates, ranked decisions, and a public visual preview.
Next.js / Postgres / mapspreview privateKataSelf-hosted fitness coach turning Apple Health exports into daily training guidance.
FastAPI / SQLite / LM Studioin progressFlowA project-aware feed that ranks videos and posts by what I am actually building.
FastAPI / embeddings / SQLitepausedMoldA conversational CAD studio for talk -> design -> visualize -> iterate.
Next.js / FastAPI / shadcn/uipreviewRallyA friend-first planning surface for turning loose maybes into a real plan.
Astro / TypeScriptpreviewVibeA music-preview demo with real audio, playback state, and a compact listening UI.
Astro / Web Audiohow this site gets made
Every surface on this site comes out of a private local workshop — a suite of generation CLIs I built (imgen, vidgen, voicegen, soundgen, notebookgen) orchestrated by webgen, with Ditto for reference capture and Playwright for browser proof. All automated, all self-hosted.
Generated media stays on disk until it earns a spot. The browser has the final say.
service ledger
Persistent Discord relay for Forge: posts run status cards, keeps the notification path alive, and routes replies back into the local dashboard.
runningobservabilityGrafana + Prometheus stackTime-series metrics, historical graphs, threshold alerting. The 'why is the cluster slow?' answer engine.
runningautomationHome AssistantSmart-home hub running my bedroom Sleep Lab — environmental tracking, humidifier control, and nightly AI briefings.
runningproductivityHomepageThe internal dashboard I land on. One page with service status, key metrics, and click-throughs to the underlying apps.
runningaiLM StudioCanonical local LLM runtime on the Zenbook/Jarvis host, serving an OpenAI-compatible API to Home Assistant, Open WebUI, n8n, and read-only agent health checks.
runningcurrentlyShipping AI tools and homelab automation, one tool at a time.
/nowlatest field note
A real backup incident caused by a multi-disk LXC, a freezer race, and the very specific kind of humility that only backup systems can teach.